Google Scale Economy Outpaces Early Rivals in Search Monetization

Acquired

Google's rapid ascent to global search dominance resulted from a self-optimizing ad auction framework engineered over highly efficient, custom distributed commodity hardware.

AI-assisted summary of Google Part I: Origins of Search. How the Best Business in Human History Happened (Audio) — original source. Check the original for full context.

Source published: 2025-06-30

Detailed notes

## Background This episode was published on YouTube by the channel Acquired on June 30, 2025. According to the page description, it initiates a multi-part series exploring the history of Google search from its roots as a Stanford research project to becoming the world's most profitable business, while also framing its modern position against emerging AI competition.

## TL;DR The episode explores the origin story and meteoric rise of Google, tracing how co-founders Larry Page and Sergey Brin combined intense ambition with structural innovations to revolutionize the search industry. It details Google's breakthroughs in distributed computing on cheap commodity hardware, the strategic hiring of engineering elites like Jeff Dean and Urs Hölzle, and the introduction of its highly lucrative AdRank and second-price auction mechanics. The walkthrough tracks Google's transformation into a financial powerhouse through high-stakes distribution partnerships with Netscape, Yahoo, and AOL, alongside the eventual launch of AdSense, laying a foundational empire before its 2004 IPO.

## Atmosphere & style The production carries the detailed, analytical, and conversational energy characteristic of hosts Ben Gilbert and David Rosenthal, blending intense business admiration with historical narrative depth. The dialogue balances high-IQ technical breakdowns of data infrastructure with a sense of high-stakes corporate drama, particularly when reflecting on 'bet the company' moments. Listeners experience a deeply engaging, retrospective journey that makes a massive tech corporate giant feel like an intimate, rapidly evolving university campus experimenting on the edge of the internet's future.

## Key takeaways - [01:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=78s) Google operates as a government-declared monopoly with a 90% market share, generating more profit than any other U.S. company, including Apple, Microsoft, and Berkshire Hathaway. - [05:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=303s) Larry Page and Sergey Brin maintained an equal-equity partnership driven by intense, calculated ambition, contrary to the narrative that they were accidental entrepreneurs. - [34:11](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2051s) The name 'Google' originated from 'Googol' ($10^{100}$), which was suggested to reflect the scale of the internet index, but Larry Page misspelled it during registration. - [01:02:06](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3726s) Google pioneered distributed computing on consumer-grade commodity hardware, using custom verification software systems to tolerate constant hardware failures and drastically cut costs. - [01:07:47](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4067s) Google's hyper-economical scaling mechanism and custom software architecture eventually allowed Google search to achieve an 87% gross margin once their business model matured. - [01:19:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4767s) Sequoia Capital and Kleiner Perkins split Google's $25 million Series A round in 1999 because the founders forced the competing venture capital firms to co-invest as a public relations strategy. - [01:28:20](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5300s) A personal affiliate test with Amazon designed by Jeff Dean and Marissa Mayer mathematically proved that intent-based search ads yielded incredibly high conversion rates. - [02:03:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7430s) Google overhauled search advertising by creating 'AdRank,' multiplying an advertiser's maximum bid by their ad's click-through rate (CTR) to use engagement as a proxy for consumer relevance. - [02:09:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7784s) Google adopted a cost-per-click model with a second-price auction system, allowing the winner to pay only one penny above the second-place bid to build long-term advertiser trust. - [02:16:10](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8170s) International search traffic grew organically to 50% of Google's total volume by 2001, prompting the company to seed global sales teams that scaled international revenue rapidly. - [02:21:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8490s) Google risked bankruptcy by signing a landmark deal with AOL, offering a $100 million revenue guarantee that ultimately unlocked massive marketplace liquidity and a virtuous cycle of increasing returns. - [02:40:33](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9633s) The Google Toolbar was deployed hyper-aggressively via browser bundles and desktop partnerships to prevent Microsoft from cutting off Google's traffic at the browser level. - [03:16:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11763s) Paul Buchheit's internal Gmail prototype, which matched folder keywords with advertisements, directly catalyzed Jeff Dean and Susan Wojcicki's development of AdSense for third-party websites. - [03:16:53](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11813s) The core organic search playbook prioritized lightning speed, an uncompromised talent bar, a clean user experience, and a strict aversion to paid inclusion. - [02:56:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10560s) Google was highly profitable by 2004 but actively sought to avoid its IPO to hide its lucrative margins from Microsoft, which controlled the Windows and Internet Explorer ecosystem. - [03:24:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12267s) An evaluation through Hamilton Helmer's Seven Powers reveals Google achieved immense Scale Economies and Network Economies, alongside severe Counterpositioning against traditional portals.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=0s) The Final Studio Episode and the Road to Radio City

Ben Gilbert and David Rosenthal kick off the show by noting a major milestone: this is the final episode they are recording in their current studio before heading to their massive live event at Radio City Music Hall. David reflects on the dramatic shift, joking that they are moving from a "very, very small audience of one" to a massive stage where, if they make a mistake, "no one will notice" because they can just re-record it like it never happened. Ben laughs and imagines yelling "strike that, let's take that again" right in front of the live crowd, joking that it would give the audience an authentic, "inside look" at how Acquired is made.

After their iconic intro music theme, Ben officially welcomes listeners to the summer 2025 season of Acquired, the podcast that explores great companies and the playbooks behind them. He introduces the main thesis of the season: artificial intelligence is the defining story of our time, marking the next trillion-dollar technology wave after personal computers, the internet, and mobile. To truly understand AI, Ben argues, one must understand Google—the company most responsible for its technical foundation and the tech wave that preceded it. This episode marks the beginning of their highly anticipated, multi-part "Google saga."

[01:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=78s) Google: The Single Greatest Business of All Time

Ben points out that Google has served as the "front door to the entire internet" for a quarter of a century. However, when it was founded in 1998, it was far from alone; a dozen other search engines already existed, operating on various business models that David describes as "not very interesting." The central question the hosts aim to answer is why Google worked, and how it evolved from a clever tech product into what Ben calls "the single greatest business of all time."

Laying out the staggering financial reality, Ben stresses that he is not being facetious: Alphabet today generates more net income and profit than any other U.S. company, surpassing titans like Apple, Microsoft, ExxonMobil, JP Morgan Chase, and Berkshire Hathaway. He attributes this "cash gusher" to three massive factors multiplied together: a super high gross margin, operating in a giant market, and possessing a government-declared monopoly with a 90% market share. David notes that while they will address the antitrust and government aspects much later in the series, the story begins with the creation of the most beautiful business ever made.

* **Market Dominance:** Google maintained a seemingly unassailable market position until the AI wars recently intensified. * **Viral Growth:** While competitors existed, Google stood out due to its clean user experience—featuring just a single search box on the homepage—and high-quality, fast search results that spread virally via word of mouth. * **The Last Search Engine:** Ben summarizes the company’s ultimate achievement by stating that while Google was nowhere near the first search engine, "it was the last" (though David jokingly notes Microsoft's Bing might take a "few percentage points of issue" with that statement).

[03:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=210s) Community Announcements and the Sell-Out Live Show

Before diving into the history, Ben directs listeners to the Acquired email list at `acquired.fm/email`, noting it is the only place to get hints about upcoming episodes, corrections, and behind-the-scenes tidbits. He also invites the community to join the active discussion in their Slack channel at `acquired.fm/slack` and plugs their interview show, ACQ2. Ben recalls their recent ACQ2 episode with Jesse Cole of the Savannah Bananas, calling it "the most fun I've ever had recording basically any podcast episode."

David takes a moment to thank their presenting partner, JP Morgan Payments, who is also sponsoring the upcoming Radio City Music Hall show on July 15th. The event is essentially sold out, with fewer than 100 tickets remaining out of 6,000 total. David urges remaining listeners to grab tickets immediately, promising an incredible lineup that includes exclusive merchandise, a pre-show fan meetup, an afterparty, and a next-day encore event at the New York Stock Exchange. Ben concludes the administrative notes with their standard disclaimer: the show is for informational and entertainment purposes only, and they may hold investments in the companies discussed.

[05:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=303s) The Hyper-Ambitious Midwestern Roots of Larry Page

Ben poses a philosophical question to start the historical narrative: "When is the first time any human being searched for anything ever?" While they joke they should ask Google, David points out that the true story of Google begins in March 1973 in Lansing, Michigan, with the birth of Lawrence Edward Page. Larry was born to Carl and Gloria Page, a household uniquely steeped in academia and computing. Carl Page Sr. was a pioneer and professor of computer science at Michigan State University (MSU) in East Lansing, having earned his PhD from the University of Michigan. Gloria Page similarly held a computer science degree from Michigan and taught programming at MSU.

The hosts emphasize how extraordinarily rare it was in the 1970s to grow up with two computer science academics as parents. Because his parents started in the field during the 1950s, Larry literally breathed the air of the upcoming PC revolution before Microsoft even came online. This environment deeply shaped Larry and his older brother, Carl Jr. (who was nine years older, studied CS at Michigan, and went on to work early on at Microsoft and Mentor Graphics).

David and Ben strongly push back against the modern perception that Larry and Sergey Brin were merely bumbling, accidental academics who stumbled into a great business. "Absolutely freaking not," David declares, dispelling the notion entirely. Their research reveals that both founders possessed a fiery, calculated ambition on par with tech icons like Bill Gates or Mark Zuckerberg.

* **The Visionary Quote:** Larry knew his path early, stating, "Probably when I was 12, I knew I was going to start a company eventually. I wanted to make the world better, and in order to do that, you need to do more than just invent things... you need to use business and entrepreneurship to make these things real." * **Early Silicon Valley Exposure:** During the 1979–1980 academic year, six-year-old Larry lived in Palo Alto while his father took a sabbatical at Stanford, leaving a lasting impression of Silicon Valley on the young boy.

[09:41](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=581s) A Fateful Meeting Funded by Charles Schwab

Larry graduated from the University of Michigan in 1995 and moved to Stanford University to pursue his PhD in computer science. It was there that he had a fateful meeting with his eventual business partner and intellectual soulmate, Sergey Brin. Folklore states that Sergey, who was already in the program, was assigned to show Larry around campus. The two immediately engaged in relentless verbal sparring, constantly bickering and challenging each other on topics ranging from civic design to transportation systems.

David shares an exclusive, previously untold story regarding this meeting, sourced from their close friend Anna Patterson (an early Google employee and former VP of Engineering). Anna was a postdoc at Stanford at the time and actually organized the new student weekend. She revealed that Larry and Sergey's very first interaction took place over drinks at the British Bankers Club in Menlo Park. The two ended up shutting down the bar that night. In a bizarre twist of fate, legendary local Stanford alumnus Charles Schwab happened to be at the bar and picked up the tab for the group of students. Ben laughs at the symmetry, noting that Charles Schwab accounts would later go on to hold billions of dollars worth of Google stock.

The hosts marvel at the rare nature of their partnership, noting that Larry and Sergey operated as completely equal co-founders with equal equity—a dynamic rarely seen sustaining over decades in tech history. While companies like Microsoft or Nvidia had multiple founders, leadership eventually consolidated around a single figurehead like Bill Gates or Jensen Huang. Larry and Sergey, however, remained true "one plus one equaled a hundred" partners.

[13:19](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=799s) The Incredible Trajectory of Sergey Brin and the Explosive Web

Sergey Brin was also born in 1973, arriving in August in Moscow, Soviet Union. Growing up Jewish in the Soviet Union was immensely difficult; Sergey's family lived in a cramped, state-allocated three-room apartment shared with his paternal grandmother. His father, an incredibly talented mathematician, attended an international conference in the late 1970s and realized he had to get his family to the West. After a grueling two-year immigration process, they moved to the United States. Sergey's father became a mathematics professor at the University of Maryland, and his mother became a researcher at NASA’s Goddard Space Flight Center.

Sergey proved to be an absolute prodigy. He graduated high school at 16, finished his undergraduate degrees in math and computer science at the University of Maryland in just three years at age 19, and immediately entered Stanford's PhD program. In the summer before arriving at Stanford, he even interned at Wolfram Research under Stephen Wolfram, working on Mathematica. David describes Sergey as Larry's absolute intellectual equal, though perhaps a bit more "zany," famously sporting a love for rollerblading.

By the fall of 1995, Larry and Sergey were working closely under PhD adviser Terry Winograd. Larry and Sergey became fascinated by the newly emerging World Wide Web. Ben shares a striking statistic from John Battelle’s book *The Search* to illustrate the sheer velocity of the era: between 1993 and 1996, the web skyrocketed from a mere 130 sites to more than 600,000. This staggering explosion represented an astronomical 723% year-over-year compound growth rate for four consecutive years, setting the stage for the computing challenge the two young men were about to tackle.

The Phenomenon of the Internet and the Stanford Inspiration

The speakers compare the early growth of the internet to the famous turning point for Jeff Bezos, who realized he had to leave D.E. Shaw to build Amazon because he saw something unprecedented happening. The internet was growing at an astonishing 700% year-over-year. While such growth can happen for a single year, seeing it persist "year over year over year for half a decade" is the kind of exponential phenomenon that justifies dropping everything and changing your entire life. Around this same time, a couple of years prior, two other PhD students at Stanford had started a directory called Yahoo. Inspired by Yahoo’s traction but noticing its limitations, Larry Page proposed working on a decentralized annotation system.

Unlike Yahoo, which relied on a human, hand-curated directory of websites, Larry wanted to create a system allowing anyone to make notes and annotations directly on websites. One of the hosts admits they missed this piece of history, despite thinking they knew Google's full origin story. When Larry brought the idea to his Stanford advisor, Terry Winograd, Terry liked the problem space of the web for a dissertation but urged Larry to refine the idea further.

From Annotations to Page Rank

As Larry collaborated with Sergey Brin to refine the concept, they quickly ran into a fundamental flaw. If a massive website like the New York Times allowed open annotations, it would be completely overrun with hundreds of thousands or even millions of user comments. They realized they needed a mechanism to "separate the wheat from the chaff" so the high-quality comments could rise to

[34:11](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2051s) Rejection from the Portals and the Search for a Name

At the time, the major internet players were completely missing the boat on search. Companies like Yahoo, Infoseek, and Lycos were obsessed with becoming "portals"—chasing eyeballs and page views rather than focusing on technology. Because they weren't technology-driven, they failed to see the potential of what Larry Page and Sergey Brin had built. In fact, Larry and Sergey tried to sell PageRank to Yahoo for a mere $1 million and were flatly rejected, marking the first of several historical chapters where Yahoo passed on buying Google. They shopped the technology all around town, even showing it to Infoseek (which was later bought by Disney and shut down), but found no takers. Frustrated but determined, they returned to the Stanford campus and decided to build a real search engine themselves.

The first order of business was changing the name. While "BackRub" accurately described the technical underpinnings of analyzing backlinks, it didn't really capture the essence of searching and was unlikely to fly with the public. They began casting about for something that encapsulated this new, superior form of search. Larry’s dormmate initially suggested the name "Whatbox"—a name that rolled right off the tongue as a place where you type questions into a box. However, they ultimately discarded it because it sounded a bit too close to a porn site.

The dormmate then suggested "Googol," the mathematical term for a 1 followed by 100 zeros ($10^{100}$), which perfectly symbolized the massive scale of the internet they wanted to index. Larry loved it, Sergey liked it, and they went to register the domain name. In a brilliant twist of historical accident, Larry misspelled it during registration as "Google" (G-O-O-G-L-E). While some legends suggest "googol.com" was simply already taken, the misspelling stuck. It ended up being a blessing in disguise, as it was far more phonetic and aligned with how everyday people would naturally spell it—long before Google could bail users out with a "Did you mean?" feature.

* **Rejected Acquisition Price:** $1 million (offered to Yahoo). * **Alternative Names Considered:** BackRub, Whatbox. * **Origin of "Google":** A misspelling of "Googol" ($10^{100}$).

---

[36:52](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2212s) Bringing the Stanford Network to Its Knees

With the name secured, Sergey designed the original homepage and crafted the first logo using GIMP, a popular open-source drawing program. The early iteration was famously bright and rainbow-colored, predating the iconic drop-shadow version. The fundamental design philosophy established during the 1997–1998 academic year—a colorful logo and a single search box on a clean white page—remains virtually identical today. By the spring quarter of 1998, the project was spreading virally. It moved rapidly from the Stanford campus to other academic communities, and eventually into Silicon Valley. Google.com was suddenly fielding 10,000 queries a day.

The traffic explosion was so intense that it brought the Stanford university network to its knees, at one point consuming roughly half of the entire university's bandwidth. What made this bandwidth consumption so stunning was how asset-light the website actually was; it consisted of a simple white page with a single image and text-only search results. People were simply using the absolute crap out of it because it was lightyears ahead of the competition.

Google worked beautifully because of two core innovations: weighting results based on backlinks (PageRank) and leveraging the descriptive anchor text of those links to index keywords. This was a massive paradigm shift from existing search engines, which blindly ranked pages based on keyword density. On other platforms, a query for "dog" would just return whatever page repeated the word "dog" the most times, regardless of authority. To keep up with the technical demands of their exploding user base, Larry and Sergey resorted to desperate measures, including "stealing" computers from the university loading docks. They would intercept hardware ordered by other researchers who weren't scheduled to start their projects for a few months, spin them up for Google, and return them just in time.

* **Daily Queries (Spring 1998):** 10,000. * **Bandwidth Usage:** Consumed roughly 50% of Stanford University's total network bandwidth. * **Key Innovations:** Relevancy ranking via backlinks and keyword indexing via anchor text.

---

[40:14](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2414s) The Legendary Seed Round and a Check to an Unregistered Company

By the end of the 1998 academic year, it was blindingly obvious that Google had to spin out into an independent company, if only to save Stanford’s crashing network. The university was highly supportive of the tech transfer, and for good reason—Stanford received roughly a 1% equity stake in the spinoff, a piece of largesse that eventually paid off handsomely. Looking for guidance on how to structure the company, Larry and Sergey approached David Cheriton, a professor in the Stanford Computer Science department. Cheriton was uniquely qualified; he had co-founded an Ethernet company called Granite Systems alongside Sun Microsystems' Andy Bechtolsheim while maintaining his professorship, and Cisco had recently acquired it for $220 million.

Cheriton emailed Bechtolsheim that evening, who immediately offered to meet at Cheriton's house at 8:00 AM the next morning on his way to the office. When the bleary-eyed grad students arrived to demo Google, Bechtolsheim was instantly hooked. "Great, I'm in. $100,000," he told them. Larry and Sergey were stunned, protesting that they hadn't even come to raise money, just to ask for advice. Bechtolsheim insisted, grabbed a checkbook from his car, wrote a check out to "Google Inc." for $100,000, and drove off. The only problem? Google Inc. did not exist yet. There were no incorporation documents, no legal entity, and no valuation—just a six-figure check thrown at two students who now faced a ticking clock to incorporate and open a bank account before the check expired.

This bizarre check served as the ultimate forcing function. Over the next couple of months, as they scrambled to set up the corporate entity and lock down Stanford's intellectual property, the seed round organically grew. Professor Cheriton decided to match Andy’s investment with $100,000 of his own. They were also introduced to Ram Shriram, a former Netscape executive who had joined the startup Junglee (which was acquired by Amazon). Shriram began advising them and injected $250,000 into the round. Shriram then asked the young founders if they wanted to meet Jeff Bezos. In 1998, Bezos was considered the "elder statesman" of the internet—even though Amazon had only been founded in 1994 and had gone public just a year prior. It was a classic "six-month-old baby hanging out with a 14-month-old" dynamic; they looked entirely different at the time, but they were destined to be classmates. Shriram hosted a meeting at his house when Bezos was visiting Silicon Valley, and Bezos casually matched Shriram's stake, throwing in another $250,000.

* **Total Seed Round Raised:** $1 million at a $10 million post-money valuation. * **The Four Seed Investors:** Andy Bechtolsheim ($100k), David Cheriton ($100k), Ram Shriram ($250k), and Jeff Bezos ($250k). * **Bezos's Return on Investment:** If Bezos never sold his seed shares, they would be worth roughly $20 billion today; even if he liquidated entirely at the IPO, he turned $250,000 into roughly $200 million.

---

[48:11](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2891s) Setting Up Shop and the State of Search in 1998

Armed with $1 million in cash, Google was officially a corporation. Given their cash-intensive business model of attempting to literally download and store the entire internet on private servers, they knew they might burn through the capital quickly. Their first order of business was securing office space, which they famously found in a Menlo Park garage owned by Susan Wojcicki. Wojcicki, then a manager at Intel, would soon become an early Google employee herself and eventually rise to become the CEO of YouTube.

With a physical headquarters established, Google was ready to unleash its product onto the wider world, stepping into a fascinating competitive landscape. To understand Google's triumph, one has to examine the state of search in 1998. The reigning heavyweight champion of search engine technology at the time was AltaVista. For anyone old enough to remember the pre-Google internet, AltaVista was actually considered quite good. Interestingly, AltaVista didn't emerge from a typical internet startup, but rather from the Western Research Laboratory of Digital Equipment Corporation (DEC) in Palo Alto—DEC's equivalent of Bell Labs.

DEC was a hardcore enterprise computing titan known for minicomputers and heavy hardware. AltaVista was born as a corporate research project specifically designed to showcase the blistering raw power of DEC's latest enterprise-class servers. The engineering breakthrough that defined AltaVista was parallel web crawling. Prior to AltaVista, all web crawlers operated on single-threaded processes—painstakingly crawling one page, finishing it, and moving to the next. Because the early internet was small, nobody had thought to parallelize the process. DEC realized that web crawling was an embarrassing parallelizable problem, which perfectly suited their massive enterprise appliances. This historical context highlights that while modern users associate Google's early dominance purely with PageRank and relevancy algorithms, the competitive vectors of search engines in 1998 were deeply rooted in raw indexing speed and architectural power.

[51:16](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3076s) The Critical Vectors of Search: Speed and Index Size

In evaluating the early days of search, two critical vectors emerged alongside relevancy: speed and index size. While modern users take instantaneous results for granted, search prior to Google's era was a slow, deliberate task. A user would kick off a query and then go do something else while waiting for it to return—much like interacting with deep-research AI models today where one might "send a query, you know, go get some lunch. Come back."

The size of a search engine's index was equally decisive. Before AltaVista introduced the parallelization of crawling, most indexes were incredibly small, rarely exceeding one million pages. Consequently, even the most advanced search algorithm was functionally useless if it only searched a tiny fraction of the live web. Furthermore, these search engines rarely updated their indexes; a claim of a million pages crawled was usually a cumulative total that simply stacked new sites over time without accounting for modifications or dead links.

* **AltaVista's Scale:** When AltaVista spun out of DEC (Digital Equipment Corporation) and launched as a commercial entity, its primary competitive edge was an index of 16 million pages. * **Information Retrieval Flaws:** Despite its massive index, AltaVista lacked relevancy and speed. Its algorithms relied heavily on keyword frequency, allowing webmasters to artificially boost rankings by spamming words like "dog food" in invisible text to create "the dog foodiest dog food page on the web."

[53:17](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3197s) The Costly Hardware and Broken Business Models of Early Search

Running a massive parallelized index required immensely expensive enterprise hardware, which AltaVista sourced directly from DEC. This highlighted a glaring industry problem: a high-quality search engine was prohibitively expensive to run, yet the market had no viable business model to support it. At the time, the sector relied entirely on low-priced, untargeted banner ads. Because of this financial mismatch, the broader tech market did not take search seriously as a standalone industry.

This structural dynamic created conflicting incentives inside DEC. The executives at DEC were not necessarily interested in pioneering the ultimate search engine; rather, they viewed AltaVista as a flashy proof of concept to help them "sell more of these boxes." Their ideal scenario was for external companies to witness AltaVista's capabilities, start their own search ventures, and buy massive amounts of high-end DEC hardware to power them.

[54:21](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3261s) The Era of Portal Juggernauts and Human Curation

In the late 1990s, algorithmic search remained an uninteresting industry because its economic upside felt strictly capped. Instead, users and investors flocked to directories and portals—with Yahoo serving as the undisputed market juggernaut. Yahoo went public in 1996 with a $1 billion market cap, and by the time google.com launched as a company in 1998, Yahoo had ballooned into a $20 billion public stock.

Founded by Stanford PhDs Jerry Yang and David Filo, Yahoo did not start as an academic or technology-driven research project, but rather as "Dave and Jerry's guide to the internet." It functioned as a hand-curated directory, resembling the Yellow Pages with rich, human-annotated descriptions explaining why a user should visit a specific site. During this era, the prevailing industry belief was that automated technology search engines would never replace human thought and curation. For the contemporary size of the web, this human-centric approach actually worked quite well.

* **The Scale Shift:** Curation was highly effective for a small web, but became entirely inefficient once the internet scaled 10,000 times larger, fracturing into hyper-specific niches that manual directories couldn't keep pace with. * **The Hybrid Model:** To cope with growth, portals like Yahoo adopted a hybrid approach. When a user ran a query, the top results were drawn from the hand-curated directory, while the lower portion of the page was backfilled by partnering with external algorithmic search engines. * **Media vs. Technology:** Yahoo operated fundamentally as a media company where search was just an afterthought "just in case." Their core business relied on human editors covering major categories to aggregate eyeballs and display banner ads, regardless of whether those ads were strictly relevant to the user's immediate search.

[57:29](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3449s) Google's Rejection of Clutter and the Scaling Crisis

Larry Page and Sergey Brin held an opposing philosophy to the messy portal model; they refused to clutter Google's homepage, insisting that their sole purpose was to help users find what they wanted and leave the site immediately. This stance raised a frightening existential question for early observers: if Google refused to keep users on-site to view banner ads, and was in fact fundamentally "dead set against using" ads at all, how could it ever make money? While Google was building the clear "front door to the internet" during a period where web traffic was exploding by 700% year-over-year, it had zero incoming revenue.

This lack of monetization created a dire infrastructure crisis. As usage grew exponentially, Google required massive amounts of capital to scale its crawling, indexing, and serving components—all of which scaled at different rates. To survive, they needed to pioneer an entirely new approach to systems architecture.

[59:16](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3556s) Recruiting Generational Talent to Fix a Broken System

Immediately after securing their angel investment round, Larry and Sergey went out into a fiercely competitive, pre-dot-com-crash talent market to recruit elite computer scientists. Because their technical challenges were so novel and exciting, they successfully landed legendary engineers who would go on to build Google's foundational infrastructure.

They recruited Urs Hölzle from academia as employee number eight, giving him the initial title of "search engine mechanic" because, quite literally, "everything was broken." Shortly after, they hired Jeff Dean away from a senior engineering role at DEC. Securing this level of elite talent during a booming market, well before Google had a proven business model, stands as a pivotal knife-edge moment in tech history that validated the sheer power of Larry and Sergey’s vision.

* **Urs Hölzle:** A Stanford PhD and UCSB professor who wrote the primary Java virtual machine used by Sun Microsystems. He ran all of Google's infrastructure from 1999 until 2023. * **Jeff Dean:** Described as "Google's Dave Cutler" and a "generational talent," Dean's extensive resume includes implementing the first version of AdWords, building AdSense, rewriting the core search pipeline five times, and co-inventing MapReduce, Bigtable, TensorFlow, and Gemini.

[01:02:06](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3726s) Distributed Computing on Cheap Commodity Hardware

Google was structurally forced into pioneering distributed computing because its core search engine required analyzing the entire text of a webpage to calculate PageRank and backlinks. The resulting index file was far too massive to fit onto any single machine, no matter how expensive or advanced. To solve this, Google's architecture broke the massive index down into highly tractable, 64-megabyte file chunks distributed across an immense web of separate disks and servers. A central master server maintained a mapping of these files, dynamically calling up only the specific chunks required to fulfill a given query in a highly parallel, concurrent manner.

With a file system designed to handle distributed data, Urs Hölzle realized they could completely bypass expensive enterprise hardware in favor of cheap, low-grade commodity components. Instead of relying on a flawless, high-end server, Google wrote intelligent software that assumed its underlying hardware would constantly fail, building redundancy directly into the system by replicating every piece of data three to five times. This difficult engineering challenge became a massive magnet for top-tier talent who wanted to solve complex architectural problems.

* **Hardware Failure Rates:** While the industry average server failure rate at the time sat around 3% to 4% per year, Google's cheap commodity hardware suffered a failure rate exceeding 10% per year, though the software layer rendered these failures invisible to the user. * **The World's Largest Computer Manufacturer:** By assembling raw individual components into custom setups rather than purchasing pre-built servers, Google rapidly scaled to technically become the largest computer manufacturer in the world.

[01:06:39](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3999s) Corkboards and Constraints: The Proto-Data Centers

The physical architecture of Google's early collocated data centers was chaotic and highly unconventional. To save money and maximize space, the engineers stripped away standard PC cases entirely. Instead, they mounted motherboards directly onto sheets of corkboard—which acted as an electrical insulator—stuffed in the RAM and hard drives, and slid the exposed setups directly into data center racks.

This extreme density was driven by a specific financial constraint: their colocation agreements charged them strictly by square footage, rather than by energy consumption or machine count. True to form, when given a constraint, Google's computer scientists optimized for it ruthlessly, packing an incredible density of raw hardware into the smallest physical footprint possible without any regard for traditional enterprise aesthetics.

[01:07:47](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4067s) Commodity Hardware and Google's Engineering Moat

Google's infrastructure was built on a strikingly messy and economical foundation: server racks stuffed with corkboards and cheap, commodity hardware strewn about rather than high-end enterprise equipment. This intentional choice allowed them to handle systemic complexity entirely in software. As search traffic surged and the web index expanded faster than any competitor's, this hyper-economical scaling mechanism allowed Google search to eventually achieve a staggering 87% gross margin once their business model kicked in.

The contrast between Google's makeshift operation and industry standards was starkly illustrated at their first co-location data center facility, a shared physical space in Santa Clara called Exodus. Google’s allocated cage sat directly adjacent to Inktomy, a competing search engine of the era. While the Inktomy cage boasted gleaming Sun Microsystems machines, pristine airflow, and immaculate cable management, Google’s setup looked like a "Frankenstein" monster. Because Google was only paying the facility based on square footage rather than power consumption, they sucked up massive amounts of electricity. Rumor has it that Google engineers even "borrowed"—or perhaps outright stole—a power circuit from the pristine Inktomy cage next door.

To cope with the high failure rates of consumer-grade hardware, Google's technical team had to design everything from the ground up, writing complex software layers to verify data integrity. For example, enterprise hardware utilizes a "parity bit" in RAM—an extra bit used for basic error checking by tracking whether data in a byte is even or odd. Consumer hardware lacks this feature, making machines cheaper by not sacrificing $\frac{1}{8}$ of the RAM to error checking, but leaving them vulnerable to data corruption from random environmental radiation flipping bits. Google's software engineers had to build custom verification systems to compensate for running on what they openly called "crap hardware." This absolute control over their infrastructure led to the creation of foundational systems like the Google File System (GFS) and MapReduce.

* **Gross Margin:** Google search eventually achieved an 87% gross margin due to infrastructure efficiencies. * **The Exodus Data Center:** Located in Santa Clara, where Google's "Frankenstein" cage operated right next to competitor Inktomy. * **Hadoop's Origins:** Yahoo later felt compelled to copy MapReduce to remain competitive, open-sourcing their version as Apache Hadoop. * **Global Speed:** By building their own cheap hardware, racks, and data centers globally, Google delivered search and ad results in roughly a quarter of a second, using these blazing-fast engineering metrics as a major brand "flex."

---

[01:12:24](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4344s) The Vague Series A Pitch Deck

All of Google's early infrastructure innovations grew directly from severe financial constraints; they simply had no money at Stanford and very little from their initial angel round. However, none of these technical feats would have mattered if they had not figured out a viable business model. In early 1999, running low on angel funds, Google hired Salar Kamangar (who would become employee number nine). As a recent Stanford undergraduate who had been blown away by the search product, Kamangar essentially banged down Google's door until they hired him.

Larry Page and Sergey Brin immediately handed Kamangar the most critical task in the company: writing the business plan and pitch deck for the Series A venture capital raise because they didn't want to do it themselves. The resulting pitch deck presented a three-pronged revenue model that was "handwavy as hell." It seemed designed to treat Google.com as an entity too precious to corrupt with immediate monetization, seeking instead to make money elsewhere to fund the core search engine.

The core of the proposed business plan was selling Google’s search technology to enterprises so corporations could search their own internal documents and intranets. This idea stemmed from the fact that Google had successfully been used to search internal Stanford internet systems. Furthermore, Larry and Sergey had already closed their very first revenue deal using this model, selling an enterprise search solution to the open-source Linux company Red Hat for $20,000.

* **The Series A Revenue Plan:** 1. **Enterprise Search:** Selling search technology to corporations (the primary projected revenue driver). 2. **Banner Ads:** Standard CPM banner ads, which were added reluctantly to appease VCs without a clear placement strategy on Google.com. 3. **Organic Search Licensing:** White-labeling organic search results to portals and directories as an OEM backfill, mimicking the business model of their data center neighbor, Inktomy.

---

[01:19:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4767s) The Legendary Series A Venture Round

Despite the somewhat harebrained nature of the business plan, it was the spring of 1999—the absolute height of the dot-com bubble—and venture capital was flowing freely. As legendary venture capitalist Michael Moritz later noted in Steven Levy's book, "nobody's feet were on the ground." Google possessed the exploding user engagement metrics that the era's market traded on. In a historic move, rival venture capital powerhouses Sequoia Capital and Kleiner Perkins decided to split the deal. Michael Moritz and John Doerr, two of the most celebrated VCs in the world, joined forces and both took seats on Google's board.

Larry and Sergey pulled off a massive public relations masterstroke by forcing these two fierce competitors to split the round, allowing each firm only a 12.5% stake in the company. Google held an in-person press conference featuring both Moritz and Doerr flanking the founders, who wore branded Google shirts. While Sequoia and Kleiner Perkins frequently split deals behind the scenes, framing this specific pairing as an unprecedented corporate marriage established a powerful narrative that Google was the hottest deal in Silicon Valley.

During the heated negotiations, Google even received a competing term sheet specifying a $150 million valuation instead of $100 million from a meeting set up by Ram Shriram. However, the founders were already "pot committed" to Sequoia and Kleiner Perkins, partially because Larry’s older brother had co-founded eGroups, which Moritz had previously funded. They closed the historic round at a $100 million post-money valuation, raising $25 million total. While a $100 million valuation for a Series A sounds quaint today, at the time it was a staggering headline equivalent to a pre-revenue startup raising at a multi-billion dollar valuation. Despite the massive cash influx, the venture capital playbook of the dot-com era demanded fast revenue to achieve a quick IPO, lighting a fire under Google to immediately find a genuine business model. Around this time, the founders met Omid Kordestani from Netscape, who initially approached them representing Netscape's interests.

* **Total Series A Raise:** $25 million. * **Post-Money Valuation:** $100 million. * **Board Seats:** Filled jointly by Michael Moritz (Sequoia) and John Doerr (Kleiner Perkins). * **Competing Valuation:** A late $150 million valuation offer was rejected to stick with chosen investors.

Omid Kordestani and the Early Revenue Struggle

Following America Online's (AOL) acquisition of Netscape, Omid Kordestani—Netscape's Vice President of Sales and Business Development—realized the environment there was becoming significantly less enjoyable. Sensing an opportunity, the formidable recruiting duo of Larry Page and Sergey Brin convinced Kordestani to join Google as Chief Revenue Officer. Tasked with turning Google's three-pronged business plan into a reality, Kordestani initially focused on selling their enterprise search model. However, he quickly discovered that customer pull for enterprise search was practically nonexistent at the time.

For the first six months of Google's existence following its venture capital funding, revenue remained painfully stagnant. The hosts recall a legendary, oft-debated quote from the era illustrating the rising anxiety of Google's backers. According to venture capital lore, Sir Michael Moritz of Sequoia Capital supposedly approached his partner Doug Leone and lamented, "Doug, we've never paid so much for so little." However, alternative behind-the-scenes accounts attribute the quote to Vinod Khosla of Kleiner Perkins, leading the hosts to joke that the line has become a "hot potato of quotes" that no one wants to officially claim. Regardless of who said it, the sentiment was undeniable: both venture firms felt they had egg on their faces after backing a Series A company at a massive $100 million post-money valuation that generated zero material revenue just as the dot-com bubble began to burst in early 2000. Despite having no business model, Google's market share was skyrocketing alongside a fiercely loyal, fervent fandom.

- Google's early business plan relied on three distinct avenues, spearheaded by Omid Kordestani. - Google commanded a $100 million post-money valuation for its Series A round without active revenue streams. - The dot-com bubble began bursting in the year 2000, compounding the pressure on Google to monetize.

Tim Armstrong, Text Ads, and the Fax Machine Era

Recognizing that slamming his head against the enterprise market was a losing battle, Kordestani shifted focus toward standard search advertising. He hired Tim Armstrong in New York to build out an east coast ad sales force, aiming directly at Madison Avenue ad agencies and direct corporate advertisers. However, Larry and Sergey imposed a strict technical constraint: all advertisements had to be text-only. While public narratives often frame this decision as a matter of "taste," it was primarily an obsession with page performance. The founders refused to let heavy images or banner ads slow down Google's lightning-fast load times.

Selling text-only advertisements proved incredibly difficult to Madison Avenue executives accustomed to vibrant visual media. At this stage, Google did not possess any self-serve web tools; ad placements were priced using a traditional Cost Per Mille (CPM) model, where advertisers paid per thousand impressions for specific keywords. The workflow was astonishingly retro: orders were not even negotiated primarily over the telephone. Instead, ad agencies sent their physical insertion orders to Google via fax. To accommodate this, Google's high-tech headquarters had to install physical fax machines so team members could manually enter the ad data and invoice clients based on page loads.

- Tim Armstrong was hired in New York to lead the ad sales push to traditional media agencies. - Ads were strictly limited to text to preserve website rendering speed and user experience. - The operational infrastructure of early Google advertising relied entirely on physical fax machines and manual order entry.

Engineering Intent: The Amazon Affiliate Breakthrough

To prove that text-based search ads could actually convert, Larry and Sergey tasked newly hired engineer Jeff Dean (who had just crossed over from DEC) with building the underlying ad-serving infrastructure without degrading the user experience. Working alongside Stanford undergraduate alum Marissa Mayer, Dean devised an ingenious method to validate the startup's monetization model using a prominent connection: Jeff Bezos. Because Amazon's founder owned a significant personal stake in Google, the search engine signed up as an Amazon affiliate.

Dean coded a system that dynamically cross-referenced live user search queries against the entire Amazon library catalog. If a user searched for a topic related to a book, Google dynamically generated a text ad containing an Amazon affiliate link reading, "Go buy this book at Amazon." While the 4% or 5% commission on a $10 book wasn't going to make Google rich, the experiment was a resounding success. It proved that search queries captured raw user "intent." By measuring the entire funnel—from ad impression to click-through rate, down to the actual purchase conversion on Amazon—Google generated ironclad mathematical proof that intent-based search ads yielded far higher conversions than random display ads scattered across the internet.

- Jeff Dean and Marissa Mayer designed the initial technical framework for Google's text ads. - Google used the Amazon Affiliate Program as a zero-risk testing ground to validate ad conversions. - The test established that capturing user intent solved a critical "math problem" for advertisers by providing superior click-through and down-funnel conversion rates.

The Netscape Deal and the Day Google Shut Down

While early ad sales brought in modest revenue, the true catalyst for Google's survival over the next year and a half came from Original Equipment Manufacturer (OEM) search portal partnerships with internet giants like Netscape, Yahoo, and AOL. In this era, portals controlled the vast majority of web traffic. Google's B2B model—essentially mirroring competitor Inktomi—involved white-labeling its organic search results to power these third-party portals in exchange for a fee. Crucially, Google required a small piece of branding at the bottom of these pages: "Powered by Google."

In late 1999, Kordestani leveraged his old Netscape connections to secure a proof-of-concept deal to backfill organic search results for Netscape's homepage directory. When Google flipped the switch to go live, the sheer volume of traffic from Netscape's millions of users threatened to instantly obliterate Google's fragile server infrastructure. Sergey Brin called Kordestani in a panic, warning him that the system was about to tip over. Faced with a potentially company-killing reputational disaster, Google made the radical decision to entirely shut down `google.com` for the day, routing 100% of their server capacity to prioritize Netscape's traffic until they could provision more machines. This temporary sacrifice paid off massively, introducing 3 million searchers a day to Google's speed and quality, and establishing the "Powered by Google" logo as the coveted "Intel Inside" ingredient brand of the internet.

- The Netscape portal deal instantly added 3 million daily search queries to Google's infrastructure. - Google took its own homepage offline in 1999 to guarantee server stability for its white-label partner. - The "Powered by Google" attribution acted as a massive growth loop, driving users away from traditional portals and directly to Google's standalone search engine.

Sponsor Break: Statsig

The hosts take a brief detour to highlight their sponsor, Statsig, drawing a parallel between Google's corporate legacy and modern product development. Google pioneered the modern software archetype, introducing engineering-first cultures, cushy campuses, and 20% time. However, they also pioneered internal data infrastructure, creating the very first advanced analytics dashboards, A/B testing frameworks, and feature controls that allowed thousands of engineers to deploy product updates safely from the bottom up.

For 15 years, subsequent tech companies spent immense resources rebuilding this internal data stack from scratch. Today, modern giants like OpenAI, Anthropic, Figma, Notion, and Brex bypass that engineering headache entirely by using Statsig. Statsig provides a standalone, warehouse-native suite of tools covering statistical experimentation, feature flags, session replays, and analytics. Operating at a scale that processes trillions of events per day, Statsig plugs directly into existing data infrastructure like Snowflake or BigQuery, allowing teams to focus on building features rather than troubleshooting metric definitions.

- Statsig offers product teams a complete experimentation and data stack out of the box, mimicking systems built internally by early tech giants. - Clients using the platform include prominent AI and SaaS enterprises such as OpenAI, Anthropic, Figma, and Notion. - The platform is warehouse-native and features a generous free tier alongside a $50,000 startup program.

The Upcoming Yahoo Landmark

Returning to the historical narrative, Ben and David set the stage for Google's ultimate ascension. By June of 2000, despite international portal acquisitions, Google still possessed very little real revenue. However, everything changed when they managed to sign a landmark partnership deal with the absolute "big kahuna" of the internet landscape: Yahoo. This monumental contract was inked precisely as the broader macroeconomic environment began to implode and the dot-com bubble burst in spectacular fashion.

[01:39:22](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5962s) The Lifeline: Yahoo's $10 Million Portal Deal

Following the NASDAQ crash in March 2000, Google secured a critical lifeline in June of that same year by signing a major deal with Yahoo. Under this agreement, Google took over all organic search result backfills on Yahoo.com, clearly branded with a "powered by Google" tagline. As a core component of the arrangement, Yahoo invested $10 million directly into Google. The speakers emphasize that this arrangement completely saved the young company, providing the financial runway and revenue necessary to survive the bleak years of the "dot winter" while they figured out their eventual AdWords business model.

The impact of the Yahoo deal was immediate and massive. On day one in June 2000, Google's traffic doubled to 14 million searchers per day, marking a material portion of overall web traffic just one year after their prior Netscape deal. By 2001, the partnership's first full year, Yahoo paid Google $7.2 million for organic search results. The speakers notes that this revenue—combined with the $10 million investment and other portal deals like Netscape—effectively bridged the company through a period where raising more venture capital was entirely impossible.

- **Venture Capital Drought:** The speakers highlight how overlooked this period is, noting that the "venture capital gravy train was over" and Google had no option to raise more money. - **Profitability Pressures:** Unlike modern startups that can operate for years without profit expectations, Google was only two years into its life but faced an incredibly expensive business to run due to heavy infrastructure and personnel costs, making immediate revenue an absolute necessity.

[01:41:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6087s) The Hunt for a "Professional" CEO and the Steve Jobs Ultimatum

As Yahoo came online, Google’s board heavily pressured Larry Page and Sergey Brin to bring in outside leadership. During the Series A funding process, investor John Doerr had "very begrudgingly extracted a promise" from the co-founders to hire a "professional CEO." Larry had been serving as CEO during the Series A and remained in the role for the next couple of years. Reluctant to give up control, the founders dragged their feet, taking a full 16 months to find a replacement—a delay the speakers joke was less about a difficult search and more about seeing how long they could get away without one.

To convince them, venture capitalists from Kleiner Perkins and Sequoia Capital ran a standard Silicon Valley playbook: taking the founders on a tour to meet the CEOs of great public tech companies to show them what professional leadership could achieve. Larry and Sergey went, but they returned entirely unimpressed. After months of touring, they gave the VCs a defiant ultimatum: they would only hire one specific person they met who actually met their exceptionally high bar: Steve Jobs.

- **The Steve Jobs Prank:** Jobs had recently returned to Apple from NeXT. While the founders likely would have accepted him if he miraculously agreed, the speakers view the demand as a way of "thumbing their nose at the VCs," essentially setting a bar of "Steve Jobs or nothing." - **Historical Irony:** The speakers chuckle at the deep irony of the founders idolizing Jobs, given the intense corporate warfare that would erupt between Apple and Google just ten years later.

[01:43:42](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6222s) Eric Schmidt Joins the Triumvirate

After a highly contentious process lasting up to 18 months, Eric Schmidt emerged as the only viable candidate acceptable to both sides. Schmidt ticked every box on the venture capitalists' checklist while remaining palatable to the founders; he was a legitimate engineer who had worked at Sun Microsystems, a proven business person, and the former CEO of a public company, Novell. Famously, he also fit the cultural Venn diagram perfectly because he went to Burning Man, just like Larry and Sergey. Schmidt officially joined Google in March 2001.

While Larry and Sergey initially harbored some resentment about the forced hiring, the arrangement ultimately worked incredibly well. Larry quickly realized that Schmidt could handle the burdensome operational aspects of a rapidly growing company—such as running the finance organization—which freed the founders up to focus on what they loved. Rather than putting the founders "out to pasture," the three formed a tight, daily standing meeting. Schmidt served as CEO, Larry became President of Products, and Sergey became President of Technology, creating a highly effective three-person leadership team.

- **The Ultimate Office Mate:** To avoid "organ rejection" from the uniquely engineering-heavy culture, Schmidt took a collaborative approach. On his first day, due to lack of space, an engineer moved into Schmidt's office to camp out with him, a classic example of early "Googliness" that acted as a successful acid test for the new CEO. - **Schmidt's Big Bet:** Joining Google in 2001 was a significant risk for a public company CEO. Although the Yahoo deal was secured, startups were deeply out of favor, and it was far from certain that Google would become the massive home run it eventually became.

[01:45:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6318s) Defining "Googliness" and the Altruistic Money Machine

The conversation shifts to defining the core culture of early Google, often referred to as "Googliness." The speakers describe it as a utopian, campus-like environment that felt more like a high-powered university than a traditional corporation. The company intentionally hired incredibly brilliant, high-IQ, but inexperienced people "who don't know what they don't know," allowing them to attempt radically novel and collaborative approaches to complex problems. The guiding operational philosophy of the entire company was defined by a "healthy disregard for the impossible."

This ambitious culture birthed their iconic 1999 mission statement, introduced in their very first post-financing press release: "to organize the world's information and make it universally accessible and useful." The speakers analyze how perfectly this statement scaled over time. It was broad enough to be inspiring and altruistic, which attracted top-tier global talent, yet narrow enough to provide a framework for what would eventually become an unprecedented "money printing machine."

- **The User-First Philosophy:** The speakers highlight a telling quote from Google's later IPO prospectus: "We believe that the most effective and ultimately the most profitable way to accomplish our mission is to put the needs of our users first." - **The Core Trifecta:** Google successfully balanced an altruistic mission, an incredible monetization model, and a strict commitment to user needs.

[01:48:47](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6527s) The Four Fatal Flaws of AdWords v1

By the spring of 2001, even though the ship had been stabilized by Yahoo's cash, Larry and Sergey turned their attention back to their advertising product and were deeply dissatisfied with its current state. Advertisers were happy, but the system suffered from four fundamental flaws that prevented it from scaling. First, ads were manually sold "by hand on Madison Avenue," meaning the pool of advertisers was tiny compared to the massive flood of daily user search queries. Second, because sales were manual, scaling the business required an unsustainable, linear scaling of the sales workforce, which threatened to make Google look like a legacy portal like Yahoo.

The third issue struck at the very heart of Google's user-experience philosophy: there was absolutely no notion of ad quality. While Google prided itself on providing the highest-quality, most efficient organic results, the ads lacked a mechanism to ensure relevance. Fourth, Google was leaving massive amounts of money on the table. They were selling highly valuable user intent keywords on a flat, arbitrary CPM (cost per thousand impressions) basis, meaning they weren't participating in the true economic value created, pricing ads based on a metaphorical "finger in the air." By the fall of 2000, they resolved to completely overhaul the system.

[01:51:13](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6673s) The "Instagram Stories" Moment: Learning from Bill Gross and Goto.com

Google's eventual overhaul—AdWords v2—was heavily inspired by an outside innovator, a move the speakers call "Google's Instagram stories moment." The pioneer was a company named Overture, originally launched as Goto.com by Bill Gross out of his Ideal Lab startup incubator. Gross introduced the company with immense flair at the TED conference in February 1998. At the time, early search engines were suffering from rapidly declining quality because bad actors were gaming basic keyword-matching algorithms using "keyword stuffing, white text on a white background," and redirecting users to irrelevant porn sites to hijack traffic.

Gross proposed a radical solution: let the free market determine relevance using dollars. He argued that the entity willing to pay the most for a keyword was highly likely to be the most legitimate, high-converting business capable of solving the searcher's pain point, while malicious spammers wouldn't be able to afford the bids. This mirrored the paid-inclusion philosophy of the physical Yellow Pages. While Larry and Sergey solved ranking mathematically on the organic side, Gross solved it financially on the paid side, creating a pure pay-per-click system that didn't even develop its own organic search technology, choosing instead to license backfills from companies like Inktomi.

- **The TED Stage Backlash:** Gross's idea of letting commercial interests dictate search ranking was so controversial in 1998 that some audience members actually booed him on stage at the TED conference. - **The Transparent Bidding Page:** The original Goto.com layout was completely transparent; it displayed paid results strictly in order of who bid the most, explicitly showing the exact price—such as 21 cents, 23 cents, or 24 cents—the advertiser was paying for that specific user click. - **Inventing the Pay-Per-Click Model:** Gross recognized that the traditional advertising model of buying impressions (CPM) was broken. Operating multiple startups at Ideal Lab, he hated paying for empty impressions and realized the internet allowed for performance-based advertising where businesses only paid when a user actually clicked. - **The Wanamaker Paradigm Shift:** This shift solved the famous historical advertising dilemma coined by John Wanamaker: "Half the money I spend on advertising is wasted. The problem is I just don't know which half."

[01:55:23](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6923s) The Economics of Cost Per Click vs. CPM

The shift from cost-per-thousand-impressions (CPM) to cost-per-click (CPC) fundamentally changed the landscape for advertisers. In a conversion-oriented mindset, an advertiser only cares about the actual click rather than mere exposure. The nuance lies in the environment: while CPM works perfectly fine for high-level brand-building situations, search engine users represent a high-intent audience where advertisers want to capture direct intent rather than distribute passive awareness. Under a CPC model, advertisers essentially receive free exposure every time an ad is displayed but not clicked.

This model proved to be incredibly lucrative right out of the gate. GoTo generated a staggering \$100 million in revenue in its very first year, vastly outperforming Google's early financial metrics and solidifying paid search as a viable business model. Crucially, the entire system operated on a self-serve website. Advertisers could log in, place a bid, and participate in a real-time auction where the highest bidder was displayed at the very top of the search results—a mechanism that sounds deeply familiar to anyone who has used modern Google advertising tools.

- GoTo achieved \$100 million in revenue in year one. - By mid-1999, GoTo had scaled to 8,000 advertisers. - In contrast, Google AdWords launched its beta program in October 2000 with only 350 advertisers.

---

[01:57:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7020s) The "Heresy" of Paid Search and the Lack of Patents

A fascinating historical wrinkle in GoTo’s story is their lack of patents for the core auction and pay-per-click mechanics. Bill Gross, the brilliant and reflective founder of GoTo, viewed the system as entirely obvious—convinced that billing per click via an auction was simply the natural way things should be done. Although lawyers flagged the lack of patents right before the company went public, GoTo was already outside the legal window of eligibility. Gross had publicly shared the mechanics on stage at the TED conference more than a year prior, rendering the ideas ineligible for patent protection.

The concept of paid search faced immense cultural resistance in the mid-1990s. An earlier search engine called Open Text attempted paid search results in 1996, but the early internet ethos was heavily rooted in an academic, utopian community that viewed commercialization as an unforgivable intrusion. Open Text was immediately "kneecapped" by users who viewed the move as heresy. This backlash created an industry-wide hangover, making competitors treat paid search as a "third rail" that internet users would never tolerate. Even Gross was booed on stage at TED in 1998 for promoting the model.

- Open Text attempted paid search results in 1996 but faced overwhelming user backlash. - GoTo missed its patent window because Bill Gross presented the ideas at a TED conference over a year prior to filing. - The early web community fiercely protected organic results from "paid inclusion."

---

[01:58:59](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7139s) The Evolution of Search Monetization and the Rise of Overture

The great irony of the early pushback against paid search is how the market ultimately evolved. When Google eventually rolled out AdWords V2, they took pains to separate the "paid icky world" from organic search by utilizing a distinctly colored sidebar explicitly labeled "sponsored." However, modern search engine layouts look exactly like Bill Gross's original vision: a single column of unified results where the top positions are sponsored ads, which are often labeled even less conspicuously than GoTo originally labeled theirs. By the time the dot-com bubble fully inflated, the internet had become commercialized enough to accept Gross's model, establishing GoTo (later rebranded as Overture) as the definitive monetization blueprint for the web.

Overture soon realized they could scale their model by bringing paid search to web portals, mirroring Google’s strategy with organic results. Transitioning into a powerhouse B2B white-label provider, Overture began powering ads for Dogpile and MetaCrawler before landing massive deals with industry titans like AOL, MSN, and Yahoo. While Google was still picking up "fun pennies on the ground" by asking for organic search placement fees, Overture was executing massive revenue-share deals. Their Yahoo partnership alone was a \$100 million deal, eventually scaling to the point where Overture powered roughly 75% of Yahoo's overall revenue. This deep financial dependency forced Yahoo to acquire Overture for \$1.6 billion in a bidding war against Microsoft.

- Overture eventually won a \$360 million settlement from Google for infringing on its smaller, self-serve tool patents. - Overture's white-label deal with Yahoo was worth \$100 million and grew to account for roughly 75% of Yahoo's revenue. - Yahoo acquired Overture for \$1.6 billion following a competitive bidding process with Microsoft.

---

[02:02:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7323s) GoTo's Attempted Acquisition of Google

Before Google built its own dominant ad machine, GoTo actually attempted to acquire the young search engine. Bill Gross envisioned a "match made in heaven": Google possessed PageRank, the absolute best algorithm for serving highly relevant organic results for informational, non-commercial queries, while GoTo possessed the premier monetization system for commercial searches. Marrying the two systems would create a unified search experience that could perfectly surface both paid and organic results.

Before raising their major venture round from Sequoia Capital and Kleiner Perkins, Larry Page and Sergey Brin approached Gross and floated an acquisition price of \$200 million. Gross thought the price was fair—offering the founders a valuation double what they were seeking from venture capitalists. However, when Gross brought the deal to the Overture board, the company was valued at \$2 billion. The board rejected the acquisition, refusing to dilute 10% of their highly valuable, revenue-generating company to buy a tiny startup with zero revenue. Had the deal gone through, it is highly unlikely that Google would have ever become the independent tech giant it is today.

- Larry and Sergey proposed a \$200 million acquisition price to Bill Gross. - The Overture board rejected the deal because Google had zero revenue and they viewed it as too dilutive to their \$2 billion valuation.

---

[02:03:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7430s) The Birth of Google's AdRank

By the fall of 2000, Larry and Sergey realized they had missed a massive opportunity by not adopting the GoTo model from the inception of AdWords. While the founders were unquestionable academic and engineering geniuses who built PageRank from scratch, their secondary superpower was their willingness to adopt the best ideas from outside their walls and run with them. In October 2000, they tasked Salar Kamangar with overhauled AdWords. The immediate priority was building an automated, self-serve system to replace Google’s slow, manual ad-ordering process and expand their advertiser base to match Overture's scale.

> "Click-through rates are a proxy for relevance."

As Salar and Larry conceptualized a uniquely "Googly" approach to paid search, they sought a way to ensure ad quality rather than building a pure pay-to-play system where the highest bidder always won. They realized they could track click-through rates (CTR) as a direct proxy for consumer relevance. By multiplying an advertiser’s maximum bid by their ad's click-through rate, Google birthed "AdRank." This formula aligned all incentives: users only saw highly relevant ads, great advertisers were rewarded with lower actual prices for maintaining high click-through rates, and Google mathematically optimized its own revenue by maximizing expected value.

- Venture capitalist Tim Draper famously validated Overture's self-serve functionality during a pitch by live-bidding on the keyword "VC" and getting into a penny-bidding war. - AdRank combines two primary signals to determine ad placement: the bid amount and the user click-through rate.

---

[02:09:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7784s) The Great Self-Serve Click Arbitrage and Second-Price Auctions

When Google launched the first iteration of its self-serve AdWords system in late 2000, it included the AdRank quality score but omitted the actual CPC billing and auction mechanics—meaning advertisers were still paying on a traditional CPM basis. Savvy early advertisers quickly discovered a massive loophole: they could manually click on their own ads repeatedly to artificially inflate their click-through rates. Because higher CTRs boosted their AdRank, their ads were displayed more frequently, and since they weren't paying per click, this internal clicking cost them nothing. For a few months, this tactic became one of the greatest arbitrage loops in internet history.

By 2001, Google closed this gap by fully adopting Overture's cost-per-click and auction model, but they introduced a brilliant structural upgrade: the second-price auction. In this system, the winner of the auction does not pay their maximum bid; instead, they are only charged one penny above the bid of the second-place finisher. For example, if a winning advertiser bids 30 cents and the second-place competitor only bids 20 cents, the winner only has to pay 21 cents for the click.

[02:11:45](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7905s) The Second-Price Auction and Storing Value

The mechanics of Google’s second-price auction mean that if a top bidder specifies a maximum bid of a dollar and the next highest competitor only bids 50 cents, the top bidder wins but only pays 31 cents. While this might appear as though Google is foolishly leaving money on the table, it is actually a calculated, long-term strategy. By not gouging their clients, Google builds lasting trust and removes the need for advertisers to constantly monitor and tweak their bids. This approach ties into a broader concept that every truly great company carries a "stored potential energy of value maximization" that it chooses not to fully exploit—with Costco being the most extreme example, and Google operating on a similar wavelength.

When Google rolled out this model at the beginning of 2002, explaining it to traditional advertisers was a massive hurdle. Clients were accustomed to simply writing a check and sending a fax to purchase a guaranteed block of impressions; transitioning them to a dynamic auction felt like entering a confusing morass. Sheryl Sandberg, who had just joined the company to manage ads, found herself "banging her head against the wall" trying to pitch the concept. In frustration, she called her mentor, former US Treasury Secretary Larry Summers (for whom she had served as chief of staff). Summers immediately recognized the system as a Vickrey second-bid auction, noting that it was backed by extensive economic literature and was the exact method the Federal Reserve used to sell Treasury bonds. Armed with this institutional validation, Sandberg used the explanation to finally get the message through to skeptical advertisers.

- **Project Sunset:** The internal code name for the painful, miserable process of sunsetting advertisers off the old fixed-rate system and transitioning them to the new AdWords pricing model. - **The Overture Contrast:** Google's competitor Overture had actually discovered the importance of click-through rates earlier. However, when Overture implemented it, their transparent pricing page became disordered, confusing advertisers who could no longer figure out what to bid for the top spot. Rather than absorbing the temporary pain of user confusion, Overture abandoned the feature and reverted to a simpler model, missing a crucial industry shift. - **Infrastructural Scale:** Executing dynamic second-bid auctions in real time for every single search query required an immense technical infrastructure that only Google possessed in 2001, utilizing their proprietary commodity hardware, distributed file systems, and scaling data centers.

[02:16:10](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8170s) Financial Explosions and International Flight paths

The financial stakes of transitioning to the new AdWords model were incredibly high. In 2001—the same year a partnership deal with Yahoo saved the business—Google generated $86 million in revenue and $10 million in profit. While these were stellar metrics for a startup navigating the dot-com winter, almost all of that money came from legacy ad systems and portal deals. Implementing Project Sunset meant putting a massive portion of their existing revenue at risk. It was a tenuous period, but the transition ultimately proved to be a spectacular success. By 2002, Google's revenue skyrocketed to $440 million, representing a five-to-six-fold increase in a single year.

Simultaneously, newly arrived CEO Eric Schmidt discovered a glaring inefficiency: 50% of Google's search traffic originated from outside the United States, yet the company had zero international ad sales. In a move that was almost a joke, Schmidt told executive Omid Kordestani to simply go to the airport on Monday morning, wait for a phone call directing him where to buy a ticket, and figure it out from there. Kordestani spent the entire year on airplanes, seeding small, autonomous startup teams across various geographies to sell the new AdWords system.

- **International Revenue Growth:** International sales went from 0% to 18% of Google's revenue in year one, grew to 22% in 2002, reached 29% in 2003, and accounts for roughly half of Google's business today. - **The China Market:** Google maintained a business footprint in China following early market friction in 2002–2004, until famously withdrawing from the country in 2010 rather than complying with search result censorship.

[02:21:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8490s) The Farce of the Yahoo Offer and Betting the Company on AOL

By mid-2002, Google realized that paid search via AdWords was performing so phenomenally well that they needed to upend their entire B2B strategy. Instead of charging web portals to display organic search results, Google decided they should be paying the portals for the right to host paid search fields and split the resulting ad revenue. This shift stands in stark contrast to late 2001, when Yahoo CEO Terry Semel sat down for dinner with Larry Page and Sergey Brin. Semel, assuming Yahoo was Google's lifelines because they paid them less than $10 million for organic results, told them, "You don't have a business, do you?" Semel offered to buy Google for $1 billion on the spot. By the summer of 2002, Yahoo saw the explosive power of AdWords, returned to the table with a $3 billion offer, and was promptly rebuffed when Google counter-offered with a dismissive $5 billion price tag.

Instead of selling out, Google went after America Online (AOL), which boasted 34 million users and had its search ecosystem split between Inktomi and Overture. Google won a landmark deal to take over the entire search and ad experience for AOL, convincing them to direct all search advertisers to a Google-hosted URL. To secure this company-making deal, Google had to agree to aggressive terms: sharing 85 cents of every dollar back to AOL, granting them warrants to buy 7.44 million shares of Google stock at $3 per share (a $22 million investment option), and providing a staggering $100 million revenue guarantee.

> "We could have gone bankrupt." > — *Sergey Brin, reflecting on the extreme financial leverage of the AOL contract.*

The $100 million guarantee was a terrifying obligation because Google did not actually have that cash in May 2002. It was a literal "bet the company" moment; if the ads failed to perform or monetize properly on AOL's network, the fixed dollar obligation would have driven Google straight into bankruptcy. The decision caused intense friction within the executive team, but the founders refused to back down. When Omid Kordestani explicitly warned them that they were wagering the entire firm's survival on a single contract, Larry Page characteristically replied: "We should be able to monetize the pages. If not, we deserve to go out of business."

[02:28:07](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8887s) Market Liquidity and the AOL Jackpot

The onboarding of AOL’s search ad inventory into Google’s system proved to be an immensely valuable strategic move, triggering what the hosts describe as a "market liquidity thing." Drawing a parallel to their previous episode on Meta—where CTO Andrew Bosworth argued to Mark Zuckerberg and Sheryl Sandberg that more ad inventory inherently yields better-performing ads—the hosts explain that an abundance of inventory is crucial. In a dynamic ranking and targeting system, having a massive pool of choices allows the algorithm to select and serve the single best candidate for a user's query.

If Google had launched its ultimate AdWords business model from a cold start without this volume, it would have faced a severe inventory problem. A thin pool of advertisers makes it nearly impossible to deliver a high-quality user experience. The hosts debate the word "magical," joking that while search ads provide no user "delight whatsoever," they are the least offensive form of advertising. Economically, however, they represent "the most magical economic transaction" ever discovered, allowing advertisers to reach the exact right person at the moment of highest intent. This unparalleled model logically explains why Google generates more profits than almost any other company.

* **The Inktomi Warning:** Inktomi commented to the *Wall Street Journal* regarding the deal, stating, "They'll learn over time that Google takes your users. It doesn't help you build your property." * **Financial Returns:** The massive \$100 million gamble paid off immediately. AOL made \$35 million in the first half-year of 2002, and \$200 million in 2003, completely blowing past Google's minimum guarantee. * **Overnight Dominance:** This single deal transformed Google into a dominant force in the paid listings market, effectively stealing the inventory from Overture, who had previously partnered with AOL.

[02:31:41](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9101s) Sheryl Sandberg's Army and the Logic of Scale

This era marks the rise of Sheryl Sandberg at Google. Having joined around the same time as Eric Schmidt, Sandberg was originally a business unit manager looking for a substantial project. Omid Kordestani approached her with a massive challenge: Google had just signed the AOL deal and suddenly possessed thousands of new advertisers across completely unfamiliar categories with zero infrastructure to service them. Sandberg was tasked with building an adaptable, highly intelligent "army of people" to manage the influx and eventually feed technical ideas back into the engineering organization to scale the operation. Her success in building out the AdWords sales function for this "monster AOL deal" served as her definitive launchpad before she eventually became the COO of Facebook.

This massive win gave Google the license to play heavy offense on traffic acquisition across the entire internet. They realized that if they could embed Google search—both paid and organic—into every corner of the web, they would build an unstoppable liquidity pool. The economic logic of search reveals that it is uniquely a winner-take-all market. While standard economies of scale allow Google to amortize fixed infrastructure costs, scale also creates an incredible secondary effect: as Google's user base grows, its revenue per unit actually increases.

* **Increasing Returns to Scale:** More bidders on any given keyword lead to better price discovery, pushing the winning bid higher than it would be in a smaller, thinner marketplace. * **Monetizing the Long Tail:** With a vast network of advertisers, Google can monetize rare, highly specific search queries that would otherwise yield no ad revenue, ensuring that almost any search results in money. * **The Virtuous Cycle:** More distribution drives search volume $\rightarrow$ search volume drives higher keyword bids $\rightarrow$ higher bids increase auction prices $\rightarrow$ higher prices generate massive revenue $\rightarrow$ massive revenue allows Google to pay even more for further distribution.

[02:36:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9410s) Playing Irrational Defense Against Microsoft

Recognizing this virtuous cycle, Google abandoned any reliance on passive, organic growth, choosing instead to aggressively gobble up the market before competitors could catch on. Their primary tactic was offering massive revenue shares to distribution partners—sometimes up to 100% or even more. Because Google monetized individual searches far better than anyone else, their payouts were inherently larger. If a competitor tried to match a 70% split, Google’s 70% absolute dollar amount was vastly superior. Eventually, it became economically impossible for smaller engines to compete without running out of money.

Google’s private company status gave them a "secret weapon" against their public counterparts. While public search engines were constrained by quarterly earnings consistency, Google could "irrationally overpay" for distribution and swallow a bad quarter for long-term dominance. They possessed a deep conviction that search was destined to be a half-trillion-dollar annual market, making it worth any amount of upfront investment to ensure they were first and biggest. This aggression became vital because they lacked their own web browser, and Microsoft—the owner of Internet Explorer—loomed as a terrifying threat that could easily redirect internet traffic if they discovered the same monetization secret.

[02:40:33](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9633s) The Strategic Genius of the Google Toolbar

To circumvent the fact that users had to manually type in "google.com," Google launched a nostalgic "blast from the past" in December of 2000: the Google Toolbar. While the public viewed it as a benevolent gift from the "benevolent Google gods," it was a calculated piece of strategic defense and offense. Acting as a browser extension long before modern extensions existed, it placed a permanent search box directly inside Internet Explorer, underneath the URL bar. This prevented Microsoft from easily cutting off Google's traffic at the browser level.

The economics behind the toolbar were staggering. Once a user installed the Google Toolbar, they averaged seven times the number of searches they did previously, making them seven times more valuable to the business. At the time, an average Google user generated roughly \$2 in annual revenue, but a toolbar user generated upwards of \$10. This delta gave Google an \$8-per-user acquisition budget to play with. Because average revenue per user (ARPU) was skyrocketing toward future values of \$20, \$50, and \$100, Google could aggressively pay software developers to bundle the toolbar. They paid every partner they could find—including major players like Adobe—to automatically check the box to install the Google Toolbar alongside their standard application installers.

[02:43:46](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9826s) The Hyperaggressive Distribution of Google Toolbar

Google grew its user base through incredibly aggressive, strategic distribution deals that are often overlooked. At the center of this strategy was the Google Toolbar. Google paid massive sums of money to software makers like Adobe (for Flash/Acrobat downloads), Real Networks, and WinZip to bundle the toolbar as a payload with their software downloads. The hosts point out that Google was essentially paying these developers to include what could technically be described as "adware, spyware, or a Trojan horse" that lived inside Internet Explorer to maximize Google's search revenue. Despite the aggressive methodology, users genuinely loved the toolbar because it gave them direct, valuable access to Google search.

To provide even more incentive and make users "sticky," Google integrated a pop-up blocker directly into the toolbar at a time when pop-up ads were ruining the web experience. They scaled this distribution by striking a direct deal with Dell to pre-install the Google Toolbar on Internet Explorer for all new Windows PCs shipping out. Furthermore, they secured a landmark deal to become the default search engine in Firefox just as the browser was rising in popularity, providing Mozilla with its primary source of revenue for decades.

* Google bundled its toolbar with popular desktop applications from Adobe, Real Networks, and WinZip. * A pop-up blocker was integrated into the toolbar to incentivize user adoption. * Dell agreed to pre-install the toolbar on Internet Explorer for newly shipped PCs. * Google became the default search engine for Firefox, anchoring Mozilla's revenue for decades.

[02:45:34](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9934s) Monetizing Google Earth and the Strategic Ballet

The acquisition of Google Earth provides another classic example of how Google leveraged distribution over direct monetization. Because Google was an ad-based business, internal conversations initially cropped up regarding how to inject advertisements directly into the virtual globe. However, they realized that Google Earth—which was a standalone desktop program at the time, completely separate from the "lame" 2D driving directions of the web-based Google Maps app—was going viral. People loved zooming in on their houses in 3D.

Instead of cluttering the beautiful interface with ads, Google simply bundled the Google Toolbar with every single Google Earth download. The hosts laugh about how the toolbar "way more than paid for itself," entirely eliminating the need for ads within the application. By the mid-2000s, Google's Average Revenue Per User (ARPU) grew to 10, 20, and 30 dollars, fueled by the fact that toolbar users were vastly more sticky than non-toolbar users. This immense revenue gave Google an unprecedented budget to continually acquire more users—a playbook that extends all the way to the tens of billions of dollars Google pays Apple today for Safari traffic acquisition costs.

* **Google Earth vs. Maps:** Originally separate; Maps was for driving directions, while the desktop Earth application was a viral, 3D interactive globe. * **ARPU growth:** Reached $10, $20, and $30 in the mid-2000s due to highly sticky toolbar users. * **The "Costco" Comparison:** The hosts compare early Google to their Costco episode, describing it as a "beautiful ballet where every piece of the Google business model works together and reinforces the other pieces." * **Google Desktop:** Another attempt at stickiness that crawled a user's local computer files and privately integrated those results into web searches. * **Sundar Pichai:** Hired as a new Product Manager in 2004, he took over the applications client team that managed the Google Toolbar.

[02:48:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10124s) The Birth of AdSense and Content-Based Targeting

As Google grew, they looked for ways to extend their strategic business model beyond the search results page. This led to the creation of AdSense, which became Google's second massive business line after AdWords. Prior to AdSense, Google could only make money when a search query occurred. While a search query represents an incredibly high-intent, high-value moment, queries don't actually happen that often. Users spend a far greater share of their time consuming static content on third-party websites across the internet where Google initially had no footprint.

Leveraging the infrastructure they had already built for PageRank—specifically their algorithmic ability to analyze organic search content, predict click-through rates, and target ads—Google realized they did not need a live search query to serve effective ads. Instead, they could run their algorithms on static publisher pages, analyze the content, and "reverse serve the keywords" from their massive existing pool of AdWords advertisers. While it shifted the model from targeting pure user *intent* to targeting page *content*, it successfully captured users who likely held loose or future intent related to what they were reading.

* **The Atomic Unit:** Prior to AdSense, the atomic unit of Google's business model was the search query. * **Inventory Constraints:** In February 2003, Google was heavily supply-constrained; advertiser demand to serve ads far exceeded the available search queries, making an expansion to third-party sites highly lucrative.

[02:51:39](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10299s) Jeff Dean's Six-Week Build and the Publisher Boom

In a characteristically legendary engineering feat, Jeff Dean coded and built the entire AdSense system in just six weeks. Google first tested the product internally on Google Groups before trying it on true third-party websites. To test it externally, Google bought display ad space from publishers and turned those visual banners into a text-based window displaying three horizontal AdWords search results. Their absolute favorite testing ground was `howstuffworks.com`, because its highly informational articles perfectly mirrored the high-intent, highly commercializable queries advertisers were already bidding on.

With Susan Wojcicki stepping in as the product manager, AdSense exploded into a massive revenue generator. Though it operated at a lower margin than first-party search pages because Google shared the majority of the revenue with the creators, it immediately brought in hundreds of millions of dollars. For small internet publishers in 2003, AdSense was revolutionary; they could simply drop in a snippet of HTML, and Google would automatically deposit money into their bank accounts. The hosts note that this liquidity and automated revenue sharing served as the direct spiritual precursor to the modern YouTube creator economy.

* **Speed to market:** Built by Jeff Dean in just six weeks and launched near the end of Q1 2003. * **Revenue Milestones:** By the end of 2003, AdSense was pulling in over 1 million dollars a day in revenue. * **Revenue split:** Reports vary between a 67% and 80% revenue share given to the website publishers, with Google keeping the smaller percentage as a "spiff."

[02:54:52](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10492s) 20% Time, Evolving Algorithms, and Custom Infrastructure

As 2003 came to a close, Google was effectively no longer capital constrained, allowing them to rapidly iterate on side projects and core systems. Through their famous "20% time" policy and the Google Labs platform, employees began launching experimental features. Google News famously was born out of an employee's personal 20% project, spurred by a strong collective hunger for rapidly updating, aggregated news topics in the wake of the September 11th attacks.

Simultaneously, Google's core organic search ranking was undergoing a massive evolution. Moving past basic PageRank and anchor text, Google began feeding user traffic data back into the algorithm, creating a powerful data network effect. If users consistently bounced away from a specific result, the algorithm learned that the page didn't belong there. Furthermore, Google never fell into the trap of thinking they had permanently "solved search." Recognizing that the exponential growth of web content meant they would never have a truly optimal index, they continuously poured money into approximating better results through system engineering, scaling from commodity hardware hacks to custom-designed data centers and pioneering software filesystems like GFS, Bigtable, and MapReduce.

* **Ranking Signals:** By 2007, Google had expanded its algorithm to use over 200 different pieces of information to rank a single query. * **Personalization:** Early personalization was limited because there was not yet a strong incentive for users to be logged into a central Google account. * **Synonym Matching:** A crucial element of search quality. Google analyzed user behavior—such as a user searching for "cool cat pictures" and subsequently modifying the query to "kitten"—to dynamically build and update a real-time synonym dictionary.

Financial Explosion and the AdSense Effect

By the end of 2002, Google experienced a fundamental business flip. Prior to June 2002, the company constantly had to make difficult financial trade-offs, but after that year, trade-offs effectively vanished. They suddenly possessed enough cash for every single project they wanted to pursue. While 2002 saw a healthy $440 million in revenue and $185 million in profits, 2003 exploded into a massive $1.5 billion in revenue and nearly $350 million in operating income.

To some listeners, this financial leap might suggest that Google's margins actually got worse, which is technically true depending on the accounting. While Google got to keep all the money from AdWords, they only kept about 20% of the money generated from AdSense. AdSense added roughly half a billion dollars in revenue in 2003 at a much lower margin, yet it was an incredibly awesome addition to the business. Meanwhile, Google AdWords steadily maintained an astonishing 85% gross margin.

* **2002 Financials:** $440 million in revenue, $185 million in profits. * **2003 Financials:** $1.5 billion in revenue, ~$350 million in operating income. * **AdSense Mechanics:** Added ~$500 million in 2003; Google retained only 20% of the revenue. * **AdWords Margin:** Maintained an 85% gross margin.

---

The Paradox of the 2004 IPO

The narrative shifts to 2004 and the historic Google Initial Public Offering (IPO). Although viewed today as a legendary success, at the time, it was widely considered infamous and horribly unsuccessful. Aside from Microsoft, there had never really been a company with so little reason to go public; Google was wildly profitable, generated immense cash, and had absolutely no need for investment capital. In fact, they had a terrifyingly good reason *not* to go public: Microsoft. Google harbored a deep paranoia that if Microsoft—who controlled the front door to the internet via Internet Explorer—found out just how wildly lucrative the search business actually was, they would deploy their massive capital to crush them.

Google’s hand was forced by the pre-JOBS Act regulatory landscape. Under the existing 500 shareholder rule, any US company crossing the threshold of 500 distinct shareholders was legally mandated to report their financials publicly anyway. Venture capital backers like Sequoia Capital and Kleiner Perkins, having just endured the dot-com crash, were not content to sit on private shares forever; they needed liquidity from their closed-end funds. Because Google never needed to raise money after their Series A, Larry Page and Sergey Brin, alongside employees via the option pool, maintained absolute voting control of the company—drawing a correlation to other ultra-capital-efficient, founder-controlled companies like Meta and Microsoft. Ultimately, going public made perfect sense to appease employees as well, instantly turning half of Google’s 2,000-person workforce into millionaires.

* **The Microsoft Threat:** Paranoia over revealing search margins to a competitor controlling Internet Explorer. * **The 500 Shareholder Rule:** Legal forcing function requiring public financial disclosures upon reaching 500 shareholders. * **VC Pressures:** Sequoia and Kleiner

[03:16:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11763s) The Genesis of AdSense

The early architecture of Gmail unexpectedly birthed one of Google's most lucrative advertising products. While working on a prototype for the email service, Paul Buchheit experimented with extracting keywords from a user's mail folder. He pulled corresponding advertisements from Google's internal ad database and displayed them alongside the content to see how well they matched. When he demonstrated this internal prototype to Larry Page and Sergey Brin, they immediately saw the broader implications, asking, "Wait, does this work on websites, too?" This specific prototyping moment for Gmail served as the direct catalyst for the development of AdSense.

[03:16:53](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11813s) The Google Playbook: Core Foundations of Success

Reflecting on the monumental growth of Google’s core search business, the hosts framework why the company achieved such dominant success. They liken this phase to "the building of the castle," anticipating that future chapters will cover the expansion of the city and nation-state around it. The foundational playbook relies on several core pillars, starting with an unmatched organic search algorithm that prioritized speed, clean user experience, and a long-standing aversion to paid inclusion.

Additionally, Google achieved world-class execution of its search advertising model. By combining an auction system, a cost-per-click model, and relevance metrics like click-through rates, they created a harmonious, self-optimizing ecosystem. In this system, advertisers are naturally incentivized to bid only on highly relevant keywords to lower their costs, seamlessly maximizing Google's expected value by serving the right ads to the right users at the exact right time.

- **Infrastructure & Engineering:** Google built an enduring advantage by reasoning from first principles and inventing entirely custom infrastructure. - **Talent Acquisition:** The company maintained an uncompromising bar, hiring exclusively world-class talent. This was heavily accelerated by the dot-com crash, which allowed Google to scoop up the industry's best minds. - **Cultural Scale:** The organizational culture blended the unconstrained, big-thinking idealism of a college campus with elite PhD brainpower. Every technical implementation detail was ruthlessly evaluated against how it would scale. - **Data Network Effects:** The core algorithm was fundamentally dependent on the massive volume of data generated by user queries, creating a powerful, self-reinforcing loop. - **The Ultimate Mission:** Their guiding corporate mission—to "organize the world's information"—proved to be neither too broad nor too narrow, pairing an altruistic public stance with the most profitable business model in history.

[03:20:26](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12026s) The Marketplace Flywheel and AI Parallels

A critical and highly unique driver of Google's success was its marketplace liquidity flywheel. Once the team recognized that the core business improved exponentially with more users and more advertisers, they realized they should spend whatever it took to expand those two pools. This dynamic functions as a "super scale economy" where scaling up does not just reduce structural costs—it dynamically increases revenue per user. In an auction-based model, higher marketplace liquidity inherently drives up the closing prices of individual micro-auctions.

The strategic landscape of Google between 1996 and 2002 features profound parallels to the modern artificial intelligence landscape spanning from 2021 to the present day. While researchers initially set out to study Google's history simply to understand the technical lineage of the transformer model, the historical landscape reveals that modern AI competitive dynamics, monetization questions, and value chains heavily rhyme with the early days of search. The era represented a incredibly narrow historical window; had Larry and Sergey started a few years earlier, the web would have been too small to require a technology-based search engine, and a few years later, the required scale would have created an insurmountable barrier to entry.

[03:24:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12267s) Evaluating Google Through the Seven Powers

Analyzing early Google through Hamilton Helmer's *Seven Powers* framework reveals that the company's competitive moats were heavily concentrated in specific areas. Counterpositioning against traditional portals like Yahoo existed because those companies were structurally dependent on portal ad revenues and a media-company identity, making them unable to pivot cleanly to a search-first user experience. While branding power was strong from an employment perspective—smart people would do anything to work there—it was weaker under the strict definition of customer pricing power, as advertisers acted as purely rational economic actors.

- **Scale Economies:** Exceptionally strong; hardware and software investments were amortized across a massive user base while simultaneously driving up ad pricing via auction density. - **Network Economies:** Highly present in the two-sided liquidity between searchers and advertisers, despite slight negative network effects among competing advertisers. - **Switching Costs:** Minimal to non-existent during this specific era, as users were not logged in and search results lacked deep personalization. - **Cornered Resource & Process Power:** Not a significant factor for the business during this foundational developmental window.

[03:29:56](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12596s) "That Company"

The ultimate defining characteristic of early Google is captured perfectly by an internal email sent on June 8, 2007, by former Google product manager Justin Rosenstein upon joining Facebook. He wrote to his colleagues to share the news that "Facebook really is that company." The phrase refers to an exceedingly rare corporate phenomenon that occurs only once in a long while—an entity on the cusp of fundamentally changing the world while remaining small enough that every individual employee exerts massive leverage on the organization. Microsoft was the first to occupy this space, Google became the definitive iteration of it for its era, and Facebook followed.

[03:31:59](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12719s) Ben's TV Show Carveouts

Ben kicks off the "Carveouts" segment with a three-way tie of excellent TV shows he watched over the last two months, noting they missed doing carveouts during their recent episode with Steve Ballmer. His most "landmark" recommendation is season two of *The Rehearsal* with Nathan Fielder. Warning listeners who are highly sensitive to spoilers to stop listening, Ben explains that he will only share details revealed in the first ten minutes of the premiere. He provides background on Fielder’s decade-old satirical show *Nathan for You*, where the comedian helped small business owners improve their operations through incredibly convoluted and counterproductive methods. Ben praises Fielder’s dry humor and intense commitment to the bit, recalling when Fielder spent hundreds of thousands—potentially millions—of studio dollars to rebrand a local coffee shop as "Dumb Starbucks" just to drive traffic.

In *The Rehearsal*, Ben explains, Fielder elevates this commitment to an unbelievable degree, dedicating years of his life to a singular goal: reducing plane crashes. Fielder believes the root cause of these crashes is pilot communication, so he builds elaborate sets and hires a fleet of actors to simulate experiences to help pilots feel more comfortable communicating. Ben teases his co-host, saying this is just "the tip of the iceberg" and that the show gets completely crazy. David is impressed, calling it an "all-time great" commitment to the bit, which prompts a humorous exchange about whether they are as committed to their own bit on *Acquired*. Ben admits they are "nowhere near" Fielder's level but finds it deeply inspiring.

Moving quickly through his other recommendations, Ben highlights *Your Friends and Neighbors* on Apple TV starring Jon Hamm. He describes it as a beautifully shot, casual, yet highly enjoyable series that follows rich people around with an unexpected twist, drawing a loose comparison to *Succession*. Lastly, he praises season two of *Andor* on Disney. While he admits the first three or four episodes start a bit slow and don't quite match the quality of the first season, he declares the final eight episodes to be "some of the best Star Wars canon that exists." David warmly commends Ben, calling him not just his best friend but his "smart friend" who gives great TV recommendations to the internet. Ben jokes that he will keep dumping these recommendations on David despite knowing David has two small kids and simply doesn't watch TV, clarifying that this list is strictly for the listeners.

- *Nathan for You* background: Satirical business consulting show featuring the famous "Dumb Starbucks" rebrand. - *The Rehearsal* Season 2 premise: Nathan Fielder tackles aviation safety by building massive simulation sets for pilots. - *Your Friends and Neighbors*: Apple TV series starring Jon Hamm featuring affluent characters and unique plot twists. - *Andor* Season 2: Disney series noted for a phenomenal eight-episode concluding run.

[03:35:16](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12916s) David's Gaming Dilemma and the Next Big Acquired Event

David takes over the carveouts, starting with a standard plug for season three of the *Gamecraft* podcast hosted by Mitch and Blake. Remembering a crossover episode they did with them years ago, David praises their ongoing dedication to the industry, asserting that if listeners love gaming and the gaming business, Mitch and Blake are "two of the best in the business." This leads into David’s second carveout, which he describes as a "real-time dilemma" that might turn into a multi-part saga on the podcast. He recalls previously giving a "preemptive carveout" expressing his excitement for the Nintendo Switch 2, which has finally launched. Although he doesn't own one yet, he has booked a shopping appointment reservation at the brand-new Nintendo store in San Francisco and eagerly anticipates playing it with his daughters when they get older.

However, David reveals a major twist in his plans: while watching reviews of the Switch 2 on YouTube, he fell down a rabbit hole of Steam Deck content and is now "desperately conflicted" about which handheld console to buy. Ben laughs, telling the "Acquiring listeners" to tune into part two of their Google episode just to find out the resolution, promising they will be on "pins and needles." David explains that while he would love to ask the audience for advice, the reality of their editing timeline means he will have already made the purchase by the time the audio drops. Instead, he promises to tweet a picture of his final choice. Transitioning to housekeeping, David plugs their upcoming live event in New York City at Radio City Music Hall, directing folks to `Acquired.fm/nyc` to experience the "ridiculous acquired experience" they are planning alongside JP Morgan Payments.

- *Gamecraft*: A gaming industry podcast recommended for its excellent third season. - Nintendo Switch 2 vs. Valve Steam Deck: David's current personal tech dilemma. - San Francisco Nintendo Store: Location where David secured a shopping reservation. - Live Event Link: `Acquired.fm/nyc` for the upcoming show at Radio City Music Hall.

[03:37:17](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=13037s) Sponsor Thanks and the Tech Dream Team Acknowledgments

As the episode winds down, Ben and David pause to express immense gratitude to the seasonal sponsors who make the show possible. They thank JP Morgan Payments for providing trusted, reliable payments infrastructure at any scale, followed by Anthropic, the creators of the Claude AI assistant. Ben notes that he used Claude a ton while prepping for this massive episode, and David chimes in to agree, stating that the tool is actively "transforming the way Acquired works." They round out the corporate thank-yous by acknowledging Statisig for product experimentation and Vercel for web development. Ben also highlights Arvin Navaratnam of Worldly Partners for writing an awesome, comprehensive breakdown of Google, which listeners can access via a link in the show notes.

The co-hosts then transition to thanking the massive lineup of tech insiders who gave up their time to help them research the history of Google. David reads off a star-studded list of former executives, engineers, and tech pioneers who granted them interviews. Ben reflects on the sheer caliber of names, joking that they had assembled a literal "dream team" capable of starting an elite tech company. David laughs, admitting that by the end of the research process, he started joking that they were "sapping billions of dollars of market cap out of the economy" just by eating up so many high-powered people's time. Deeply appreciative of the community's generosity, the duo say goodbye to their listeners, and the episode fades out to its signature theme song, "Who Got the Truth?".

- **Sponsors Mentioned**: JP Morgan Payments, Anthropic (Claude), Statsig, and Vercel. - **Research Contributor**: Arvin Navaratnam (Worldly Partners). - **Individuals Thanked**: Craig Silverstein (Google's first employee), Anna Patterson, Omid Kordestani, Alan Eustace, Clay Bavor, Bret Taylor, Jeff Dean, Jen Fitzpatrick, Danny Sullivan, Nick Fox, Paul Buchheit, Bill Gross, Wesley Chan, and Esar Lipkovitz.

***

http://googleusercontent.com/interactive_content_block/0

## Notable quotes > [05:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=303s) Probably when I was 12, I knew I was going to start a company eventually. I wanted to make the world better, and in order to do that, you need to do more than just invent things... you need to use business and entrepreneurship to make these things real. — Larry Page

> [40:14](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2414s) Great, I'm in. $100,000, — Andy Bechtolsheim

> [01:21:40](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4900s) Doug, we've never paid so much for so little. — Michael Moritz

> [01:45:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6318s) to organize the world's information and make it universally accessible and useful.

> [01:45:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6318s) We believe that the most effective and ultimately the most profitable way to accomplish our mission is to put the needs of our users first.

> [01:51:13](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6673s) Half the money I spend on advertising is wasted. The problem is I just don't know which half. — John Wanamaker

> [02:03:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7430s) Click-through rates are a proxy for relevance.

> [02:21:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8490s) We could have gone bankrupt. — Sergey Brin

> [02:21:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8490s) You don't have a business, do you? — Terry Semel

> [02:21:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8490s) We should be able to monetize the pages. If not, we deserve to go out of business. — Larry Page

> [02:28:07](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8887s) They'll learn over time that Google takes your users. It doesn't help you build your property. — Inktomi

> [03:29:56](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12596s) Facebook really is that company. — Justin Rosenstein

> [03:31:59](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12719s) some of the best Star Wars canon that exists. — Ben

> [03:37:17](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=13037s) sapping billions of dollars of market cap out of the economy — David

## Mentioned Ben Gilbert, David Rosenthal, Radio City Music Hall, Google, Alphabet, Apple, Microsoft, ExxonMobil, JP Morgan Chase, Berkshire Hathaway, Bing, Jesse Cole, Savannah Bananas, JP Morgan Payments, New York Stock Exchange, Lawrence Edward Page, Carl Page, Gloria Page, Michigan State University, University of Michigan, Mentor Graphics, Bill Gates, Mark Zuckerberg, Stanford University, Sergey Brin, Anna Patterson, British Bankers Club, Charles Schwab, Nvidia, Jensen Huang, Wolfram Research, Stephen Wolfram, Mathematica, Terry Winograd, John Battelle, Amazon, Jeff Bezos, D.E. Shaw, Yahoo, Infoseek, Lycos, Disney, GIMP, David Cheriton, Granite Systems, Sun Microsystems, Andy Bechtolsheim, Cisco, Ram Shriram, Netscape, Junglee, Susan Wojcicki, YouTube, AltaVista, Digital Equipment Corporation, Bell Labs, Jerry Yang, David Filo, Urs Hölzle, Jeff Dean, Dave Cutler, MapReduce, Bigtable, TensorFlow, Gemini, Exodus, Inktomy, Google File System (GFS), Apache Hadoop, Salar Kamangar, Red Hat, Linux, Sequoia Capital, Kleiner Perkins, Michael Moritz, John Doerr, eGroups, Omid Kordestani, America Online (AOL), Doug Leone, Vinod Khosla, Tim Armstrong, Marissa Mayer, Statsig, OpenAI, Anthropic, Figma, Notion, Brex, Snowflake, BigQuery, NASDAQ, Eric Schmidt, Novell, Burning Man, AdWords, Overture, Goto.com, Bill Gross, Ideal Lab, TED, John Wanamaker, GoTo, Open Text, Dogpile, MetaCrawler, MSN, Adobe, Real Networks, WinZip, Mozilla, Firefox, Dell, Safari, Tim Draper, Sheryl Sandberg, Larry Summers, Terry Semel, Andrew Bosworth, Sundar Pichai, AdRank, AdSense, Google Toolbar, Google Earth, Google Maps, Google Desktop, Google Groups, Google News, Google Labs, howstuffworks.com, Internet Explorer, Vickrey auction, Costco, BackRub, Gmail, Paul Buchheit, Hamilton Helmer, Justin Rosenstein, Facebook, The Rehearsal, Nathan Fielder, Nathan for You, Dumb Starbucks, Your Friends and Neighbors, Jon Hamm, Succession, Andor, Steve Ballmer, Gamecraft, Mitch, Blake, Nintendo Switch 2, Steam Deck, Valve, Claude, Vercel, Arvin Navaratnam, Worldly Partners, Craig Silverstein, Alan Eustace, Clay Bavor, Bret Taylor, Jen Fitzpatrick, Danny Sullivan, Nick Fox, Wesley Chan, Esar Lipkovitz

[Source](https://www.youtube.com/watch?v=v1nkN0sM2wc)

中文简报

谷歌规模经济在搜索变现上超越早期对手谷歌迅速崛起为主导全球搜索的霸主，得益于其在高效定制的分布式商用硬件上构建的自优化广告竞价框架。本期内容探讨了谷歌的起源故事与迅速崛起，追溯了联合创始人Larry Page与Sergey Brin如何将强烈的野心与结构性创新相结合以彻底变革搜索行业。内容详细介绍了谷歌在廉价商用硬件上实现分布式计算的突破、战略性招揽Jeff Dean和Urs Hölzle等工程精英，以及引入高利润的AdRank和次高价拍卖机制。梳理过程跟踪了谷歌如何通过与Netscape、Yahoo和AOL展开重大分发合作，以及最终推出AdSense，转型为一家财务巨头，并在2004年IPO前奠定了帝国基石。谷歌作为被政府认定的垄断企业运营，拥有90%的市场份额，创造的利润超过包括Apple、Microsoft和Berkshire Hathaway在内的任何其他美国公司。 (1:18) 与外界认为他们是“偶然创业者”的说法相反，Larry Page和Sergey Brin在强烈且深谋远虑的野心驱动下，始终保持着均等股权的合作关系。 (5:03) “Google”一名源自“Googol”（$10^{100}$），该名称原本用来体现互联网索引的庞大规模，但Larry Page在注册时拼错了。 (34:11) 谷歌开创了在消费级商用硬件上进行分布式计算的先河，利用定制的校验软件系统来容忍频繁发生的硬件故障，从而大幅削减成本。 (62:06) 谷歌超高经济效益的扩展机制与定制软件架构，最终使得谷歌搜索在商业模式成熟后达到了87%的毛利率。 (67:47)

Reading summary

Acquired hosts Ben Gilbert and David Rosenthal argue that Google's rise into one of corporate history's most profitable monopolies was neither accidental nor purely academic, but the result of solving harsh infrastructure limits with commodity hardware, perfecting search monetization via AdRank auctions, and aggressively acquiring distribution to build an insurmountable liquidity moat. Distributed Software on Fragile Commodity Hardware (62:06) While early search leader AltaVista relied on costly enterprise hardware from Digital Equipment Corporation without a sustainable business model, Google faced severe capital constraints. Founders Larry Page and Sergey Brin recruited key engineers, including Urs Hölzle and Jeff Dean, who developed a distributed computing architecture tailored to low-cost hardware. Instead of depending on expensive high-reliability servers, Google split its massive index into 64-megabyte chunks across networks of cheap commodity drives, writing fault-tolerant software that assumed hardware would continuously fail and replicating data three to five times for redundancy. This architecture converted operational austerity into a durable structural advantage. While competitors paid premium prices for enterprise hardware and pristine data centers, Google ran server racks cushioned by corkboards in crowded colocation spaces. According to the hosts, this first-principles software design reduced scaling costs so drastically that Google Search ultimately sustained an 87% gross margin once monetization matured. Aligning Monetization Through AdRank and Auctions (123:50) Google initially struggled with monetization, attempting to sell enterprise intranet search and processing manual Cost Per Mille (CPM) banner ads via fax. The turning point arrived after studying Bill Gross’s GoTo.com (later Overture), which pioneered pay-per-click (CPC) search auctions. Gross demonstrated that high-intent search queries could be monetized directly by auctioning keyword positions to commercial bidders, replacing untargeted impressions with commercially relevant listings. Rather than adopting GoTo's pure highest-bidder ranking, Salar Kamangar and Larry Page developed AdRank in late 2000. AdRank multiplied an advertiser’s bid by their ad's click-through rate (CTR), treating user clicks as a direct proxy for relevance. After curbing early click-arbitrage loopholes, Google paired AdRank with a second-price auction mechanism where winning bidders paid only one cent above the second-highest bid. This mathematical alignment rewarded higher quality ads, pleased users, and optimized revenue, accelerating Google from $86 million in revenue in 2001 to $440 million in 2002. Aggressive Distribution and Marketplace Liquidity Moats (151:41) Google recognized search as a winner-take-all market powered by increasing returns to scale: higher search query volumes attracted more advertisers, creating denser bidding pools that dynamically increased revenue per search. To preempt Microsoft before the software giant realized search's true profitability, Google aggressively pursued syndication deals. It backfilled organic and paid search results for major portals like Netscape, Yahoo, and AOL, offering generous revenue shares—often 85% or higher—to lock in default placement. Under Sheryl Sandberg, Google rapidly built an operations unit to support the surge of advertisers brought in by the massive AOL partnership. Simultaneously, Google secured direct distribution on client PCs by bundling the Google Toolbar with applications from Adobe, WinZip, and Real Networks, establishing factory pre-installs with Dell, setting default search on Mozilla Firefox, and bundling the toolbar with Google Earth. These distribution tactics drove average revenue per user (ARPU) to between $10 and $30, funding continuous distribution expansion. Cash Expansion, AdSense, and Forced Public Listing (171:39) In early 2003, Jeff Dean coded AdSense in six weeks, extending Google's ad auction model to third-party publishers like HowStuffWorks. Managed by Susan Wojcicki, AdSense shared roughly 80% of revenue with website owners. Although it carried lower margins than core AdWords, AdSense brought in roughly $500 million in 2003 alone, propelling total company revenue to $1.5 billion with nearly $350 million in operating income and ending Google's cash constraints. By 2004, Google's immense profitability meant it had no business need for public capital. The founders actively feared an Initial Public Offering because public financial disclosures would alert Microsoft to the staggering margins of search. However, pre-JOBS Act regulations mandating public reporting once a company exceeded 500 shareholders, alongside venture capital investors seeking fund liquidity, forced Google's 2004 IPO. Page and Brin retained dual-class voting control, insulating their long-term, engineering-driven strategy from short-term market pressures. The hosts conclude that Google's core era was defined by a unique historical window between 1996 and 2002, where the web was large enough to require algorithmic search but small enough for a capital-constrained startup to index completely. They argue that this era's strategic dynamics—infrastructure efficiency, auction design, and aggressive distribution flywheels—closely parallel the competitive battles unfolding across the modern artificial intelligence landscape.

精炼阅读摘要

Acquired 主播 Ben Gilbert 与 David Rosenthal 指出，谷歌的崛起绝非学术天才偶然撞大运的偶发产物，而是一场由颠覆性技术架构与极端商业进取心共同促成的精密演进。在互联网爆炸式扩张的狭小历史窗口期，谷歌凭借开创性的低成本分布式计算、对外部拍卖商业模式的改良融合，以及极具侵略性的分发策略，缔造了人类商业史上盈利能力与飞轮效应最为惊人的搜索帝国。廉价硬件与软件容错的技术壁垒 (62:06) 谷歌在创立伊始便面临网页数量指数级爆发的技术瓶颈。当时主流搜索引擎 AltaVista 依赖昂贵的 DEC 大型企业级服务器展示硬件算力，而雅虎（Yahoo）则受制于人工目录分类的扩展极限。拉里·佩奇（Larry Page）与谢尔盖·布林（Sergey Brin）在斯坦福大学研发出基于反向链接权重（PageRank）与锚文本索引的核心算法，实现了检索相关性的质的飞跃。然而，全面解析网页文本所产生的庞大索引文件，在当时没有任何单一商用主机能够容纳。为突破物理极限，谷歌吸纳了乌尔斯·霍泽尔（Urs Hölzle）与杰夫·迪恩（Jeff Dean）等顶尖工程人才，开创了革命性的分布式计算架构。工程团队将海量索引切分成 64MB 的离散数据块分散存储，由主服务器并行调度。面对当时服务器行业年均 3% 至 4% 的硬件故障率，霍泽尔反其道而行之，彻底抛弃高成本的专用企业设备，采用极度廉价的普通通用硬件搭建机架，完全通过软件将数据复制三到五份实现冗余容错。这一架构不仅化解了初创期的资金短缺，更为日后搜索业务在激增流量下维持高达 87% 的惊人毛利率奠定了成本基础。门户伙伴关系与泡沫破裂中的生存桥梁 (99:22) 在探索出可行商业模式之前，谷歌曾经历漫长的生存焦虑。1999 年第九号员工萨拉·卡曼加（Salar Kamangar）草拟的首份 A 轮商业计划书方向模糊，其尝试向大型企业销售站内搜索软件的方案在市场上遭遇彻底冷遇，一度引发红杉资本与凯鹏华盈（Kleiner Perkins）对高估值投资的强烈质疑。当时主流市场严重依赖单价低廉且缺乏精准定向的横幅广告，行业普遍对算法搜索的独立变现价值持怀疑态度。面对困境，首席商务官奥米德·科尔德斯塔尼（Omid Kordestani）果断转向向门户网站输出白标搜索结果的 OEM 模式。继 1999 年为网景（Netscape）提供服务并因流量激增被迫单日关停自身网站后，谷歌在 2000 年 6 月迎来了决定命运的转折点——与行业霸主雅虎签约，为其提供标有“Powered by Google”的自然搜索技术。在 2000 年纳斯达克暴跌引发的“互联网寒冬”中，这笔包含 1000 万美元直接股权投资以及 2001 年雅虎支付的 720 万美元服务费，构成了谷歌最关键的造血来源，在风险投资全线冰冻的环境下为公司争取到了跑通商业闭环的时间。拍卖机制演化与 AdRank 商业飞轮 (123:50) 面对早期依靠传统传真订单销售每千次展示成本（CPM）纯文本广告的低效局面，谷歌迎来了商业维度的根本蜕变。播客主播将其称为“谷歌的 Instagram Stories 时刻”：佩奇与布林敏锐借鉴了比尔·格罗斯（Bill Gross）在 Idealab 与 GoTo.com（后更名为 Overture）首创的按点击付费（CPC）与实时竞价体系。但与 Overture 纯粹“价高者得”导致劣质垃圾内容泛滥不同，谷歌将“点击率即相关性”确立为底线，由卡曼加等人研发出 AdRank 体系。 AdRank 将广告主的最高竞价与广告点击率（CTR）相乘作为排序依据，让高相关性、高转化率的广告得以享受更低的实际展示成本，实现了用户体验、广告主回报与平台期望收益的三方协同。2001 年，谷歌进一步引入博弈论中的次价拍卖机制，规定胜出者仅需支付次高出价加一美分，彻底封堵了早期按展示计费下的虚假刷量漏洞。这一升级引发了爆发式增长：谷歌年收入从 2001 年的 8600 万美元激增至 2002 年的 4.4 亿美元（净利润 1.85 亿美元）；2003 年，伴随杰夫·迪恩六周内写出的第三方分润网络 AdSense 的上线，营收更是跃升至 15 亿美元，完成了由学术项目向商业印钞机的终极蜕变。激进分发扩张与反制微软的垄断防御 (151:41) 跑通高转化率的 AdWords 广告引擎后，谷歌彻底扭转了原本向门户收取技术费的 B2B 逻辑，转而通过激进的流量获取成本（TAC）横扫全网。在 2002 年夺取美国在线（AOL）3400 万用户搜索入口的关键战役中，谷歌不惜让利 85% 并出具 744 万股认股权证，由雪莉·桑德伯格（Sheryl Sandberg）建立起规模庞大的广告运营队伍进行支撑。由于谷歌单位搜索的变现效率远超同行，即便向合作伙伴让利 80% 到 100%，其给付的绝对金额依然能压垮任何对手，形成了“用户与广告主越多，微观拍卖竞价池越深，单位用户变现价值（ARPU）反而越高”的超规模经济。这种不计短期账面成本的激进扩张，深层动因是对微软及其掌握的 Internet Explorer 浏览器入口的巨大防范心理。作为当时无需向华尔街公开季度财报的非上市公司，谷歌通过巨资将 Google Toolbar 深度捆绑至 Adobe Acrobat、Real Networks 和戴尔电脑的出厂系统，甚至将其打包进风靡一时的 Google Earth 独立客户端中，依靠内置阻截弹窗广告等实用功能牢固锁定用户，随后又与 Firefox 达成默认搜索协议。这套不惜代价购买分发渠道的防御打法，不仅成功挡住了微软的生态钳制，更演变为现代互联网流量垄断战术的原型。主播总结指出，谷歌在 1996 至 2002 年间构筑的搜索帝国，恰好处在一个难以复制的历史窗口期：若起步过早，网络体量过小无需复杂技术算法；若延后数年，海量网页带来的基础设施门槛将使后来者望尘莫及。尽管因触及 500 名股东监管上限及早期风投退出的被动压力，谷歌于 2004 年开启了一场争议不断、并不情愿的 IPO，但其以第一性原理打造技术底座、博采众长并重构商业利益链的演化逻辑，不仅奠定了现代科技巨头的范式，更为主播审视当今生成式人工智能在分发垄断、基础设施开销与变现路径上的全新角逐提供了深具穿透力的历史镜鉴。

Detailed notes — zh

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=0s) 录音室的最后时刻与 Radio City 现场秀前瞻节目在一片轻松幽默的闲聊中拉开序幕。Ben Gilbert 和 David Rosenthal 提到，这是他们在前往纽约无线电城音乐厅（Radio City Music Hall）举办大型现场秀之前，在常规工作室录制的最后一期节目。从只有一两个人的安静录音棚，突然跨越到面对数千名观众的宏大舞台，两人打趣道：在录音室里如果说错了话，随时可以喊一句“掐掉这段，重新来一遍”，就像什么都没发生过一样；但在无线电城的舞台上可没法这么干，哪怕假装喊一句“重录一次”，台下的观众也只会当成是原汁原味、毫无修饰的现场花絮。随着标志性的片头曲响起，两位主播正式宣布开启《Acquired》2025 年夏季赛季。他们将这一季的焦点投向了商业历史上最重要、也最不可思议的企业之一。 ### [00:46](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=46s) 商业史上最伟大的印钞机：为什么是 Google？在 Ben 和 David 看来，人工智能（AI）无疑是当今时代的终极叙事，也是继个人电脑（PC）、互联网和移动互联网之后，又一个板上钉钉的数万亿美元级别技术浪潮。而要真正理解 AI，就必须追溯并理解为现代 AI 技术奠定底层基础、同时统治了上一轮科技浪潮的巨头——Google。如今 Google 已经稳坐整个互联网的“大门”长达 25 年之久。但回到 1998 年它刚创立时，市场上已经存在十几个成熟的搜索引擎，当时绝大多数商业模式都极其平庸无趣。本期节目要解答的核心问题正是：为什么偏偏是 Google 脱颖而出？它是如何从一项巧妙的技术和一个优秀的产品，演化成人类商业史上最赚钱的终极商业机器的？David 特别强调，这绝非夸大其词：如今 Google（及其母公司 Alphabet）创造的净利润超过了任何一家美国上市公司，不仅超越了苹果和微软，也碾压了埃克森美孚、摩根大通和伯克希尔·哈撒韦。这种超强赚钱能力源于三个天文数字的乘积： - 极其高昂的毛利率； - 规模庞大且持续扩张的全球市场； - 牢不可破的垄断地位（根据美国政府的反垄断认定，Google 在搜索市场拥有高达 90% 的份额）。尽管由于当下的 AI 竞争白热化，资本市场对 Google 的未来产生了质疑，导致其估值倍数相较于微软、苹果和英伟达存在上万亿美元的折价，但在过去的四分之一世纪里，它的地位几乎坚不可摧。世人都津津乐道于它极简的主页搜索框、极致的速度和用户体验，但优秀的产品仅仅是冰山一角。Google 虽然远不是人类历史上的第一个搜索引擎，却成了“最后一个搜索引擎”（微软的 Bing 或许有几个百分点的市场份额想要表达不同意见，但无伤大雅）。 ### [03:30](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=210s) 播客动态与 Radio City 现场活动预告在正式展开历史长卷前，两位主持人向听众介绍了节目的最新动态与赞助商信息。他们提醒听众订阅官方邮件列表（acquired.fm/email），这是获取下期节目线索、内容更正以及幕后花絮的唯一渠道；同时欢迎大家加入官方 Slack 社区交流探讨。此外，他们的衍生访谈节目《ACQ2》近期采访了萨凡纳香蕉棒球队（Savannah Bananas）的创始人 Jesse Cole，被两位主播评为录制过程中最欢乐、最精彩的一期访谈。本期节目的赞助伙伴是摩根大通支付（JPMorgan Payments）。摩根大通也是他们即将于 7 月 15 日在纽约无线电城音乐厅举办的盛大现场活动的首席合作伙伴。这场容纳 6000 人的现场演出门票几乎售罄（仅剩不到 100 张）。活动期间摩根大通将展示其前沿的支付科技，现场还会推出限定周边，并在次日于纽约证券交易所（NYSE）举行返场特别活动。 ### [05:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=303s) 拉里·佩奇的童年：浸润在计算机学术与硅谷启蒙中故事正式从 1973 年 3 月密歇根州兰辛市（Lansing）拉里·佩奇（Larry Page）的降生讲起。拉里是卡尔·佩奇（Carl Victor Page）与格洛丽亚·佩奇（Gloria Page）夫妇的第二个儿子。拉里的童年环境在当时乃至整个计算机史上都极其罕见：他的父亲卡尔是密歇根州立大学（MSU）的计算机科学教授，拥有密歇根大学的计算机博士学位；母亲格洛丽亚同样毕业于密歇根大学计算机专业，并在密歇根州立大学担任编程讲师。这意味着在 20 世纪 70 年代初——个人电脑与微软尚在孕育破晓的阶段——拉里·佩奇就生活在一个双亲皆为计算机学术专家的家庭中，计算机科学对他而言如同空气一般自然。更为关键的是，在 1979 至 1980 学年，拉里六七岁时，他的父亲前往斯坦福大学度过了一个学术休假年（Sabbatical）。全家人搬到了帕洛阿尔托（Palo Alto），早期的硅谷风貌给年幼的拉里留下了不可磨灭的深刻印记。家庭的影响不仅限于父母。拉里那位年长 9 岁的哥哥小卡尔·佩奇（Carl Page Jr.）同样在密歇根大学主修计算机，毕业后前往美国西海岸，先后在微软早期阶段以及俄勒冈州的 Mentor Graphics 工作，随后也投身硅谷的创业浪潮，并在后续 Google 的早期融资中发挥了不可忽视的作用。 ### [07:51](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=471s) 破除神话：非偶然的商业雄心与奔赴斯坦福 David 和 Ben 在这里严肃地澄清了一个广为流传的误区：大众普遍认为拉里·佩奇和谢尔盖·布林是两个不食人间烟火、误打误撞闯入商业世界的书呆子学者，仿佛 Google 只是一个偶然演变成伟大商业帝国的斯坦福科研项目。事实恰恰相反，在两位主播的深入调研和采访中，所有受访者都一致指出了他们两人极其强烈的野心——不仅是对技术的野心，更是对商业巨舰的野心。尽管拉里的性格与马克·扎克伯格（Mark Zuckerberg）大相径庭，但他骨子里想要建立一家改变世界的庞大企业的渴望，与扎克伯格乃至上一代的比尔·盖茨（Bill Gates）处于同一水准。拉里曾在日后回忆道： > “大概在我 12 岁的时候，我就知道自己终究会去创办一家公司。我想让世界变得更美好，而为了做到这一点，你需要的不仅是发明东西。你必须借助商业和企业家精神让这些构想落地成真，仅仅发明它们是远远不够的。” 这番话背后蕴含着拉里很早就参透的两层商业本质： - 企业是将先进思想与创新技术推向普罗大众的最佳载体； - 在资本主义社会体系中，商业组织是积累利润的核心工具，只有源源不断积累利润，才能将其重新投资，去支撑更为宏大、更具野心的工程构想。 1995 年，拉里从密歇根大学本科毕业，前往斯坦福大学攻读计算机科学博士学位。正是这次跨越，让他遇到了命中注定的合伙人、挚友与商业灵魂伴侣——谢尔盖·布林（Sergey Brin）。彼时谢尔盖已经在斯坦福的博士项目中就读，并负责带领新生参观校园，两人的命运在此刻正式交汇。 ### [10:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=603s) 初次相遇：言语交锋与查尔斯·施瓦布买单的“初次约会” 拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）的初识充满戏剧性。当时谢尔盖负责带领新生参观斯坦福大学，试图吸引佩奇加入该博士项目。尽管佩奇是个初来乍到的“新人”，但他几乎在每一个细节上都在向谢尔盖发起挑战，不断反驳提出：“难道这样做系统不会更好吗？”两人一路就城市规划、交通网络和市政设计激烈辩论，言语间暗自较劲、互不服输，试图证明自己才是更聪明的那个。关于他们相识的场景，市面上有诸多版本。主播通过与谷歌早期资深员工安娜·帕特森（Anna Patterson，后曾任谷歌工程副总裁）求证，揭开了一段从未公开披露的秘辛：安娜当时是斯坦福的博士后，正是她组织了那场新生周末活动。两人的破冰并非发生在白天的校园参观中，而是在门洛帕克（Menlo Park）的英国银行家俱乐部酒吧（British Bankers Club）喝饮品时。那一晚，拉里和谢尔盖一边争执一边聊得火热，硬是喝到了酒吧打烊。更具传奇色彩的是，当晚为他们这桌斯坦福学生买单的人，竟是住在此地、经常光顾该酒吧的斯坦福著名校友——查尔斯·施瓦布（Charles Schwab，嘉信理财创始人）。施瓦布常常看到斯坦福学生就豪爽地表示“这顿我请了”，无意中资助了拉里与谢尔盖的“初次约会”。讽刺而又美妙的是，几十年后，嘉信理财平台上的账户托管了价值数百亿美元的谷歌股票。 - 两人一见如故，迅速建立起一种既言辞交锋又高度互补的深厚友谊与合作关系。 - 纵观商业史，像佩奇与布林这样数十年如一日平起平坐、股权接近对等且真正做到“1 + 1 = 100”的创始人搭档极其罕见（即便比尔·盖茨与保罗·艾伦、英伟达早期三人组，后期也迅速分出绝对主导者；巴菲特与芒格中，芒格也从未在伯克希尔全职工作且持股远逊；或许只有 Capital Cities 的汤姆·墨菲与丹·伯克可与之类比）。 --- ### [13:20](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=800s) 谢尔盖·布林的成长历程与极客基因谢尔盖同样出生于 1973 年（8月，比佩奇稍晚几个月），出生地是比密歇根更为严寒的莫斯科。在当时的苏联，身为犹太人处境极其艰难，一家人挤在莫斯科一套由国家分配的三居室公寓里，与祖母同住。谢尔盖的父亲是一位极有天赋的数学家。1977至1978年间，父亲参加了一场国际数学会议后，敏锐地意识到必须带家人逃离苏联、前往西方。历经整整两年的艰难移民申请，全家最终成功抵达美国。定居美国后，谢尔盖的父亲成为了马里兰大学的数学教授，母亲则进入 NASA 戈达德太空飞行中心（Goddard Space Flight Center）担任研究员。与佩奇父母皆为计算机科学教授的家庭背景相映成趣，在 20 世纪 80 年代，同时拥有如此顶尖理科与科研家底的孩子凤毛麟角。谢尔盖展现出惊人的早熟天资： - 16 岁高中毕业； - 进入马里兰大学，仅用 3 年时间（19 岁）便拿下数学和计算机科学双学士学位； - 随后顺利考入斯坦福大学攻读计算机博士； - 在进入斯坦福前的那个夏天，他还在开发了 Mathematica 的 Wolfram Research 公司实习（师从斯蒂芬·沃尔夫勒姆，Stephen Wolfram）。性格上，谢尔盖不仅在智力上完全与佩奇势均力敌，而且更具天马行空的搞怪色彩（zany）。他热爱轮滑，这也是日后谷歌文化中滑旱冰、玩各种新奇玩意儿的源头。在日后的分工与公司基因上，佩奇更专注于打磨能够形成商业闭环的精密产品，而谢尔盖则主导了 Google Glass、高空跳伞直播等充满未来极客气质的前沿探索项目。 --- ### [15:52](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=952s) 互联网爆炸式增长与最初的“网页批注”构想 1995 年秋天，拉里·佩奇正式来到斯坦福，与谢尔盖的关系愈发融洽，他的博士导师是人机交互领域的先驱特里·温诺格拉德（Terry Winograd）。当时，万维网（World Wide Web）正处于史无前例的指数级大爆发时期。据约翰·巴特尔（John Battelle）在《搜索》（*The Search*）一书中的统计： - 1993 年至 1996 年间，全球网站数量从 130 个激增至超过 60 万个； - 4 年间的复合年均增长率高达 **723%**； - 这种前所未见的指数级扩张，正是促使杰夫·贝佐斯（Jeff Bezos）毅然辞去 D. E. Shaw 高薪职位创办亚马逊的催化剂。受两年前同样出自斯坦福的雅虎（Yahoo!）启发，佩奇在导师指导下最初构思的博士课题并非做搜索引擎，而是一个“网页去中心化批注系统”。雅虎依赖人工编辑、中心化地将网页分类并附上评价；佩奇则设想让全网用户可以直接在任意网站上添加批注和笔记，读者在浏览网页时点击即可查看其他人留下的真知灼见。 ```mermaid flowchart TD accTitle: 从网页批注到 PageRank 的演进过程 accDescr: 展示拉里·佩奇最初构想的去中心化网页批注系统如何因垃圾评论筛选问题演变为针对网页本身的 PageRank 排序算法。 subgraph Phase1["最初构想：去中心化批注"] A["浏览网页内容"] --> B["用户自由发表批注与笔记"] end subgraph Phase2["遭遇瓶颈：信息过载与噪音"] B --> C{"热门网站面临成千上万评论"} C --> D["如何区分良莠？ 谁有资格对权威大站做批注？"] end subgraph Phase3["顿悟跃迁：PageRank 诞生"] D --> E["需要一套算法： 对评论及来源网站进行权威度排序"] E --> F["导师建议聚焦排序： 跳过混乱的批注，直接给网页排序"] F --> G["核心洞察：将该算法直接用于网页搜索排名"] end ``` 当佩奇将这个想法提交给温诺格拉德时，导师认为网络课题很有前景，但建议他进一步打磨。在与谢尔盖讨论后，两人敏锐地发现了该系统的致命缺陷：一旦面对《纽约时报》或雅虎等超大流量网站，评论区瞬间会被成千上万的用户占领、充斥垃圾信息。系统迫切需要一种机制来“淘尽黄沙始见金”，把高质量、具有权威性的评论推到顶端。佩奇后来回忆道：“我们并不是打算做搜索引擎，我们是为了解决批注问题才做了一套排序系统。我们想给全网做批注……但你怎么决定谁有权在雅虎这样的大站上批注？我们必须找出一种方法，挑选出人们该看哪条批注，这就意味着我们得先搞清楚哪些网站的评论才算具备权威性。” 当佩奇带着排序算法的构思再次找到导师时，温诺格拉德指出批注本身过于繁琐混乱，不如单纯聚焦于计算机科学中最具价值的排序难题。拉里·佩奇随即迎来了那个改变整个互联网进程的关键跃迁：不再去给批注排序，而是直接利用这套权威度机制对全网的网页本身进行排序，并将其用于解决搜索问题。由此，以拉里·佩奇（Larry Page）姓氏命名的 PageRank 算法雏形正式诞生。 ### [20:07](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1207s) 学术引文机制与 PageRank 的灵感来源佩奇和布林最初的核心构想，是通过评估网站的权威度和可信度来进行排名。对于身处斯坦福大学的学术研究者而言，这种机制其实非常亲切——它几乎就是学术界既有评价体系的翻版。一篇学术论文的重要性取决于什么？不仅取决于被引用的绝对数量，更取决于被哪些论文引用。如果顶级期刊上的重量级论文引用了你的研究，你的论文权重自然水涨船高。在万维网诞生之前，“引文分析”（academic citations）本身就已经是一个成熟的研究领域。播客中调侃道，这就像好莱坞极度热衷于拍摄好莱坞题材的电影一样，学术界也沉迷于撰写“关于论文的论文”。 ```mermaid flowchart TD accTitle: 学术引文与网页链接在权威度传导上的同构关系 accDescr: 对比学术论文引文与万维网超链接，展示权威度如何通过反向链接和锚文本进行传递。 subgraph Academia["学术界引用模型"] P1["高权威学术论文 / 顶级期刊"] -->|"引用 (Citation)"| P2["目标论文"] P3["普通论文"] -->|"引用 (Citation)"| P2 end subgraph Web["网页链接模型 (PageRank)"] W1["权威网站 (如斯坦福大学主页)"] -->|"超链接 (含锚文本)"| W2["目标网页"] W3["普通博客 / 个人主页"] -->|"超链接 (含锚文本)"| W2 end ``` 从学术引文跨越到网页排名，他们还需要补齐一个关键的映射逻辑：网络世界里的“引文”到底是什么？答案就是“超链接”（hyperlink）。超链接在结构上完全等同于学术引文，甚至在信息维度上比传统引文更为优越，因为超链接携带了一项极具价值的元数据——“锚文本”（anchor text）。今天人们只需按下快捷键 `Command + K` 就能轻松为一段文字添加超链接，这段嵌入 HTML 的文字便浓缩了链接发起者对目标页面的评价。在实际场景中，他人引用你时所用的描述文字，往往比你自己写在网页上的自我介绍更加精准、客观。拉里·佩奇基于这一逻辑构建出早期的网页排名排序，结果令人惊艳：最具权威性的网站高居榜首，垃圾网页则被自动过滤到底部。导师特里·威诺格拉德（Terry Winograd）随即鼓励道：“就把这个作为你的博士毕业论文课题吧。” ### [23:29](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1409s) “反向电话簿”难题与唯一的历史时间窗口要让 PageRank 的理论转化为现实，佩奇面临着一个底层网络架构上的技术死结：当时的万维网协议是单向设计的，任何一个网页都只记录自身指向外部的出链（outgoing links），根本无法反向查询“谁链接了我”。主持人用了一个绝妙的比喻：在手机里查看自己的通讯录轻而易举，但想要知道“谁的通讯录里存了我的号码”，仅凭自己的手机根本无法实现。唯一的破解手段，就是把全天下所有人的通讯录都拷贝一份，逐一比对并进行全局反向溯源。这就引出了一个在当时看似疯狂的决定：拉里必须写一个程序，把整个互联网的所有网页全部抓取下来，在本地建立副本，然后再反向计算所有页面的入链关系。这也造就了 Google 发展史中无法复刻的独特时间窗口： - **历史节点的偶然性**：1996 年前后，万维网的整体体量尚处于一个刚刚好能被高校单台服务器与简陋硬盘阵列容纳的狭窄区间。 - **超限的成本壁垒**：这项任务对于学生科研项目而言虽然极其冒险，但在工程上依然可行；如果再推迟一年，全网抓取并建立完整索引的硬件成本就会飙升至数千万美元；若推迟两三年，这一成本将突破数亿美元，个人或小型学术团队根本无力承担。 ### [25:20](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1520s) BackRub 诞生：Java 的挫折与 Python 救火 1996 年下半年，也就是佩奇进入斯坦福的第一学年下半学期，佩奇与布林在斯坦福内网设立了一个主页，正式启动这一极具野心的项目。因为这项技术的核心依赖于“反向链接”（backlinks），他们将项目命名为“BackRub”。随后，他们上线了二级域名站点 `backrub.stanford.edu`。佩奇最初使用 Java 编写了第一版爬虫和 PageRank 算法，但代码漏洞百出，系统基本无法稳定运转。无奈之下，他们向斯坦福的朋友斯科特·哈桑（Scott Hassan）求助。哈桑的编程水平远胜佩奇和布林，他迅速用 Python 彻底重构了代码，才终于让系统顺畅跑通。值得一提的是，哈桑从未成为 Google 的正式员工，因为当时这纯粹只是一个斯坦福校内的科研实验，连“Google”这个名字都尚未出现。当时的 BackRub 页面非常简陋，顶部挂着一张诡异的黑白人体背部按摩照片，配以鲜红的文字说明：“BackRub 是一个旨在遍历万维网的网络爬虫。目前，我们正致力于开发改进网页搜索引擎的技术。” - 早期 BackRub 并不是面向大众的搜索引擎，而是一个爬虫与 PageRank 算法的原型实验系统。 - 此时佩奇的职业规划仍是按部就班读完博士再考虑创业，类似于马克·扎克伯格在 Facebook 运营数年后才意识到这会成为一家全职公司。 ### [27:34](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1654s) 邻座的四亿美元神话：eGroups 与商业化觉醒就在佩奇和布林沉浸于博士研究课题时，身处的时代浪潮却在急剧加速。1996 至 1997 年的硅谷正处于互联网泡沫急剧膨胀的黄金期，“身处硅谷，出去创办一家公司就如同义务一样，每个人都想去狠狠砸碎这只财富彩罐（piñata）”。帮助佩奇用 Python 重构系统的功臣斯科特·哈桑便果断选择离开学校下海创业，成立了一家名为 eGroups 的公司。令人意想不到的是，与哈桑共同创立这家公司的合伙人，正是拉里·佩奇的亲哥哥卡尔·佩奇（Carl Page）。eGroups 不仅拿到了红杉资本（Sequoia Capital）和迈克尔·莫里茨（Michael Moritz）的顶级风险投资，更在短短几年后以约 4 亿美元的惊人估值被当时如日中天的雅虎（Yahoo）收购，摇身一变成了后来的 Yahoo Groups。眼看着帮自己写代码的技术伙伴和亲哥哥在隔壁成立公司、融资并套现数亿美元，留在实验室里的佩奇和布林受到了极大的心理冲击。他们开始认真反思：既然 BackRub 排名算法能输出如此出色的结果，这项技术本身显然蕴含着巨大的商业价值，或许不能仅仅把它当成一份博士论文。 ### [28:57](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1737s) 1997 年春天的技术兜售与 Excite 会面到了 1997 年春季学期即将结束之际，佩奇和布林做出了一个折中决定：他们打算把 BackRub 的底层技术直接打包卖给当时市面上的现有搜索引擎。两人的如意算盘非常务实——把技术作价一笔巨款转让出去，以顾问身份协助对方上线部署，然后拿着这笔钱心无旁骛地回到斯坦福完成博士学业。彼时的系统虽未经过大规模工业级的高并发淬炼，索引库也相对有限，且尚未开发出面向终端大众的交互式查询产品，但其权威度排序的核心算法已经通过了小规模验证。1997 年春夏之交，两人四处拜访潜在买家，几乎敲遍了硅谷所有头部互联网门户与搜索公司的技术大门。当时活跃在市场上的产品名录极其繁复： - **主流门户与先驱**：Yahoo、Lycos、Excite、Infoseek、AltaVista - **其他代表性系统**：HotBot、Inktomi、Archie、Gopher 在这轮漫长的商业兜售中，佩奇和布林距离技术变现最近的一次接触发生在新兴搜索门户 Excite 身上。据史蒂文·利维（Steven Levy）的经典著作《In the Plex》记载，两人终于成功争取到了与 Excite 决策层面对面展示技术的关键会面。 ### [30:38](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=1838s) Excite 的授权谈判与致命的分歧：留存时长 vs 搜索效率拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）在寻找商业落地的过程中，设法与 Sun Microsystems 的传奇联合创始人、当时在 Kleiner Perkins 与约翰·多尔（John Doerr）共同掌舵的顶级风投家维诺德·科斯拉（Vinod Khosla）取得了联系。当时科斯拉正是门户网站 Excite 的董事会成员。他们敲定了一项初步协议：Excite 拟出资约 100 万美元（包含部分现金与 Excite 股票），向两人授权 BackRub 搜索技术。按照原计划，拉里和谢尔盖将在那个夏天去 Excite 上班，帮他们部署 BackRub、彻底重构其搜索引擎，等到秋季学期再返回斯坦福。在交易敲定的临门一脚阶段，他们为 Excite 的 CEO 进行了一场并排对比测试（Side-by-side test）：一边是 Excite 原有的搜索算法，另一边是 BackRub。BackRub 的搜索结果极其精准，用户搜什么就能立即得到想要的结果，一点即达。反观 Excite 原本的算法体验很差，用户必须在网站里来回点击、翻页翻找，在站内停留很长时间。然而，Excite CEO 看完演示后的反应却令人目瞪口呆：“我们疯了才会换你们的算法！我赚钱靠的就是把用户留在我的网站上。我不希望他们离开网站，你们简直不可理喻，赶紧走人，交易取消。” 这段发生在 1996 至 1997 年间的插曲，淋漓尽致地揭示了早期互联网商业模式中最核心的利益冲突。当时的门户网站（Portals）清一色依赖横幅广告（Banner Ads）按千次展示付费（CPM, Cost Per Thousand Views）变现，所有战略重心都是尽可能提高网站的页面浏览量（Page Views）和留存时间。而 BackRub 极致高效的技术无异于直接腰斩站内 PV，把流量双手奉送给外部网站。对当时的 Excite 而言，采用更好的搜索技术不仅不是优势，反而直接颠覆了自身的盈利逻辑。 ```mermaid flowchart TD accTitle: Excite 与 Google 早期商业模式冲突对比 accDescr: 展示了传统门户网站的流量闭环与 Google 高效搜索跳转机制的根本冲突 subgraph Portal["90年代门户网站模式 (Excite / Yahoo)"] A["劣质搜索结果"] --> B["用户在站内反复翻找/跳转"] B --> C["增加页面浏览量 (Page Views)"] C --> D["展示更多 Banner 广告 (CPM 变现)"] end subgraph BackRub["BackRub / Google 模式"] E["输入关键词"] --> F["PageRank + 锚文本精准排序"] F --> G["首位结果即目标内容"] G --> H["用户点击并立即离开网站"] end Portal -.->|"商业逻辑冲突：好搜索破坏了广告库存"| BackRub ``` --- ### [33:43](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2023s) 碰壁各大门户与从 BackRub 到 Google 的诞生 1997 年秋天，拉里和谢尔盖不得不返回学校。所有快速变现的谈判全部告吹，拉里索性决定：“去他的吧，既然我认为搜索应该这样运作，那我们就自己在斯坦福把它做出来。”正如拉里后来的回忆：“没人愿意出大价钱买 BackRub。我们拿到了些报价，但金额都很低，所以我们决定回斯坦福继续做。那些公司根本不会专注在搜索上，他们都想做门户，只想要眼球和 PV。他们不懂搜索，他们也不是做技术的人。”事实上，他们当时曾尝试以 100 万美元的价格将 PageRank 卖给雅虎（Yahoo!），但遭到拒绝——这也是雅虎多次错失收购 Google 历史中的第一幕；此外，他们还向 Infoseek（后来被迪士尼收购并关闭）以及 Lycos 推销过，同样四处碰壁。回到校园后，两人着手打造真正属于自己的搜索引擎。首先要解决的就是产品名称问题，“BackRub”虽然描述了背后的“反向链接”技术，但听起来完全不像一个搜索工具。在确定最终名称之前，他们甚至还考虑过一个叫“WhatBox”的名字，意为用户在里面输入问题的框，但因为听起来太像成人网站而迅速被弃用。最终，拉里的宿舍室友建议使用数学术语“Googol”（代表 10 的 100 次方，即 1 后面跟 100 个零），象征着组织海量互联网信息的雄心。关于名称拼写的流变，流传着两种说法：一种是拉里在注册域名时拼错成了“Google”；另一种说法则是原域名“googol.com”已经被他人抢注。谢尔盖随后使用开源绘图软件 GIMP 设计了首个 Google 首页和彩色 Logo。尽管设计简陋且排版一度显得杂乱，但“白底背景 + 彩色 Logo + 一个单一搜索输入框”的核心极简架构，自 1997 年确立后一直沿用至今。 - **各大门户拒购记录**：雅虎（报价 100 万美元被拒）、Excite（估值 100 万美元的现金加股权交易告吹）、Infoseek、Lycos。 - **废弃候选名称**：“WhatBox”，因名称联想不良被放弃。 - **数学词源**：原词为“Googol”（$10^{100}$），最终登记并确定为“Google”。 --- ### [37:43](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2263s) 流量瘫痪斯坦福网络与剥离建司前夕进入 1997–1998 学年后，BackRub 正式演化为运行在斯坦福校园网上的 Google。搜索机制实现了两大关键飞跃：一是基于反向链接（Backlinks）权重的算法排序，二是结合超链接上的锚文本（Anchor Text）精准描述关联网页。这彻底终结了当时其他搜索引擎只通过页面内部词频暴力匹配的做法——此前如果有人搜“狗”，充斥着大量无意义“狗”字的低质页面就会霸占榜首。优质的搜索体验迅速引发病毒式传播。到了 1998 年春季学期，google.com 的日均查询量已经突破 1 万次。这一产品先是在斯坦福内部传开，接着风靡全美高校与学术社区，并迅速渗透进整个硅谷。由于流量激增，Google 常常直接搞垮斯坦福的校园网络；在最夸张的时期，光是 Google 这一项服务就占用了整个斯坦福大学近一半的网络带宽。要知道，当时的 Google 没有任何复杂的图片和重型脚本，完全是一张轻量级的纯白页面和纯文本搜索结果，如此 asset-light 的架构能吃掉半座大学的带宽，足以证明访问量已经达到了惊人的量级。庞大的基础设施也给两位创始人带来了严峻的技术挑战。由于缺乏预算，拉里和谢尔盖甚至不得不干起“顺手牵羊”的营生：当时其他实验室订购的新电脑送到学校收发货卸货区（Loading dock）后，如果对方的项目还有几个月才启动，拉里和谢尔盖就会偷偷把机器搬回机房组装到 Google 集群里，先跑几个月，等到原团队要用时再悄悄送回去。到了 1998 学年结束时，形势已经非常明朗：Google 必须正式脱离学校独立成为一家商业公司。斯坦福大学虽然一直在慷慨地提供基础设施庇护，但服务器和带宽的极限已经逼近临界点，校方明确要求他们搬离网络并自筹资金。作为技术转移和知识产权剥离的回报，斯坦福获得了这家新公司约 1% 的股权，在随后的岁月里赢得了不可估量的丰厚回报。 ### [40:51](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2451s) 斯坦福教授大卫·切瑞顿与清晨的门廊会面拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）在为如何将项目公司化而发愁时，找上了斯坦福计算机科学系的教授大卫·切瑞顿（David Cheriton）。切瑞顿不仅是学术大牛，更是位传奇的连续创业者——他与 Sun Microsystems 联合创始人安迪·贝托尔斯海姆（Andy Bechtolsheim）此前创办了以太网交换机公司 Granite Systems，而思科（Cisco）刚刚以 2.2 亿美元将其收购，切瑞顿在保留教职的同时早已实现财富自由。拉里和谢尔盖心想：“大卫是我们的导师，他最懂怎么搞产业化。”切瑞顿便建议道：“你们不如和安迪聊聊，看怎么把这个项目剥离独立出去。” 切瑞顿当天傍晚就给贝托尔斯海姆发了邮件。贝托尔斯海姆当即回复：“明早我很忙，不过早上 8 点在我去办公室的路上，我们在你家碰头如何？”由此拉开了硅谷风投史上极具传奇色彩的 Google 种子轮融资序幕。 ### [41:49](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2509s) 赞助商插播：J.P. Morgan Payments 播客主持人插播介绍了节目的主要赞助商摩根大通支付（J.P. Morgan Payments）。主持人提到双方的深厚合作，从此前在 Chase Center 举办的线下活动，到即将前往 Radio City 的合作。作为全球规模最大的支付业务特许经营机构，摩根大通支付支撑着全球前 20 大企业中的 18 家，《财富》500 强中超过 90% 的企业都是其客户。其每天处理的支付体量高达 10 万亿美元，处理超过全美 50% 的电商交易规模，已成为全球金融体系的底层支撑平台，并为初创企业与大型跨国集团提供防欺诈、外汇兑换（FX）及营运资金解决方案等全套开发者 API 与企业级基础设施。 ### [44:02](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2642s) 一张写给不存在公司的 10 万美元支票第二天清晨 8 点，拉里和谢尔盖挣扎着从斯坦福宿舍爬起来，赶往切瑞顿位于帕洛阿尔托（Palo Alto）市中心的住宅。贝托尔斯海姆驱车赶到，雷厉风行地对他们说：“我很赶时间，快给我演示下你们搞出了什么。”在看完了 Google 搜索的原型演示后，贝托尔斯海姆立刻被震撼了，当场表态：“太棒了，我投 10 万美元。”拉里和谢尔盖当场愣住了，连忙解释：“我们没打算融资啊，我们只是来咨询成立公司的建议。”贝托尔斯海姆丢下一句：“很好，我去车里拿支票。” 贝托尔斯海姆径直开出了一张抬头为“Google Inc.”的 10 万美元支票，塞给两个年轻人后跳上车扬长而去。然而在当时，名为“Google Inc.”的公司在法律上根本还不存在。贝托尔斯海姆的态度十分洒脱：“这你们自己去解决，不是我的问题，反正面额有效。”没有任何投资法律协议，没有估值条款，甚至连接受资金的法人主体都没有。这张支票成了不可逆的“倒逼机制”（forcing function），迫使拉里和谢尔盖火速设立法定实体、向斯坦福大学买下知识产权（IP）授权、开设银行账户，赶在支票过期前将其承兑。 ```mermaid flowchart TD accTitle: Google 种子轮融资脉络图 accDescr: 展示从最初的 10 万美元支票到最终敲定 100 万美元种子的 4 位投资人结构 subgraph Origin["起点：帕洛阿尔托门廊会面"] A["Larry & Sergey 演示 Google"] --> B["Andy Bechtolsheim 开出 $100K 支票 （抬头写给尚未成立的 Google Inc.）"] end subgraph Legal["倒逼公司化成立"] B --> C["注册 Google Inc. / 解决 Stanford IP 授权 / 开设银行账户"] end subgraph Investors["四位种子轮天使投资人"] C --> D1["Andy Bechtolsheim: $100K"] C --> D2["David Cheriton: $100K"] C --> D3["Ram Shriram: $250K"] D3 --> D4["Jeff Bezos: $250K （占种子轮 25%）"] end subgraph Result["种子轮收官"] D1 & D2 & D3 & D4 --> E["投后估值 $10M，融资金额 $1M"] end ``` ### [45:25](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2725s) 杰夫·贝佐斯入局：百万美元种子轮与门洛帕克车库在办理注册与知识产权移交的几个月里，这轮融资像滚雪球般壮大。大卫·切瑞顿自己决定跟投 10 万美元。随后，拉里和谢尔盖结识了前网景（Netscape）高管拉姆·什里拉姆（Ram Shriram）。当时什里拉姆刚将自己加入的创业公司 Junglee 卖给 Amazon，手握充足流动资金的他不仅为两位创始人提供顾问支持，还自掏腰包投资了 25 万美元。接着什里拉姆问他们：“你们想见见杰夫吗？” 这里的“杰夫”指的正是杰夫·贝佐斯（Jeff Bezos）。在 1998 年的时间节点上，Amazon 创立于 1994 年并已于 1997 年上市，贝佐斯在互联网界宛如一位成熟的“长者”。当贝佐斯再次来到硅谷时，三方在什里拉姆的家中碰面。贝佐斯和贝托尔斯海姆一样干脆利落，直接问什里拉姆：“拉姆，你投了多少？25 万？那我也跟 25 万。”最终，Google 的种子轮融资金额锁定在 100 万美元，投后估值 1000 万美元。 - **极度集中的天使天团**：整轮 100 万美元仅由 4 位投资者构成——安迪·贝托尔斯海姆（10 万美元）、大卫·切瑞顿（10 万美元）、拉姆·什里拉姆（25 万美元）和杰夫·贝佐斯（25 万美元）。贝佐斯一人就独占了四分之一的额度。 - **回报计算**：贝佐斯从未披露他后来是否减持了 Google 股票。如果他始终一股未卖，这笔 25 万美元的投资在今天价值约 200 亿美元；即便他在 2004 年 IPO 时全部变现，那笔 25 万美元也暴赚至近 2 亿美元。完成融资后，拿到百万现金的 Google 彻底脱离斯坦福孵化阶段，租下了门洛帕克（Menlo Park）一间民宅的车库作为最初的办公室。车库的主人正是时任英特尔（Intel）经理的苏珊·沃西基（Susan Wojcicki）——她不仅很快成为 Google 的早期核心员工，日后更掌舵成为 YouTube 的首席执行官。 ### [48:58](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=2938s) 1998 年的搜索版图：AltaVista 与 DEC 的硬核基因手握 100 万美元现金的 Google 必须面对高昂的服务器抓取成本，也必须直面 1998 年激烈的搜索市场环境。彼时雅虎（Yahoo）与 AOL 稳居门户霸主地位，而专注于纯粹搜索引擎的代表玩家则是 AltaVista。在 Google 横空出世之前，AltaVista 的使用体验其实相当优秀。 AltaVista 的出身极其特殊——它并非源自纯粹的互联网初创企业，而是来自巨型硬核硬件计算先驱 DEC（数字设备公司，Digital Equipment Corporation）设立在帕洛阿尔托的西部研究实验室（Western Research Laboratory）。DEC 是创造了小型机神话、孕育出 Windows NT 架构师戴夫·卡特勒（Dave Cutler）的硬件巨头。AltaVista 最初其实是 DEC 用来炫耀其顶级企业级服务器运算与并发性能的“展示品”。 - **突破性架构（并行抓取）**：在 AltaVista 之前，所有网络爬虫（Web Crawler）几乎都是单线程顺序执行，爬完一个页面再去请求下一个。 - **技术转折**：由于互联网在 90 年代中后期呈现指数级扩张，AltaVista 率先洞察到抓取网络是可以高度并发（parallelizable）的，利用 DEC 强悍的多处理器硬件优势实现了史上首个大规模并行爬虫架构。这也启发了行业重新审视搜索引擎的核心竞争要素——搜索引擎之间的较量绝非仅仅取决于排序算法，底层基础架构的并发索引能力同样至关重要。 ### [51:12](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3072s) 搜索的三大维度：速度、索引规模与关键词堆砌的黄昏在评估搜索引擎时，人们通常认为创新、相关性和排名算法最为核心，但实际上还有另外两个至关重要的维度：**速度（Speed）**与**索引规模（Index Size）**。在今天，毫秒级返回搜索结果已被视作理所当然，但在谷歌诞生前夕，搜索完全是另一种体验——用户提交一个查询后，必须起身去干点别的，静静等待结果返回。这就像今天人们运行复杂的 AI 深度研究（Deep Research）任务一样：“丢一个 Prompt，去吃个午饭，吃完回来再看结果。” 另一个致命属性是索引覆盖面：你所检索的网页数据库究竟有多大？在 AltaVista 引入并发/并行爬虫技术之前，市面上所有主流搜索引擎的索引规模都小得可怜，最大者也不过包含约 100 万个网页。这意味着即便你拥有全世界最聪明的排序算法，如果爬取到的全网站点占比微乎其微，搜索引擎也谈不上实用。更滑稽的是，当时各家引擎极少全量更新索引，所谓“已收录 100 万页”不过是历史累积数字，他们只是不断追加新网址，几乎不做动态刷新。当 AltaVista 脱离 DEC（Digital Equipment Corporation）作为独立商业实体上线时，其最大的卖点正是索引规模——它坐拥惊人的 1600 万个网页索引，瞬间碾压所有竞品。然而，AltaVista 在相关性和响应速度上依然表现糟糕。当时包括 AltaVista 在内的信息检索算法高度依赖关键词词频：如果某个关键词在页面上出现的次数多，算法就认为该页面更相关。这也导致了极度荒谬的作弊手法——如果你想让自己的网页在“狗粮”（dog food）这一查询词下排到第一，只需在页面所有角落用隐形文字或明文疯狂堆砌“狗粮”，硬生生把页面做成全网“最狗粮的狗粮页面”（dog foodiest dog food page）。 - **传统检索痛点**：以词频为核心的简单匹配规则催生了大量关键词农场与隐形文本作弊，搜索极易被垃圾网页操纵。 - **静态累积索引**：早期引擎通常不重新抓取和清理旧页面，数据新鲜度极低。 - **AltaVista 的跃升**：凭借并行爬虫机制将索引库推升至 1600 万页，但检索结果的质量和速度仍未解决。 --- ### [53:09](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3189s) 昂贵的硬件与鸡肋的商业模式：搜索在当时的尴尬处境 AltaVista 背靠 DEC，拥有全行业最顶级但也极其昂贵的硬件设备。如果 AltaVista 想作为一家独立企业持续运转，这种依赖 DEC 昂贵服务器的并行架构成本将高到令人咋舌。更致命的是，当时整个搜索行业根本没有跑通健康的商业模式：变现渠道几乎完全依赖廉价、粗放且缺乏精准定位的横幅展示广告（Banner Ads）。这就解释了为什么当年的硅谷主流并不看好独立搜索公司。做好搜索不仅极其烧钱，而且看不到合理的商业回报上限。甚至连 DEC 内部高层的心态也非常微妙：他们真正关心的根本不是打造全世界最优秀的搜索引擎，而是怎样把自家的 Alpha 服务器硬件卖出去。AltaVista 在 DEC 眼中更像是一个展示服务器极限算力与并发性能的“概念验证（Proof of Concept）”，DEC 期望其他创业公司能看中这套架构，进而大批量采购 DEC 昂贵的硬件机箱。 - **商业模型缺失**：搜索流量仅能通过无差别的横幅广告按曝光变现，单位流量价值极低。 - **高昂的基础设施开销**：支撑海量网页爬取与高并发检索所需的专业服务器硬件，令独立搜索公司的毛利极其难看。 - **硬件厂商的算盘**：母公司 DEC 推动 AltaVista 的初衷是售卖高利润服务器，而非押注互联网信息服务。 --- ### [54:21](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3261s) 雅虎的分类目录霸权与“差不多就行”的搜索哲学在 20 世纪 90 年代末，搜索非但不是一门好生意，甚至也不是网民的主流入口。当时统治互联网的是门户网站（Portals）与人工分类目录（Directories），其中的绝对霸主是雅虎（Yahoo!），而非 AltaVista、Excite、Lycos 或 Infoseek。雅虎于 1996 年上市时市值约 10 亿美元，而到了 1998 年谷歌刚成立时，雅虎已成长为市值超 200 亿美元的庞然巨兽。与极度依赖硬核技术的搜索引擎不同，雅虎虽然由同为斯坦福博士生的杨致远（Jerry Yang）和大卫·费罗（David Filo）创立，但它的起点不是深奥的学术研究，而是二人整理的“杰里和大卫的万维网指南”（Jerry and David's Guide to the World Wide Web）。雅虎的核心产品是一份由人工精挑细选、精心标注的互联网分类导航，类似附带详尽点评的“互联网黄页”。在当时，大众普遍深信机器算法永远无法取代人类对优质内容的鉴赏力。连拉里·佩奇最初构思的也是基于人工标注的学术网页排名。在互联网早期体量尚小的时候，人工编辑确实高效而准确；但随着网页数量迎来万倍暴增、长尾小众需求爆发，分类目录的层级体系迅速崩溃。 ```mermaid flowchart TD accTitle: 90年代末雅虎混合模式与谷歌纯粹架构对比 accDescr: 对比雅虎以人工目录为核心、搜索做垫底的媒体门户模式与谷歌以算法和爬虫基础设施为核心的技术模式 subgraph YahooModel ["雅虎的门户/媒体模式"] Y1["人工分类目录 (主要展示)"] --> Y2["横幅广告展示 (停留变现)"] Y1 -. 兜底回填.-> Y3["第三方搜索引擎 (次要补充)"] end subgraph GoogleModel ["谷歌的技术/检索模式"] G1["分布式爬虫与存储"] --> G2["PageRank 核心索引"] G2 --> G3["极简输入框 / 秒级精准直达"] end ``` 到了 1998 年，面对体量膨胀的网络世界，雅虎等门户纷纷妥协，转向了“混合模式”：用户在雅虎检索时，排在最顶部的永远是人工审核收录的分类目录，只有当目录无法覆盖时，底部才用合作搜索引擎的结果做“兜底回填”（Backfill）。这是一种典型的“差不多就行”（Good Enough）的技术妥协： - 雅虎本质上是一家媒体公司，商业目标是靠丰富的内容把用户留在站内看横幅广告，技术只要不掉链子即可。 - 谷歌则截然相反：拉里和谢尔盖痴迷于用最优雅、终极的技术方案去彻底解决问题，对半吊子的妥协毫无兴趣。 --- ### [57:28](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3448s) 谷歌的反骨困局：拒绝横幅广告与暴增的基础设施成本雅虎的模式是堆砌丰富内容并展示横幅广告，而拉里和谢尔盖对这种做法嗤之以鼻。他们坚决拒绝让主页充斥杂乱信息，谷歌的唯一目标是帮助用户以最快速度找到想去的地方并离开主页。这引发了投资人巨大的困惑：如果搜索引擎不在站内强留用户看广告，商业模式到底是什么？拉里和谢尔盖极度厌恶广告，尤其是对用户体验毫无价值的横幅硬广。从投资人的视角来看，当时的谷歌呈现出一种令人费解的撕裂状态：一方面，全球互联网以每年 700% 的速度爆炸式扩张，谷歌以极佳的体验迅速成为网民公认的互联网正门；但另一方面，全行业唯一的变现路径（横幅广告）却被两位创始人坚决否定。更严峻的危机在于不断恶化的算力开销。随着使用量几何级爆发，系统所需的物理基础设施急剧攀升，但在零收入的情况下，每一笔硬件投入都是纯现金流出。一个完整的搜索架构必须拆解为多个独立扩容的系统： - 抓取全网内容并切片解析的分布式爬虫系统； - 存储全网链接并完成相关性计算的核心索引库； - 高并发处理用户实时查询的网页检索服务器。这三大模块对资源的需求完全不同，随着流量暴增，技术架构全面逼近崩溃边缘。 --- ### [59:34](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3574s) 基础设施救赎：传奇工程师 Urs Hölzle 与 Jeff Dean 的加入这构成了谷歌之所以能成为商业奇迹的第二大支柱。如果说第一大支柱是以 PageRank 为代表的搜索相关性与排序算法，那么第二大支柱就是能以极低成本高效支撑海量规模扩张的**底层基础设施**。拿到天使轮融资后，拉里和谢尔盖的第一要务就是招揽顶尖计算机科学家重构这套濒临崩溃的代码。他们很快吸引了两位日后奠定谷歌技术帝国版图的图腾级人物：**乌尔斯·霍泽尔（Urs Hölzle）**与**杰夫·迪恩（Jeff Dean）**。乌尔斯·霍泽尔作为第 8 号员工加入谷歌。他是斯坦福计算机博士、UCSB 教授，此前曾编写了被 Sun 公司采纳为官方标准的 Java 虚拟机（HotSpot JVM）。当拉里和谢尔盖将他从象牙塔拉入初创期混乱的谷歌时，他的首个正式职位头衔极其接地气地被称为**“搜索引擎修理工”（Search Engine Mechanic）**——因为当时系统里“几乎所有东西都是坏的（everything was broken）”。从 1999 年至 2023 年长达二十余年的时间里，乌尔斯始终执掌谷歌的整个基础设施体系。几乎同一时期从 DEC 高级工程师岗位被挖来的杰夫·迪恩，则被节目主持人誉为“谷歌的戴夫·卡特勒（Dave Cutler）”。杰夫·迪恩不仅亲手编写了第一版 AdWords，搭建了 AdSense，将核心搜索流水线从头重写了多达 5 次，还深度参与主导并联合发明了 Bigtable、MapReduce、TensorFlow 以及当下的 Gemini 模型。如今他统领 Google DeepMind 与 AI 研发部门，至今仍在网络上公开维护自己的个人简历，其工程成就堪称硅谷一代传奇。 - **Urs Hölzle 的贡献**：确立了以廉价 PC 硬件搭建高可用分布式集群的哲学，带领谷歌架构平稳度过最凶险的流量暴涨期。 - **Jeff Dean 的技术履历**： - 核心变现系统：实现 AdWords 第一版，主导构建 AdSense。 - 搜索骨架：先后 5 次重写谷歌核心检索管线。 - 分布式计算基石：联合发明 Bigtable 与 MapReduce，重塑了整个现代大数据产业。 - 现代 AI 浪潮：联合主导 TensorFlow 与前沿多模态大模型 Gemini 的研发。 ### [01:01:14](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3674s) 互联网泡沫期的顶级人才磁石在 1998 到 1999 年间，Google 招募到了这批最初的核心工程师。回过头来看，这简直不可思议：在当时，Google 的各项技术构想能否跑通、能否横向扩展，前景其实非常渺茫；正是这批天才工程师的加入，才最终把一切变成了现实。很多人事后会轻描淡写地认为，Google 只是在 2000 年互联网泡沫破裂（Dot-com crash）后，凭借自身稳固的商业模式顺手抄底、收割了全硅谷最顶尖的工程师。但事实恰恰相反，在 1998 和 1999 年那个狂热的“淘金热时代”（go-go times），泡沫根本还没破裂，人人都想去最火的创业公司发财。Larry Page 和 Sergey Brin 却能在这个炙手可热的人才市场中把顶尖高手招致麾下。这一步正是决定 Google 生死存亡的关键历史拐点。能够在最疯狂的市场中打动这些技术大牛，充分说明了 Larry 和 Sergey 愿景的感染力、想法本身的令人兴奋，以及他们技术路线的极致新颖。当然，产品本身优秀、用户增长迅速确实让 Google 看起来很有趣，但真正吸引这群顶级程序员的，是源自斯坦福的那套极其独特、前所未见的技术架构和计算挑战。对顶级工程师来说，这才是最具致命吸引力的攻坚课题。 ### [01:02:29](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3749s) 逼出来的分布式计算：海量索引与 64MB 切片 Google 面临的核心难题在于，为了让 PageRank 算法生效并建立完整的网页索引，它所需要处理的数据体量远超当时市面上的任何搜索引擎。其他引擎可能只抓取网页的标题或前几行文本，而 Google 为了计算网页排名、提取海量的反向链接（Backlinks），必须把整个网页内容完整抓取并处理。这导致其索引文件庞大到无论购买多么昂贵、规格多么顶级的单台大型服务器，都绝对无法塞下。 Google 被迫走上了大规模分布式计算与分布式文件系统的道路。既然单机装不下，工程团队就把巨大的索引切分成数以万计的、易于处理的小文件——每个被称为一个“Chunk”（切片），大小固定为 64 MB。这些 64MB 的小切片被分散存储在无数台普通的机器、硬盘、服务器乃至遍布全球的不同数据中心里。系统设立了一个独立的 Master 服务器（主控服务器），专门维护一张全局元数据映射表（Master Mapping），记录每一个 Chunk 具体存放在哪台机器的哪个物理位置上。 ```mermaid flowchart TD accTitle: Google early distributed search index architecture accDescr: Diagram showing how incoming queries query a master mapping server to access 64MB chunks across commodity disks. subgraph QueryFlow["查询检索层"] Q["用户检索请求 (Query)"] --> M["Master 服务器 (Master Mapping 表)"] end subgraph ChunkStorage["分布式切片存储层 (GFS 前身)"] M -.->|"仅定位所需 64MB 切片位置"| D1["机器 A / 硬盘 1 Chunk #1042 (64MB)"] M -.->|"并发读取"| D2["机器 B / 硬盘 2 Chunk #3891 (64MB)"] M -.->|"并发读取"| D3["机器 C / 硬盘 3 Chunk #7205 (64MB)"] end subgraph ParallelReturn["结果聚合"] D1 --> R["并发汇总切片数据并返回结果"] D2 --> R D3 --> R end ``` 当一条检索请求进入系统时，计算节点无需加载整个索引文件，只需向 Master 索取该查询相关的特定 Chunk 物理地址，然后直接以高度并发的方式从各个机器上拉取对应的 64MB 数据并进行计算。在计算层看来，系统仿佛拥有访问整个文件的抽象接口，而底层传输的实际上仅仅是被切碎、精准调取的微小数据块。这一架构让海量索引的实时查询在工程上成为可能。 ### [01:04:39](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=3879s) 廉价消费级硬件、软木板机架与 87% 的毛利率如果数据本身已经被切碎且可分布在任何地方，那么底层的物理硬件其实不需要任何昂贵的名牌设备。早在斯坦福时期，Larry 和 Sergey 就因为缺钱而四处淘廉价的消费级主板和硬盘。等到 Urs Hölzle 等人加入后，他们果断把这个思路推向极致：干脆全面采用最便宜的消费级标准配件（Commodity hardware）。这种廉价硬件当然质量差、极易损坏甚至经常烧坏，但这完全没关系——在分布式架构下，软件层直接把每份数据做 3 到 5 份的多副本冗余容错设计。软件工程师们面临的挑战变成了：“当底层硬件完全不可靠、没有企业级服务器保障时，我该如何设计出一套高可用的系统？”这激发了工程师们极大的研发热情。数据表明，当时行业内企业级服务器的年硬件故障率大约在 3% 到 4% 左右，而 Google 由于大量使用消费级零件，年故障率超过了 10%。但在多副本软件容错机制下，机器随时宕机、硬盘随时报废对前端服务完全毫无影响。随着 Google 疯狂扩张，即使在私有化时期，按消耗的计算组件规模计算，他们实际上已经隐形地成为了“全球最大的计算机制造商”——因为他们从不购买组装好的整机，而是成批采购主板、内存、硬盘等散件自己堆砌。为了节省成本，Google 在早期的托管机房（Colocation facility）里做出了各种叹为观止的奇葩举动： - **摒弃机箱与软木板架构**：他们根本不买电脑机箱，而是直接把主板钉在廉价的软木板（Corkboard）上，插上内存与硬盘，直接塞进机架。软木是天然的绝缘体，能防止裸露的电路板相互导电短路。 - **钻机房计费规则的空子**：早期他们在圣克拉拉（Santa Clara）租用的是 Exodus 托管机房。当时的租赁合同是按**占用面积（平方英尺）**计费，而不是按功耗或机器台数计费。作为顶级计算机科学家，面对这一边界约束，他们将“空间密度”优化到了极致：既然不算电费，那就彻底去掉机箱隔断，把软木板机架密密麻麻塞满整个空间。 - **邻居对比与“借电”传闻**：Google 简陋如“科学怪人弗兰肯斯坦”（Frankenstein）般的杂乱机架，正巧紧挨着当时如日中天的竞争对手 Inktomi。Inktomi 的机房区域井井有条，摆满了光鲜亮丽的 Sun Microsystems 昂贵服务器，布线极其优雅、散热通风空间充裕；而 Google 这边则如同垃圾堆般塞满了狂暴吸电的软木板裸机。据说 Google 甚至因为功耗太大，一度偷偷“借用”（borrowed）了隔壁 Inktomi 机架的电源插座。 - **压倒性的单位经济模型**：这套被逼出来的野路子让 Google 能以极低成本横向扩展，伴随搜索流量暴增，单位边际成本远低于对手。一旦广告变现的商业模式跑通，这种基础设施优势直接转化成了高达 87% 的惊人搜索毛利率。 ### [01:09:25](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4165s) 内存校验位与残酷的“宇宙射线”挑战 Jeff Dean 曾分享过一个生动的技术细节，用以说明企业级硬件与消费级廉价硬件的本质区别：在企业级硬件中，内存（RAM）通常带有奇偶校验位（Parity bit）或 ECC 错误检查机制。奇偶校验位会给每个字节额外增加 1 bit 数据用于校验——通过统计该字节内 1 的数量是奇数还是偶数，把校验位设为 0 或 1。如果某位被宇宙中飞行的随机电离辐射撞击而意外翻转（Bit flip），系统就能立刻识别出数据已被损坏。其代价是机器成本上升，并且牺牲了约 1/8 的内存容量。而在 Google 使用的消费级廉价内存上，根本没有这种硬件级的奇偶校验机制。这意味着宇宙射线或其他电磁干扰随时可能悄无声息地篡改内存中的数值，而硬件底层却毫无察觉。为了保证计算的准确性，Google 必须在纯软件层面自己搭建复杂的检验与多重验证机制。这也逼迫 Google 的工程师们构建出一整套前所未有的底层系统软件栈：在绝对不可靠的“破烂硬件”之上，硬生生用纯软件代码构筑起一套坚不可摧的高可靠系统。 ### [01:10:55](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4255s) 约束逼出的底层基建创新与极致速度依托廉价普通硬件（commodity hardware）运行系统的现实，加上团队极其强悍的工程技术实力，迫使谷歌从最底层开始重构一切。从计算系统、文件系统、数据中心、机架到具体硬件，全部由自己从零设计。他们打造出了举足轻重的底层技术，例如谷歌文件系统（GFS）和 MapReduce。这项技术后来对整个行业产生了深远影响——雅虎为了保持竞争力不得不仿制 MapReduce，并将其开源为 Apache Hadoop。虽然 Hadoop 后来主要由雅虎外部的开源社区维护和发展，但其最初正是源自对谷歌技术的追赶与复刻。自主定制机架与数据中心不仅大幅压低了成本，还让谷歌得以在全球范围内迅速铺设分布式数据中心。这种全球化布局带来的直接结果，就是向世界各地的用户几乎毫无延迟地瞬间返回搜索结果与广告。速度由此成为谷歌最引以为傲的招牌。他们开始在搜索结果页顶部显眼地标注查询耗时（通常仅需四分之一秒左右），并高调展示庞大的索引规模。这在很长一段时间里是谷歌顶级的“技术秀肌肉”：在数十亿网页的浩瀚索引中以闪电般的速度命中海量结果并直接把数字亮给所有人。这既是工程师文化的极致体现，也是在向外界宣示无人能及的技术壁垒。所有这一切惊人的底层创新，本质上都倒逼自最初在斯坦福大学以及仅靠天使轮融资时的窘境——当时他们兜里没钱，更没有任何造血和盈利途径。 - **自主技术栈**：从 GFS（Google File System）到 MapReduce，全套自研计算与分布式存储架构。 - **Hadoop 的渊源**：雅虎因竞争压力仿制 MapReduce 并将其开源为 Hadoop，奠定了后来的大数据生态。 - **搜索速度展示**：查询结果页上标注的索引网页数量与查询耗时（约 0.25 秒），成为谷歌早期最标志性的技术背书。 --- ### [01:12:57](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4377s) 播客研究工具与早期财务数据溯源在深入探讨谷歌如何摸索出人类商业史上最赚钱的商业模式之前，节目提到了播客制作中的研究工具 Anthropic 及其 AI 助手 Claude。制作团队在查阅数十年的采访记录、SEC 监管文件和技术白皮书时深度使用了 Claude。主播分享了一次切身体会：在向 Claude 查询谷歌早期财务数据时，AI 给出的一组数字与参考书目上的记录不一致，但 AI 给出的数字内部逻辑始终完全自洽；经过仔细比对，主播才发现谷歌后来对早期财务报表做过重述修正（restated financials），而 Claude 准确捕捉到了这一细微修正，避免了节目在后续内容中出现关键事实错误。此外，Claude 的 Extended Thinking（深度思考）模式在解析谷歌当年长篇累牍的招股说明书（Form S-1）时展现出极强的长文档综合分析能力。结合对 Google Workspace 等工具的调用，AI 在协助理清谷歌 25 年纵横交错的历史脉络中发挥了巨大的检索与校验作用。 --- ### [01:15:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4500s) 九号员工萨拉尔·卡曼加与荒诞的商业计划书 ```mermaid flowchart TD accTitle: Google Series A Three-Pronged Pitch Strategy accDescr: 谷歌A轮融资计划书中提出的三条营收假想路线 subgraph CoreGoal ["核心诉求：保护 Google.com 纯粹性"] A["Google.com 搜索体验"] end subgraph RevenueProngs ["A轮商业计划书的三条变现路线"] B["1. 企业级内部搜索 (主推) 将搜索技术售卖给企业内部网/文档"] C["2. 展示横幅广告 CPM (妥协) 跟风市场常规玩法，缺乏具体落地细节"] D["3. 门户网站白标授权 OEM (对标) 向雅虎等门户输出搜索结果 (对标 Inktomi)"] end A -.-> B A -.-> C A -.-> D ``` 即便拥有惊艳的基础设施和极具成本效益的扩展能力，到了 1999 年初，谷歌天使轮筹集到的资金依然行将见底。正是在这个节点，刚从斯坦福大学本科毕业不久的萨拉尔·卡曼加（Salar Kamangar）作为第 9 号员工加入了谷歌。卡曼加在斯坦福读书时就深度使用过谷歌搜索，被其卓越体验彻底折服，于是拼命敲开谷歌的大门争取职位。入职之后，拉里·佩奇和谢尔盖·布林立刻把全公司眼下最生死攸关的任务交给了这位新员工——公司快没钱了，必须出去募集风险投资（A 轮），但拉里和谢尔盖打心底里不想去写商业计划书和路演 PPT，于是直接甩手交代：“你去把商业计划书和 Deck 写出来。” 卡曼加领命后与拉里、谢尔盖反复沟通，最终拼凑出了一份向风险投资人交差的商业构想。这个方案描绘了三条极其模糊甚至充满空想色彩的变现路径： 1. **企业内网搜索（Enterprise Search）**：这是整个计划书里被描绘为最核心、预期收入最高的支柱业务。其逻辑是将谷歌先进的消费级搜索技术卖给各大企业，帮助企业搜索内部文档和内网。这一想法的由来是因为在斯坦福时期，BackRub 与谷歌就曾被用来搜索斯坦福校内网络；而且在 A 轮融资前，拉里和谢尔盖成功把这项技术以 2 万美元卖给了开源 Linux 公司红帽（Red Hat），达成了谷歌史上的第一笔收入。当时团队心想：“既然有客户掏钱，那这里一定有个巨大的市场！”更深层的心态在于，两位创始人极度珍视 Google.com，把它当成不容玷污的圣地，因此极力渴望在主站之外通过其他技术授权赚钱来补贴 Google.com。 2. **展示横幅广告（CPM Banner Ads）**：为了迎合 VC 的传统预期而勉强写上的退路。“既然投资人非要看广告变现，那我们就写会卖 CPM 横幅广告”，但整份计划书中完全没有思考任何细节——广告究竟放在结果页还是搜索框旁边？谁也不知道，因为两位创始人当时内心深处根本不想做这件事，甚至故意在 Deck 中保持模糊以防向 VC 透露太多。 3. **门户网站白标搜索技术授权（OEM Search）**：向雅虎、AOL 等大型门户网站和分类目录提供底层网页搜索结果，作为它们的补充备选引擎。当时硅谷的 Inktomi 已经凭借这种商业模式建立起了相当规模的业务（Inktomi 甚至还是谷歌在圣克拉拉 Exodus 数据中心里的隔壁邻居）。然而，整个市场上能买得起这项服务的大门户总共也就 5 家核心客户，全部潜在客户加起来也不过 50 家左右。带着这份充满空想的商业计划书，谷歌于 1999 年春季走向了风险投资市场。彼时正值互联网泡沫的最高峰，热钱在市场上疯狂涌动。红杉资本的迈克尔·莫里茨（Michael Moritz）后来在史蒂文·利维（Steven Levy）的书中贴切地评价那个时期：“当时没有任何人的双脚是踩在地上的（Nobody's feet were on the ground）。”尽管商业模式漏洞百出，但凭借疯狂飙升的用户量、粘性与爆炸式增长指标，谷歌迎来了互联网泡沫中最关键的融资转折点。 ### [01:19:57](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4797s) 红杉与凯鹏华盈联手：不可思议的 1 亿美元估值 A 轮融资在互联网泡沫狂飙突进的眼球经济时代，Google 无疑成了风险投资圈最炙手可热的香饽饽。当时风投界最富传奇色彩的两位顶级投资人——红杉资本（Sequoia Capital）的迈克尔·莫里茨（Michael Moritz）与凯鹏华盈（Kleiner Perkins）的约翰·杜尔（John Doerr）罕见地联手破例，平分了这笔交易。Larry 和 Sergey 开出了极其苛刻的条件：两家机构各占 12.5% 的股份，必须联合领投，且两位大佬必须同时进驻 Google 董事会。坊间甚至举办了一场正式的新闻发布会，莫里茨和杜尔亲自出席站台，Larry 和 Sergey 则穿着印有 Google 标识的 T 恤亮相，这一度被外界奉为 Google 公关策略上的神来之笔。尽管红杉和凯鹏华盈此前合投过不少项目，但两位掌舵人同坐一家早期初创公司的董事席位依然极为罕见。谈判过程中甚至有竞争对手在 Ram Shriram 的引荐下开出了 1.5 亿美元估值的投资意向书（Term Sheet），但 Larry 的哥哥此前联合创立的 eGroups 正是莫里茨投资的项目，创始团队打定主意要与红杉和凯鹏华盈绑定。最终，Google 敲定了 2500 万美元融资，投后估值高达 1 亿美元。在当时即便是处于互联网泡沫的巅峰期，一家仅有数位天使投资人、几乎没有任何收入的初创公司拿到 1 亿美元的 A 轮投后估值，也震撼了整个硅谷。 - **融资金额与估值**：总计融资 2500 万美元，投后估值（post-money valuation）达到 1 亿美元，红杉与 Kleiner Perkins 各获得 12.5% 的股权。 - **竞争报价**：在谈判中途，团队曾收到另一份由 Ram Shriram 引荐、估值高达 1.5 亿美元的 Term Sheet，但 Google 最终仍坚持选择两大顶级风投机构。 - **时代反差**：当时 1 亿美元的早期估值轰动程度，堪比当下一家刚拿到少量天使轮投资的公司直接以数十亿美元估值完成 A 轮融资。 --- ### [01:23:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=4983s) 商业化困局与“我们从未花这么多钱买到这么少的东西” 手握 2500 万美元巨额资金并不意味着高枕无忧，那个时代的投资法则十分粗暴：注资、迅速跑出营收、火速上市。彼时，Larry 和 Sergey 从网景（Netscape）挖来了销售与业务拓展副总裁欧米德·科德斯塔尼（Omid Kordestani）担任首席营收官（CRO）。网景刚刚被 AOL 收购，Omid 正寻求新的挑战。来到 Google 后，他肩负着把创始团队商业计划书中的设想落地的重任，首当其冲的就是向企业推销“企业级搜索”（Enterprise Search）解决方案。然而在 1999 年下半年长达六个多月的时间里，市场上几乎没有任何客户拉动（customer pull），商业化推进举步维艰。随着时间迈入 2000 年，互联网泡沫开始显现破裂迹象，而 Google 几乎依然没有跑通任何收入模型。红杉与凯鹏华盈的高管们如坐针毡。红杉掌门人道格·莱昂内（Doug Leone）后来透露，迈克尔·莫里茨曾对他坦言：“道格，我们从未花过这么多钱，买到过这么少的东西（We've never paid so much for so little）。”虽然有幕后传闻称这句话最早可能出自凯鹏华盈的维诺德·霍斯拉（Vinod Khosla）之口，但这句犹如烫手山芋般的感叹，精准反映了风投机构面对 1 亿美元天价投资却颗粒无收时的难堪与焦虑。即便 Google 的搜索份额在飞速攀升、深受早期铁杆用户喜爱，迫在眉睫的营收压力也已如悬顶之剑。 ```mermaid flowchart TD accTitle: Google Early Monetization Dilemma accDescr: 展示 Google 从 A 轮巨额融资后面临的商业化困境及战略转向流程 subgraph Capital ["资本狂欢与重压"] A["完成 2500 万美元 A 轮融资 估值 1 亿美元"] --> B["泡沫破裂前夜的营收考核压力"] end subgraph Failure ["早期遇阻"] B --> C["聘请 Omid Kordestani 担任 CRO"] C --> D["尝试推销企业级搜索产品"] D --> E["市场买力几乎为零 '花最多钱买到最少东西'"] end subgraph Pivot ["路径修正"] E --> F["转向关键词广告业务"] F --> G["严苛限制：仅限纯文本广告 保障加载速度与体验"] end ``` --- ### [01:26:05](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5165s) 纯文本广告的坚持与麦迪逊大道的传真机时代面对企业搜索的死胡同，精明的 Omid 果断转向第二个商业方向：广告。他从纽约招募了蒂姆·阿姆斯特朗（Tim Armstrong），着手搭建广告销售团队，开始在搜索结果页顶部出售广告位。然而，Larry 和 Sergey 提出了一个在当时极不寻常且近乎苛刻的要求：Google 绝对不能投放任何横幅广告（Banner Ads）或图片广告，所有广告必须是纯文本。两位创始人对外宣称这是为了视觉审美与品味，但根本原因在于性能——任何图像都会拖慢页面加载速度，而速度是 Google 极致体验的核心生命线。这套纯文本设想让习惯了华丽视觉展示的麦迪逊大道（Madison Avenue）广告巨头们大跌眼镜。当时的销售模式仍然极其传统：按照千人印象成本（CPM）计费，广告主花钱买断某个特定关键词，只要用户搜索该词，页面就会展示相应文字广告并按展示次数开账单。更滑稽的是，当时并不存在任何在线自助服务或数字化管理后台，广告订单甚至不是通过电话沟通，而是完全依赖纸质传真。为了接单，Google 总部不得不特意添置传真机，手动录入来自传统广告代理商的投放指令。 - **纯文本底线**：创始人坚决拒绝图片与横幅广告，将页面渲染效率和搜索加载速度置于商业妥协之上。 - **古旧的售卖方式**：采用传统的 CPM 展示计费模式，依靠线下销售团队人工对接广告代理商。 - **传真机接单**：所有插入订单（Insertion Orders）全部通过物理传真机接收，再由工程师和运营人员手动录入系统配置。 --- ### [01:28:20](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5300s) 意图的力量：Jeff Dean 与亚马逊联盟链接的概念验证 Google 向广告界兜售的核心王牌是用户的“搜索意图”（Intent）。刚刚从 DEC（迪吉多）跳槽加入 Google 的传奇工程师杰夫·迪恩（Jeff Dean），接到的第一个任务便是与刚从斯坦福毕业加入公司的玛丽莎·梅耶尔（Marissa Mayer）合作，开发出一套能在不损害搜索核心体验的前提下展示纯文本广告的技术架构。为了验证大规模文本广告的可行性与用户响应率，他们灵机一动想到了亚马逊：作为 Google 早期天使投资人的杰夫·贝佐斯（Jeff Bezos）正是亚马逊的创始人。Google 顺理成章地注册成为了亚马逊联盟（Amazon Associates）的推广合作伙伴。Jeff Dean 编写了一套动态生成程序：当用户在 Google 搜索框输入任何内容时，系统会自动与亚马逊庞大的图书目录进行实时比对，只要查询与某本书籍相关，搜索结果中就会动态展示一条纯文本推广链接，诱导用户“去亚马逊购买此书”并带上 Google 的联盟返利追踪码。这场早期初创公司验证商业模式的试验取得了意想不到的成功，不仅有力地证实了搜索意图的惊人转化价值，也为日后席卷全球的搜索广告帝国奠定了最初的雏形。 ### [01:29:52](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5392s) 亚马逊联盟测试：意图驱动广告的下漏斗数据飞跃在探索广告变现的初期，Google 进行了一项巧妙的测试：挂上亚马逊的联盟营销链接（Amazon affiliate link）。从单纯的佣金金额来看，一本 10 美元的书哪怕提成 4% 或 5%，区区几十美分对 Google 当时的财务状况根本起不到决定性作用。然而，这次实验的真正价值在于向广告主提供无可辩驳的数据证明——Google 能够捕捉用户的真实购买意图，并通过匹配关键词为商家导流高转化价值的流量。广告转化的漏斗通常包含三个关键步骤：展示曝光（Impression）、点击跳转（Click）和页面内转化（Conversion）。如果仅测试点击率（CTR），搜索广告的表现本就已经远超互联网上随处乱飞的随机展示广告，因为主动发起搜索的用户带着明确意图（High intent）；但亚马逊联盟链接的独特之处在于，它让 Google 直接拿到了最底层的“下漏斗”（downfunnel）数据。Google 能精确知晓给出的曝光量最终在亚马逊促成了多少本实体书的交易。凭借这种端到端的数据闭环，Google 能够底气十足地向麦迪逊大道（Madison Avenue）的传统广告主推销：“虽然我们这只是纯文字广告，但这本质上是一个极具说服力的数学题，你们一定会爱上这组转化数据。” ```mermaid flowchart LR accTitle: Search Ad Conversion Funnel accDescr: The three-stage funnel of Google's initial affiliate ad test demonstrating high downfunnel conversion. subgraph Funnel["广告转化漏斗与验证"] A["展示曝光 (Ad Impression)"] --> B["点击跳转 (Click-Through / 高CTR)"] B --> C["下漏斗转化 (Amazon 购书成交 / 高转化率)"] end ``` - 商业逻辑验证：这次小规模尝试被主播形容为“绝对天才的测试”，成功为第一代 Google Ads 商业模式奠定了实证基础。 - 销售推进：虽然早期广告系统建立并开始在麦迪逊大道斩获广告代理商客户，但在接下来的 12 到 18 个月里，推动 Google 规模飞跃的真正核心引擎却是 OEM 门户合作交易。 --- ### [01:31:38](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5498s) 门户网站 OEM 贴牌交易与 Inktomi 模式在 Google 自主研发的搜索竞价变现体系尚未完全成型之前，互联网依然处于门户时代（Portals era）。像 Netscape、Yahoo 和 AOL 等门户巨头占据着全网绝大多数的流量。此时 Google 探索的一大收入支柱，就是效仿当时著名的搜索引擎供应商 Inktomi 的商业模式：充当 B2B 供应商，将自身的自然搜索技术白标/贴牌（White-labeling）出售给第三方门户，门户网站则按调用量向 Google 支付服务费。对当时的 Google 而言，这种合作具有双重战略意义：它不仅带来了亟需的现金流收入，更是一匹特洛伊木马——通过嵌入各大流量中心，Google 得以借助数以千万计的网民流量训练大众使用其搜索引擎。最关键的是，Google 坚持在每一页搜索结果的底部标注一行小字：“Powered by Google”（由 Google 提供技术支持）。尽管当时 AOL 拥有数千万拨号上网用户、规模远比 Google 庞大，但 Google 正悄悄借用这些巨人的肩膀完成对整整一代互联网用户的品牌心智启蒙。 - 合作属性：属于典型的 OEM 贴牌搜索服务，由 Google 提供底层的网页抓取索引与自然排名，向门户收取技术服务费。 - 关键推手：1999 年从 Netscape 跳槽至 Google 负责业务拓展的高管奥米德·科尔德斯塔尼（Omid Kordestani），迅速回到老东家 Netscape / AOL 阵营，敲定了一笔概念验证（POC）性质的合作。 - Netscape 的战略背景：Netscape 刚推出了旨在对标雅虎的目录服务，被 AOL 收购后，搜索成为其流量变现与门户战略的重要一环。 --- ### [01:34:01](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5641s) 惊险一刻：1999 年为保 Netscape 流量而关闭 Google 官网当 Netscape 正式接入 Google 搜索后，潮水般涌入的流量几乎瞬间压垮了 Google 早期脆弱的服务器基础设施。当时团队像老鹰一样死死盯住后台分析监控，联合创始人谢尔盖·布林（Sergey Brin）急促地给奥米德打电话预警：流量即将爆表，整个服务器集群濒临崩溃。在那个决定公司生死存亡的紧要关头，如果无法按约兑现服务，不仅会彻底丢掉 Netscape 的大单，更会让 Google 丧失所有企业级信誉，未来的门户大单和商业设想都将化为泡影。于是团队做出了一个今天看来不可思议却又无比理智的断然决定：**在当天彻底关闭 google.com 官网，将所有服务器算力全额调配给 Netscape 用户的搜索请求**，直到能架设上线更多实体机器为止。在今天，Google 几乎是全天候高可用性、永不宕机的代名词；然而在 1999 年，他们却毅然关停了自家的门户大门，全心全意服务 Netscape。事实证明，这一策略极其英明：Netscape 的合作单日就为 Google 导入了 300 万独立搜索用户。起初，Google 还需要在商务谈判中苦苦恳求门户网站保留“Powered by Google”的水印；但由于其索引庞大、响应极快，这行小字迅速成为了搜索领域的“Intel Inside”，反而成了各大门户标榜自身搜索品质的金色招牌。借助 Netscape 的成功标杆，奥米德在随后的时间里接连签下一批海外与中小型门户，并逐步将目光瞄准了当时互联网真正的终极霸主——Yahoo（雅虎）。 --- ### [01:36:48](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5808s) 现代软件工程的基石与赞助商 Statsig 探讨在转向 Google 拿下雅虎的历史节点前，节目展开了一段关于现代软件工程演进的讨论。Google 在许多层面上是现代意义上第一家真正定义的现代软件企业——它不仅塑造了工程师至上的企业文化、奢华舒适的园区体验以及著名的“20% 自由时间”，更深刻定义了数据驱动产品研发的标准范式：包括大规模数据分析看板、A/B 测试实验框架以及特性开关（Feature flags）机制。过去十五年间，硅谷各大科技巨头都在不惜成本在内部重复造轮子来搭建这套基础设施。而赞助商 Statsig 则将曾专属于前十名巨头的全套数据实验平台产品化，服务于 OpenAI、Figma、Brex、Notion、Anthropic 等新生代 AI 与软件公司。其产品天然适配 Snowflake、BigQuery 等现代数据仓库，每天处理数万亿级事件，免去了团队自行造轮子与对齐指标的研发负担。节目由此收束话题，重新回到 2000 年 6 月 Google 走向规模爆发的关键历史转折点。 ### [01:39:15](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=5955s) 拯救危局的 Yahoo 世纪交易与泡沫寒冬的救生索 2000 年 3 月，纳斯达克指数冲上历史顶峰，随后互联网泡沫以摧枯拉朽之势轰然破裂。然而就在整个世界仿佛陷入停滞与崩溃的关口，Google 却在 2000 年 6 月与当时互联网无可争议的霸主 Yahoo 签下了一份扭转乾坤的世纪大单。根据协议，Google 将全面接管 Yahoo 门户网站上所有自然搜索结果的垫底填充（backfills），并在搜索结果旁标注“powered by Google”的品牌标识；更关键的是，作为该合作的一部分，Yahoo 还向 Google 注入了 1000 万美元的股权投资。这笔交易彻底挽救了 Google 的生命线。在接下来的数年里，正是 Yahoo 的 1000 万美元注资加上技术服务收入，支撑着 Google 熬过了漫长残酷的“互联网寒冬”（dot-com winter），并为他们赢得了宝贵的时间去摸索出日后点石成金的 AdWords 商业模式。协议生效的第一天，Google 的日搜索请求量便瞬间翻倍，直接跃升至每天 1400 万次。到了 2001 年——即该搜索交易完整履行的第一个整年——Yahoo 单向 Google 支付的自然搜索技术服务费就高达 720 万美元。这一节点在科技商业史上往往被后人所低估。当时随着泡沫破裂，硅谷风险投资的“免费午餐列车”彻底停运，Google 根本没有任何通过新一轮风投继续融资的可能。外界彼时甚至开始着急地追问：“他们到底什么时候才能把营收开关打开？”事实上，此时的 Google 仅仅成立了两年左右，而当今的初创公司在成立前几年几乎不会被要求具备造血能力。然而，Google 运营着由昂贵人才和庞大计算基础设施支撑的重资产网络系统，既然无法再从资本市场拿钱，商业化变现与自我造血便成了活下去的唯一道路。 - **交易签署时间**：2000 年 6 月（正值 2000 年 3 月纳斯达克触顶崩盘之后）。 - **合作核心条款**：接管 Yahoo.com 的自然搜索结果，展示“powered by Google”标识；Yahoo 投资 1000 万美元。 - **流量爆发式增长**：交易生效首日，Google 搜索量暴增至每天 1400 万次。 - **现金流造血能力**：2001 全年 Yahoo 向 Google 支付了 720 万美元技术服务费，加上 Netscape 等其他门户交易，构成了抵御资本寒冬的坚固桥梁。 --- ### [01:41:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6087s) 拖延16个月的 CEO 物色战：“除非是史蒂夫·乔布斯” 在 Yahoo 交易紧锣密鼓推进的同时，董事会正不断向拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）施加巨大压力，要求他们聘请一位职业 CEO。早在当初 A 轮融资谈判时，KPCB 的约翰·杜尔（John Doerr）就非常勉强地从这两位年轻创始人手中硬生生抠出了一个承诺：日后必须引入一位所谓的“成年人/专业 CEO”。在融资完成后以及随后的数年里，拉里一直兼任 CEO，但他和谢尔盖对交出指挥权极度抗拒，故意磨洋工、拖延节奏。这一寻帅过程竟然足足耗费了 16 个月。播客主播们笑称，这绝不仅是因为找不到合适的人选，更主要是两位创始人想看看自己到底能把这件事情拖延多久。在此期间发生了一则极具戏剧色彩、堪称硅谷传奇的逸事。当时，约翰·杜尔以及来自红杉资本的迈克·莫里茨（Mike Moritz）拿出了风投机构说服年轻创始人的经典套路：带着拉里和谢尔盖周游硅谷，逐一拜访各大顶尖上市科技公司的明星 CEO，试图通过言传身教告诉他们“看吧，一位优秀的职业 CEO 能为你们的事业带来多么巨大的飞跃”。然而，两位心高气傲的创始人转了一大圈，见过硅谷名流之后完全不为所动，甚至觉得这些大牌高管平庸乏味。数月巡游结束后，他们回到红杉和 KPCB 面前，神情严肃地摊牌：“好吧，在走访的所有人里，我们认为只有一个人达到了我们的门槛，如果我们一定要请一位外部 CEO，我们只接受他来掌舵 Google。” 投资人们急切地追问是谁，拉里和谢尔盖掷地有声地回答：“史蒂夫·乔布斯（Steve Jobs）。”当时乔布斯刚刚通过苹果收购 NeXT 重返苹果不久，不仅是两位创始人的精神偶像，更是当时硅谷神一般的存在。两位创始人或许是在以此半开玩笑地戏弄风险投资人，传递出“我们的标准极其苛刻，要么是乔布斯，要么免谈”的信号。这也构成了科技史上极其讽刺的一幕——谁能想到在十年之后，苹果与 Google 将在移动操作系统领域爆发一场势不两立的世纪大战。 ```mermaid flowchart TD accTitle: 早期 Google 寻找职业 CEO 的拉锯过程 accDescr: 展示红杉与 KPCB 施压创始人寻找外部高管，创始人提出乔布斯后最终妥协选定施密特的过程。 subgraph Pressure["董事会施压与拖延战术"] A["A 轮协议承诺：聘请'职业 CEO'"] --> B["创始人消极怠工拖延 16 个月"] B --> C["投资人安排'硅谷明星 CEO 巡访之旅'"] end subgraph Climax["极具反差的交锋"] C --> D["创始人挑剔回绝所有候选人"] D --> E["提出唯一符合标准的人选：史蒂夫·乔布斯"] end subgraph Resolution["最终达成共识"] E --> F["各方妥协与权衡"] F --> G["锁定唯一符合交集的黑马：埃里克·施密特"] end ``` --- ### [01:43:40](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6220s) 埃里克·施密特登场与“三驾马车”权力架构拉锯战持续了十七八个月后，埃里克·施密特（Eric Schmidt）终于浮出水面，成为了当时几乎唯一能让投资人和创始团队双方都欣然接纳的破局者。施密特身上具备近乎完美的复合背景：他本身是正统计算机博士出身的硬核工程师，曾长期担任 Sun Microsystems 的首席技术官（CTO）；同时他又具备深厚的企业经营与上市治理经验，刚刚卸任企业软件巨头 Novell 的 CEO。更为传神的是，在价值观与文化认同的维恩图（Venn diagram）上，施密特完美契合了拉里与谢尔盖的所有挑剔条件——他甚至和他们一样，每年都会去内华达沙漠参加“火人节”（Burning Man）。 2001 年 3 月，施密特正式受邀加盟 Google。虽然拉里和谢尔盖起初对被迫让位仍心存芥蒂，但实际运转后却展现出了惊人的默契。拉里敏锐地意识到，随着公司体量急剧膨胀，CEO 职责中有大量繁琐、枯燥的事务是他极度厌恶的——比如管理庞大复杂的财务体系、法务规范以及日常行政琐事，而施密特恰好能够游刃有余地承担起这些职责。在公司最需要规模化扩张、建立成熟商业闭环的关键拐点上，这种互补被证明是无价之宝。施密特入主 Google 绝非传统意义上的“职业经理人架空创始人并将创始人扫地出门”。相反，他们建立起了一种稳固紧密的“三驾马车”（Triumvirate）联合治理机制：施密特出任首席执行官（CEO），拉里·佩奇担任产品总裁（President of Products），谢尔盖·布林担任技术总裁（President of Technology）。三人每天都会举行固定的站立例会（daily standing meeting），在高度互信的氛围下共同掌舵公司的每一个重大决策，形成了远超双人搭档的超强合力。 --- ### [01:45:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6318s) 独特的 Googliness 文化与“整合全球信息”的宏伟蓝图当施密特于 2001 年春季步入 Google 时，迎接他的是一种极其罕见且近乎排外的工程师原生文化——即日后被称为“Googliness”的组织精神。如果施密特带着传统跨国企业高管的官僚习气或强硬手腕空降，Google 的有机体势必会爆发剧烈的“器官排斥反应”。施密特极其聪明地以极度谦逊的姿态融入其中，专注倾听与学习。在入职初期甚至第一天，由于办公室空间极度匮乏，一位普通工程师径直搬进了施密特的办公室与他同桌办公，而这位身经百战的上市公司前 CEO 竟然毫无架子地与这名工程师做了好几个月的室友，顺利通过了 Google 文化的“压力测试”。早期的 Google 散发着一种乌托邦式的理想主义色彩，其运作机制比起商业公司更像是一所生机勃勃的世界顶级大学实验室： - **无知者的无畏创新**：偏爱招聘缺乏传统行业成见、极具天赋与智商的新人，因为他们“不知道什么事情是做不成的”，从而能够以颠覆性的全新逻辑去解决棘手难题。 - **对不可能之事抱有健康的蔑视**（A healthy disregard for the impossible）：这句话成为了早期 Google 团队行动的核心行为准则，无论构想听起来多么不可思议，只要方向正确就值得全力以赴。 - **不可思议的平均智商**：全员拥有极高的智力水平与强烈的协同精神，这让同期其他按部就班的商业公司相形之下甚至宛如死气沉沉的“邪恶帝国”。早在 1999 年完成首轮风投融资发布的第一份新闻稿中，Google 就确立了自己的终极使命：“整合全球信息，供大众使用，使人人受益”（To organize the world's information and make it universally accessible and useful）。这个使命定位精准到了极致：它既不过于宽泛抽象，也不流于逼仄狭隘；它在道德层面上充满理想主义与利他情怀，能够吸引全球最顶尖的心智投身其中；更为关键的是，它天然蕴含着骇人听闻的商业化变现潜力——只要能将人类社会的信息有效组织编排，掌握核心入口的人就能轻而易举地缔造出一台无与伦比的印钞机。在后来的 IPO 招股书（Prospectus）中，Google 进一步阐释了这种价值观与商业力量的统一：“我们相信，实现公司使命最有效、也是最终最有利可图的方式，就是将用户的需求置于首位。”对当时的施密特而言，在 2001 年互联网泡沫破灭、创业公司人人自危的严冬里，放弃成熟上市公司 Novell 的掌门席位，转投一家胜负未卜、虽然有流量却前途未知的创业公司，无疑是一场巨大的职业豪赌。但正是因为全心接纳并押注了这种宏大的理想主义蓝图与组织文化，施密特才真正推开了这家未来科技霸权的大门。 ### [01:49:11](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6551s) 初代广告模式的四大缺陷与 Google 的瓶颈 2000 年秋天，拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）将注意力重新转回广告业务时，对当时的现状感到非常不满。尽管初期的商业化尝试已经开始运转，广告主也比较满意，但这套模式存在致命的结构性缺陷：第一，广告依然严重依赖传统麦迪逊大道式的线下人工销售。这意味着潜在广告主的规模极其狭窄，远远无法匹配 Google 搜索引擎上汇聚的海量用户搜索意图。要想扩大收入规模，Google 只能拼命从现有的少数大客户手里压榨出巨额预算。第二，纯人工销售完全不具备技术扩张性（scalability）。广告客户每增加一批，销售团队的人头就必须同步扩张，这让 Google 看起来越来越像他们最鄙视的传统门户 Yahoo。第三，在用户体验层面，当时的广告系统完全没有“广告质量”的概念。Google 给用户最核心的价值主张是提供最优质、最高效、最精准的搜索结果，然而那些粗暴展示的广告与这一目标背道而驰，没有任何机制能保证广告内容对用户真正有用。第四，Google 白白把大把利润留在了牌桌上。他们手握高价值的搜索意图关键词，却只能按固定的千次展示成本（CPM）出售展示位，既无法参与后续转化的经济价值分配，对关键词的定价也全凭感觉“随风乱猜”。Google 意识到，必须从底层彻底重构一套全新的广告引擎。 - **人工直销瓶颈**：早期广告由销售人员逐一敲定，无法承接长尾需求，团队扩张成本极高。 - **质量脱节**：广告排序缺乏质量与相关性筛选，损害了 Google 赖以生存的用户体验。 - **商业价值流失**：单纯按 CPM 计费，无法捕捉搜索意图背后的真实商业转化价值。 --- ### [01:51:08](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6668s) Goto.com 与比尔·格罗斯的自由市场拍卖实验促使 Google 打造 AdWords 第二版（AdWords v2）的灵感来源，堪称 Google 历史上的“Instagram Stories 时刻”——他们直接借鉴并完善了该领域先驱 Goto.com（后更名为 Overture）的模式。Goto.com 由传奇创业孵化器 Idealab 的创始人比尔·格罗斯（Bill Gross）创立。1998 年 2 月，就在 Google 即将正式上线之际，比尔在 TED 大会上公布了这一极具颠覆性的构想。当时的搜索引擎普遍面临算法作弊的危机。早期按文本频率匹配的简单算法已经失效，互联网上充斥着关键词堆砌、白底白字隐蔽代码以及利用色情网站劫持流量的恶劣行径。比尔·格罗斯提出了一个极为激进的自由市场假设：搜索结果的排序应当完全交由资金出价决定。谁愿意为某个关键词支付最高的价格，谁往往就是该关键词下最好的商业搜索结果；而那些无关的垃圾网站由于转化率低下，根本无力长期支付高昂竞价。这就像电话黄页（Yellow Pages）中的付费收录逻辑一样，只有能真正解决痛点的合法企业才会高价上榜。拉里和谢尔盖是通过 PageRank 从“自然搜索”端解决了相关性排序，而比尔·格罗斯则试图通过“付费拍卖”端解决相关性。Goto.com 早期甚至完全没有自主研发的自然搜索技术，整个页面全部由付费广告组成；用户若继续往下滑动，看到的自然搜索结果则是从 Inktomi 等第三方引擎购买回来的保底内容。虽然比尔在 TED 演讲时遭到了台下听众的猛烈批评甚至嘘声，但事实证明，他的商业洞察走在了整个时代的前列。 ```mermaid flowchart LR accTitle: Goto.com 竞价排名运作流程 accDescr: 描述广告主竞价与用户搜索结果按价格排序展示的机制 subgraph 竞价环节 A["广告主自主出价"] --> B["关键词实时竞价池"] end subgraph 呈现环节 B --> C["搜索结果按出价高低降序排列"] C --> D["透明公开展示每次点击价格 （如 21¢ / 23¢ / 24¢）"] end subgraph 计费转化 D --> E{"用户是否点击？"} E -- 是 --> F["广告主支付单次点击费用（CPC）"] E -- 否 --> G["免费展示曝光（零成本）"] end ``` --- ### [01:54:03](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=6843s) 按点击付费（CPC）与自助式竞价的爆发 Goto.com 的首创机制包含两大核心支柱：其一是透明的实时关键词竞价排名。Goto.com 没有任何复杂的暗箱算法，搜索结果严格按照出价高低从高到低排列，并且极其坦诚地直接在网页上向所有人展示价格，用户可以清晰地看到某个链接标价“21 美分”、“23 美分”或“24 美分”。其二是对计费模式的根本颠覆——全面抛弃千次展示费用（CPM），开创按点击付费（Cost Per Click，即 CPC / PPC）。比尔·格罗斯在孵化 Idealab 旗下众多公司时，对传统展示广告的浪费深恶痛绝，他只想为真实发生的点击掏钱。这直接攻克了零售巨头约翰·沃纳梅克（John Wanamaker）百年来留下的商业名言：“我在广告上花的一半钱都浪费了，问题是我不知道是哪一半。”（"Half the money I spend on advertising is wasted. The trouble is I don't know which half."）在精准追踪的互联网环境下，效果广告（performance-based advertising）让广告主彻底告别了浪费。在搜索这类高意图场景中，广告主追求的不是品牌曝光，而是精准捕获用户的购买意图。这套系统还开创了全自助服务（Self-serve）：广告主只需登录网站，便可自行选词、设置出价、参与实时拍卖。凭借极低的准入门槛与高度自动化的飞轮，Goto.com 取得了惊人的爆发式增长： - **年收入破亿**：上线仅一年，Goto.com 的年化营收就飙升到了 1 亿美元，商业化速度远超早期的 Google。 - **客户规模悬殊**：到 1999 年中，Goto.com 已拥有 8,000 家广告主；相比之下，Google 直到 2000 年 10 月推出初代 AdWords 测试版时，也仅有 350 家广告主。 - **闪电上市**：成立仅一年左右，Goto.com 便成功登陆公开资本市场。 --- ### [01:57:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7020s) 错失的专利与乌托邦时代的商业阻力既然 Goto.com 拥有如此无敌的商业模式与先发优势，为何没能成为最终的绝对霸主？核心原因之一在于其致命的专利失误。比尔·格罗斯在与播客主播沟通时十分坦诚。作为一个纯粹的技术天才与商业奇才，他当时单纯地认为“按点击计费”和“关键词拍卖”是显而易见的常识，既然符合客观商业规律，就理应这么做。直到公司筹备上市前夕，法务团队才惊觉必须为此申请专利保护；然而为时已晚，比尔早在一年多前就已经在 TED 大会的舞台上面向公众详细公开了这些机制，彻底超出了美国专利申请的一年宽限保护期，从而丧失了申请专利的资格。事实上，探索付费搜索模式的先例并不止比尔·格罗斯一人。早在 1996 年，一家名为 OpenText 的搜索引擎就曾尝试引入付费搜索结果。然而在 1990 年代中期的互联网极早期，网络世界仍然笼罩在学术界浓厚的乌托邦与共享主义氛围中，强烈的反商业化情绪导致任何付费排名的尝试都会遭遇行业舆论的狂轰滥炸。即便到了 1998 年比尔在 TED 演讲时，现场依然有人大声报以嘘声；而 1996 年 OpenText 试水时的遭遇则更为惨烈。这一激进理念直到互联网泡沫中晚期商业活动爆发，才终于迎来了真正破土动工的历史契机。 ### [01:58:29](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7109s) 付费搜索的早期恶名与历史讽刺在互联网早期，付费搜索（Paid Search）一度被整个行业视为不可触碰的“高压电线”（third rail）。在 Open Text 早期尝试被舆论痛击之后，整个科技圈都留下了一种普遍宿醉般的恐惧，认为互联网用户绝不可能容忍付费搜索。当时的早期网络道德观念认为：怎敢用付费收录来污染纯粹的自然搜索结果，甚至把广告明目张胆地放在最显眼的位置？最初，拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）也持有这种近乎原教旨主义的纯粹观念。然而历史充满了巨大的讽刺。当 Google 后来推出 AdWords V2 时，为了表明自身的清白与底线，特意将广告设计成带有不同背景色底纹的侧边栏，并打上极其显眼的“Sponsored”（赞助）标签，刻意营造出“那是肮脏的付费世界，这边才是 Google 纯净美丽的自然搜索”的割裂感。但回看如今的 Google 搜索界面，现实几乎完全变成了比尔·格罗斯（Bill Gross）当年设想的样子：单列呈现，排在最前面的几条全是赞助内容，其标注甚至比当年 GoTo 标注得还要隐蔽，随后才是自然搜索结果。曾经被痛斥为大逆不道的模式，最终成了统治当今互联网商业化和搜索变现的绝对主导模型。 - 90年代中期商业化土壤尚未成熟，直到互联网泡沫充分膨胀，网络才具备足够的商业接受度让比尔·格罗斯率先试水。 - GoTo（后更名为 Overture）向世人证明了搜索乃至整个互联网到底该如何变现，直接启发了 Google。 --- ### [02:00:21](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7221s) Overture 的 B2B 帝国与雅虎的百亿美元依赖虽然 Overture 后来针对自助工具申请了一些较小范围的专利（Google 最终因专利侵权向其赔付了高达 3.6 亿美元），但其最核心的两大发明——CPC（按点击付费）与竞价拍卖（Auction），作为商业理念已事实上面向全行业开放。在接下来的两年内，Overture 意识到不仅可以在自己的站点做付费搜索，更可以将这套付费模型输出给门户网站（Portals）。正如 Google 当时在到处签自然搜索代理协议一样，GoTo 也开始在各大门户签下付费搜索变现协议，并由此彻底转型为一家 B2B 公司，随后正式改名为 Overture。 ```mermaid flowchart LR accTitle: Overture 门户分发与商业变现架构 accDescr: 展示 Overture 如何将 CPC 竞价广告系统通过白标分发至各大门户网站并形成收入分成 subgraph OvertureCore["Overture 核心引擎"] Auction["CPC 关键词竞价拍卖"] SelfServe["广告主自助服务平台"] end subgraph Portals["下游门户网站分发"] MSN["MSN / AOL / Dogpile"] Yahoo["Yahoo (占其收入约 75%)"] end Auction --> Portals SelfServe --> Portals Portals -.-> RevShare["巨额广告收入分成 (Rev-share)"] RevShare -.-> OvertureCore ``` Overture 迅速席卷了搜索分发市场，先后为 Dogpile、MetaCrawler 等提供底层广告，接着拿下了 MSN、AOL 这样的重量级巨头，最终更是拿下了雅虎（Yahoo）。单单与雅虎签下的就是一笔价值上亿美元的超级大单。彼时的 Google 还沉迷在为了几分钱的自然搜索 API 查询费苦苦求人的境地，而 Overture 早已通过白标广告分成（white label rev share）赚得盆满钵满。到后来，雅虎公司整体营收中有相当惊人的一部分（约 75%）均由 Overture 驱动的付费搜索广告贡献。 - 互联网泡沫破裂后，雅虎市值暴跌，但在严重依赖 Overture 现金流的被动局面下，雅虎在与微软的竞价战中最终以 16 亿美元的天价全资收购了 Overture。 - 雅虎当时被掐住了命脉（over a barrel），若不出手收购这家为其贡献绝大部分营收的供应商，其商业根基将完全无法自主。 --- ### [02:02:06](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7326s) 险些改写历史的收购秘闻：GoTo 曾想买下 Google 在 Overture 被雅虎收购之前，还有一段鲜为人知的行业秘史：GoTo 曾有机会以极低的价格全资收购 Google。比尔·格罗斯本人认为这桩交易绝对是天作之合——Google 拥有基于 PageRank 的世界顶尖自然搜索算法，能完美解决信息类（informational）的非商业查询；而 GoTo 则拥有无可匹敌的商业关键词付费竞价系统，擅长处理交易类（commercial）查询。两相结合，就能打造出一个融合信息与商业的最完美搜索引擎。在拉里和谢尔盖引入红杉资本（Sequoia）与凯鹏华盈（Kleiner Perkins）联合领投的那轮融资之前，两人曾主动找到比尔·格罗斯，开出了 2 亿美元的收购报价。比尔对此非常认可，认为这群年轻人确实做出了非凡的技术，2 亿美元的估值十分合理（这一估值几乎是当时那轮极高融资估值的两倍）。然而，当时估值已达 20 亿美元且已上市的 Overture 董事会却否决了这一提议。董事会的逻辑非常典型且短视：“我们凭什么要把自己这家极其重要、极具价值且高利润公司的 10% 股份，拱手让给一家没有任何收入的小作坊？”董事会认为这笔交易会极大稀释现有股东的权益，断然予以拒绝。如果这笔交易在当时侥幸达成，世人所熟知的 Google 帝国大概率将不复存在。 --- ### [02:03:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7430s) 窃取火种与 Ad Rank 的诞生到了 2000 年秋天，拉里和谢尔盖终于能够集中精力打磨 Google 自身的广告系统。当看到 GoTo 和 Overture 的商业模式大获成功时，两人不免有些懊悔，反思自己当初为何走弯路浪费了时间。拉里和谢尔盖固然是天才，但他们更可怕的第二项“超能力”在于：**绝不执着于点子必须出自自己**；只要在外部或内部发现更优秀的想法，他们能够迅速将其吸收并执行到极致。2000 年 10 月，他们指派萨拉·卡曼加（Salar Kamangar）牵头重构优化 AdWords。当时团队面临的第一个迫切任务是彻底放弃人工销售，构建一套**完全自助（self-serve）的广告投放系统**。只要还依赖人工接单，Google 就绝不可能实现按点击计费的技术架构，也无法将海量长尾的中小广告主纳入池中。 - 风险投资人蒂姆·德雷珀（Tim Draper）在考察 Overture 时曾现场验证其自助服务能力：他在笔记本上当场竞价关键词“VC”，以 1 美分、2 美分的出价亲自掀起了一场竞价战，随后决定领投 Overture。当时 Overture 拥有 8,000 多家广告主，而 Google 纯靠人工拉单仅有区区几百家。萨拉·卡曼加和拉里·佩奇在推进的同时，内心敏锐地察觉到 Overture 的模式固然先进，但还欠缺某种纯粹度，它只解决了商业模式的 75%，并且依然存在被恶意投机和劣质广告滥用的漏洞。团队开始思考：既然自然搜索端有无比精妙的 PageRank 算法保证网页质量，能否在付费广告端也设计一套机制来确保广告质量？最终的突破在于引入广告的**点击率（Click-Through Rate, CTR）**。Google 意识到，广告排位不应该单纯变成谁出钱多谁就排第一的“完全付钱上位”（pay-to-play）模式；相反，应该将广告的实际点击转化效果作为权重信号引入排序体系。这就是 **Ad Rank（广告排名）** 机制的雏形：它将源自 Overture 的自助竞价、按点击付费（CPC）与用户真实的点击率结合起来，形成了全新的商业反馈循环。 ### [02:08:05](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7685s) Ad Rank与商业模型的数学最优解 ```mermaid flowchart LR accTitle: Google Ad Rank 核心逻辑 accDescr: 展示出价与点击率结合生成广告排名并实现三方共赢的流程 subgraph Inputs["竞价与质量输入"] Bid["出价 (Max Bid)"] CTR["预估点击率 (CTR / 相关性)"] end subgraph Engine["核心算法机制"] Rank["Ad Rank = 出价 × 点击率"] Pricing["次高价格拍卖 (Vickrey Auction)"] end subgraph Outcomes["多方激励对齐"] Google["Google: 期望收益数学最优 (EV)"] Adv["优质广告主: 享受更低实际扣费"] User["搜索用户: 仅看到最相关的高价值广告"] end Bid --> Rank CTR --> Rank Rank --> Pricing Pricing --> Google Pricing --> Adv Pricing --> User ``` Google 在广告排名系统中引入了关键创新——广告排名（Ad Rank）。该算法由两个核心要素决定：一是广告主的最高出价（Bid），二是用户的实际点击率（CTR）。点击率在本质上是广告相关性（Relevance）的最佳代理指标。这一机制带来了一个精妙的商业闭环：将广告主的单次点击出价与点击概率相乘，从数学上讲，正是 Google 实现自身期望收益（Expected Value）最大化的最优解。这种设计彻底改变了广告主的博弈环境。如果某家商户的产品与搜索关键词天然契合、更受用户青睐，那么由于其点击率极高，即便出价比竞争对手低，依然能够赢得拍卖并获得靠前的排位。这就把各方利益完美对齐（All incentives are aligned）： - 用户始终能看到最相关、最有价值的商品与服务，避免了垃圾广告的骚扰； - 优质广告主凭借出色的相关性获得了“以更低出价赢得优质曝光”的特权； - Google 则在保障用户体验的前提下，达成了平台整体变现效率的最大化。 --- ### [02:09:46](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7786s) 2000年秋的初版 AdWords 与“点击自家广告”套利 AdWords 的落地是一个循序渐进的演进过程。2000 年秋季，Google 首次推出的 AdWords 版本包含了自助投放（Self-serve）和点击率质量因子（Ad Quality），但并未同步推出按点击付费（CPC）或动态竞价机制。当时广告主依然按照传统的千人展示成本（CPM）付费，Google 率先把技术难度最高的质量排名算法完成了上线。这一混合阶段迅速引发了互联网历史上极其罕见的套利奇观。既然广告排名高度依赖点击率，而广告主计费却只按展示次数（CPM）计算，各家广告主很快心领神会：疯狂雇人或自行点击自己的广告。每次点击不会产生额外的单次扣费，却能人为刷高 CTR，进而让算法误以为该广告极度相关，换来更多主搜索结果的黄金曝光。一时间，“狂点自家广告”成了当时 ROI 最高的投放黑客手段。尽管存在漏洞，但这次初试啼声让 Google 彻底确信：引入 CTR 的排名逻辑具备惊人的市场需求，他们准备好进入下一阶段。 --- ### [02:11:06](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=7866s) 借鉴 Overture、次高价格拍卖与 Sheryl Sandberg 的请教迈入 2001 年后，Google 借鉴了 Overture 的按点击付费（CPC）与拍卖模型，但在机制设计上完成了一次关键超越：引入“次高价格拍卖”（Second-Price Auction，即 Vickrey Auction）。在该规则下，出价最高的胜出者无需支付全额出价，只需支付第二名出价再加上 1 美分。例如： - A 出价 0.20 美元，B 出价 0.30 美元，C 出价 0.50 美元； - 最终 C 赢得广告位，但实际仅需支付 0.31 美元。表面上看，Google 似乎放弃了每跳一次点击多赚十几美分的机会，但这是一种极富远见的长线价值沉淀。它让广告主信任平台不会被动抬价勒索，也免去了广告主每分每秒盯着竞价面板频繁微调出价的心智负担。这契合了一种商业哲学：所有真正伟大的公司（如 Costco）体内都封存着一种“未完全压榨变现的潜能（Stored potential energy）”，正是这种克制铸就了无可动摇的飞轮。这种现代机制在 2002 年初全面上线时，传统广告主却难以理解。过去投放广告只需开一张支票、发一份传真就能买下一大包曝光，如今面对这套复杂的算法竞价无所适从。负责广告业务推广的 Sheryl Sandberg 倍感吃力，不得不打电话向她的导师——前美国财政部长 Larry Summers 求助。Larry 听后一针见血地指出：“这在经济学上叫维克里第二价格拍卖（Vickrey auction），不仅有海量的学术文献支撑其最优性，美国联邦储备委员会出售国债也是这个逻辑。你就这么去跟广告主解释。” --- ### [02:14:07](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8047s) 日落计划的阵痛、Overture 的折戟与底层工程壁垒虽然次高价格模型最终被证明是划时代的，但推行过程伴随着巨大的阵痛。Google 内部将迁移过程命名为“日落计划”（Project Sunset），强行淘汰原有的旧计费体系，要求客户全盘接受新机制。 Overture 实际上也曾尝试引入点击率，但由于其商业标签是“绝对透明”，页面上直白地展示每个排位的对应价格；引入点击率加权后，页面上的价格顺序瞬间被打乱，广告主陷入混乱并大声抗议，质疑 Overture 从透明竞价沦为了“黑盒系统”。最终 Overture 无法承受转型之痛，彻底放弃了点击率维度，这也为它日后的落败埋下了伏笔。除了商业魄力，底层技术基础设施更是天堑般的壁垒： - Google 必须在用户键入搜索词的短短几十毫秒内，在海量分布式集群上动态并发计算成千上万次第二价格拍卖； - 在 2001 年的硬件条件下，只有 Google 依靠廉价商用服务器节点构建的分布式计算系统（GFS、分布式索引架构）具备支撑这一算力的工程实力。 2001 年全年，凭借与雅虎等门户网站的合作底气，Google 实现了 8600 万美元的营收和 1000 万美元的净利润，在互联网泡沫破裂的寒冬中罕见地实现扭亏为盈。然而，这 8600 万美元几乎全系于旧有的门户合同与旧广告模式之上。“日落计划”意味着将公司刚立足的生命线置于险境，管理层顶着巨大的短期收入崩盘风险才完成了这场惊心动魄的商业跃迁。与此同时，刚刚加盟担任 CEO 的 Eric Schmidt 惊讶地发现：虽然 Google 全球超过 50% 的搜索查询来自美国本土以外，但公司在海外的商业变现几乎是一片空白。Schmidt 几乎是半开玩笑地对负责业务拓展的 Omid Kordestani 说：“你周一一大早直接去机场，到时候我打电话告诉你买去哪里的机票，随便选个国家，我们去把那里的市场搞定。”Google 就此在多个海外市场组建起精悍的初创开拓团队，正式拉开了国际化变现的大幕。 ### [02:17:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8264s) 国际业务扩张与商业模式的逆转在埃里克·施密特（Eric Schmidt）力主开拓国际市场的战略下，负责商务拓展的奥米德·科德斯塔尼（Omid Kordestani）几乎整整一年都在飞机上度过，在世界各地搭建起一套让海外代理机构通过新系统分销 Google 广告的销售网络。这套打法立竿见影：第一年，国际收入在总营收中的占比就达到了 18%，2002 年攀升至 22%，2003 年达到 29%，如今更是构成了 Google 业务的半壁江山。虽然 Google 早年也曾进入中国市场，但在经历数年围绕内容审查与网络攻击的拉锯后，最终在 2010 年选择彻底退出而非妥协。但在中国之外的全球绝大多数地区，从 2001 年施密特敲定国际化战略起，Google 便建立起了一个运转极其平稳、扩张迅猛的全球商业机器。 2002 年成了 Google 全面转向自主竞价广告系统（AdWords）的蜕变之年。这一模式的成功超出了所有人的预期：Google 2002 年的全年总营收暴增至 4.4 亿美元，相比 2001 年的 8600 万美元实现了五到六倍的爆发式增长。更关键的是，这不仅仅是数字的飞跃，更是商业基本盘的彻底换轨。在 2001 年那 8600 万美元的收入中，很大一部分来自为各大门户网站提供后台自然搜索服务的技术外包费。然而到了 2002 年年中，Google 敏锐地意识到付费搜索与 AdWords 的吸金效率高得惊人，于是产生了一个颠覆性的战略转变：停止让门户网站花钱购买 Google 的搜索技术，转而主动向门户网站付分成，换取在它们页面上独家提供付费搜索广告的权利。这一策略直接促成了 2002 年夏天震惊业界的 AOL（美国在线）世纪大单。 - **国际收入飞速爬坡**：国际收入占比从首年的 18% 跃升至 2002 年的 22%，再到 2003 年的 29%，直至成为贡献超 50% 收入的支柱。 - **业绩实现超 5 倍爆发**：年营收从 2001 年的 8600 万美元暴涨至 2002 年的 4.4 亿美元。 - **商业范式根本倒转**：从“门户向 Google 付费购买有机搜索”转变为“Google 为换取流量向门户分发付费搜索广告并给予高额分成”。 --- ### [02:19:31](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8371s) 赞助商插播：Vercel 与全栈 AI 时代的代码即时交付在深入揭秘 AOL 交易细节之前，节目插播了对赞助商 Vercel 的介绍。Vercel 如今已成为驱动现代 Web 开发以及本轮生成式 AI 浪潮的基础设施骨干。在以往的传统开发流程中，工程师必须兼顾写代码和部署运维这两件完全不同的琐事，而 Vercel 提出的“框架定义基础设施”（Framework-defined infrastructure）彻底打破了这道隔阂，能将开发者的代码自动化转化为全球分布式部署的高可用应用。借助 Vercel，企业代码交付的节奏发生了质的飞跃。以往顶尖 Web 公司大多只能做到每日一次的代码更新，而如今像 DoorDash 和 Notion 这类企业已经可以在 Vercel 上实现持续不断的高频交付，做到“AI 生成代码有多快，线上部署就有多快”。Vercel 过去被视为前端开发的代名词，如今已全面演进为支持后端和智能体（Agentic）工作负载的全栈 AI 平台。其旗下的 AI 代码生成产品 v0 拥有 250 万用户，累计生成了超过 1 亿条消息，平台平均每秒可完成 6 次完整的应用生成，并且整个系统完全跑在 Vercel 自己的基础设施之上。 --- ### [02:21:35](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8495s) 雅虎特里·塞梅尔的傲慢出价与 50 亿美元反击为了理解 2002 年 Google 世界被改写的速度有多么震撼，节目回顾了几个月前发生的一段极富戏剧性的往事。2001 年底，好莱坞传媒巨头出身的雅虎新任 CEO 特里·塞梅尔（Terry Semel）与拉里·佩奇（Larry Page）、谢尔盖·布林（Sergey Brin）共进晚餐。当时塞梅尔居高临下地对两位创始人说，雅虎为了采购 Google 的自然搜索结果每年支付不到 1000 万美元，并以此断言：“既然我们是你们最大的客户，而且只付了不到 1000 万美元，那说明你们根本就没有成型的商业模式，你们的生意根本做不大。” 当时 Google 仍处于 IPO 前极度保密的状态，外界根本不清楚其真实流水。面对塞梅尔的质疑，拉里和谢尔盖只是含蓄地表示正在秘密筹备一些极具前景的新业务——他们心中所想的正是即将上线的 AdWords 系统。然而塞梅尔完全没意识到即将来临的风暴，直接当场开出 10 亿美元的高价意图吞并 Google。然而仅仅过去一年，随着 Google 营收在 2002 年一举突破 4.4 亿美元，形势彻底逆转。看到 Google 恐怖的吸金能力后，雅虎慌忙提价到 30 亿美元意图再次求购；面对这一报价，拉里和谢尔盖反手甩出了 50 亿美元的要价。这笔世纪收购最终不欢而散，也标志着雅虎彻底失去了全资拥有 Google 的最后机会。两位创始人提出的 50 亿美元实际上更像是一记充满讽刺的逐客令：当时雅虎自身市值遭受互联网泡沫重创，如果真以 50 亿美元的估值推进并购，在结构上将演变为 Google 对雅虎的反向收购（Reverse Takeover）。收购告吹后，雅虎转而在不久后以 16 亿美元的天价买下竞价排名鼻祖 Overture，这也吞掉了雅虎当时很大一部分市值。 ```mermaid flowchart TD accTitle: 雅虎与 Google 的收购博弈流程 accDescr: 展示2001年底特里·塞梅尔10亿美元报价到2002年50亿美元反讽要价的演变过程 subgraph Phase1["2001 年底：傲慢的误判"] A["塞梅尔与创始人晚宴 认为 Google 年入不足千万无商业前景"] --> B["当场开出 10 亿美元收购报价"] B --> C["Google 暗中研发 AdWords 并拒绝收购"] end subgraph Phase2["2002 年中：惊人的逆转"] C --> D["Google 营收翻超 5 倍至 4.4 亿美元"] D --> E["雅虎恐慌性加价至 30 亿美元求购"] E --> F["Google 反讽开价 50 亿美元（等同反向收购雅虎）"] F --> G["谈判破裂：雅虎转而以 16 亿美元收购 Overture"] end ``` --- ### [02:25:02](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8702s) 押上公司身家的豪赌：拿下 AOL 流量命脉眼见全新 AdWords 的飞轮彻底转动，Google 迅速将目标锁定在了当时全网的流量霸主——拥有高达 3400 万活跃用户的 AOL（美国在线）。在宽带尚未普及的 2002 年，AOL 是绝大多数普通网民接入互联网的绝对入口。当时 AOL 的搜索服务处于割裂状态：自然搜索由 Inktomi 提供技术支持（双方自 1999 年起已合作三年），而付费广告搜索则由 Overture 在过去两年中独家提供。Google 此时带着“自然搜索算法 + 自助竞价系统”的全套完整方案杀入竞争，试图以一己之力同时吃下 AOL 的有机结果与广告两块业务。最终 Google 赢得了这场角逐，并在合同中设立了一个颠覆性的关键条款：所有的搜索广告必须全部经由 Google 的后台界面统一录入投放。即使是与 AOL 拥有长期合作关系的品牌大客户，AOL 也只能把他们引导至 Google 的专门页面下单。作为交出搜索广告控制权的对价，AOL 获得了极尽优厚的条件： 1. **超高广告分成**：Google 承诺将从搜索广告中赚取的所有收入，按照“每一美元分给 AOL 85 美分”（85% 分成比例）返还给 AOL。 2. **股票认股权证（Warrants）**：AOL 获得了以每股 3 美元的行权价认购 740 万股 Google 股票的特权，总计约合 2200 万美元的潜在投资额度。 3. **1 亿美元保底承诺（Revenue Guarantee）**：无论这套广告系统最终表现如何、哪怕全盘崩溃，Google 都必须在一年内至少向 AOL 支付 1 亿美元现金。这笔交易在 Google 内部引发了极其剧烈的争议。问题的致命之处在于：在 2002 年 5 月双方进行合同谈判时，整个 AdWords 才刚刚显现成功迹象，Google 账户上甚至根本没有 1 亿美元现金。正如谢尔盖·布林事后坦承的那样：“我们当时可能会因此破产。”（"We could have gone bankrupt."）这项协议在本质上是一种激进的财务杠杆操作，Google 承担了高达 1 亿美元的刚性兑付义务。虽然拉里和谢尔盖对广告系统的变现效率拥有极致的数学信心，但一旦在 AOL 的庞大流量上出现任何水土不服、导致广告点击率和单价达不到预期，Google 就将从原本小有盈利的初创公司直接跌入万劫不复的破产深渊。这一极具魄力的重大决策，最终完全依靠拉里与谢尔盖力排众议的坚决态度才得以强行落地。 ### [02:27:54](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=8874s) 押注全公司的 AOL 豪赌与双边流动性飞轮当谷歌决定与 AOL（美国在线）签约并承诺巨额保底收益时，负责业务拓展的奥米德·科德斯塔尼（Omid Kordestani）对拉里·佩奇直言：“如果这么做，你是在拿整家公司去赌。”拉里的回答极其果断且充满魄力：“我们理应有能力将这些页面变现。如果做不到，那我们活该倒闭。”正是这种魄力，促成了合作中最核心的条款——所有广告主必须统一接入谷歌的系统。这也印证了后来 Meta 首席技术官安德鲁·博斯沃思（Andrew Bosworth，昵称“Boz”）向扎克伯格和雪莉·桑德伯格提出的洞察：“广告越多，广告就越好”。在基于算法排序和精准定向的动态展示系统中，系统内拥有的广告库存越庞大，算法在挑选最佳广告时的候选集就越丰富，广告匹配度自然越高。将 AOL 庞大的搜索广告需求与库存一次性导入谷歌系统，在战略上带来了无可估量的价值。本质上，这是一场关于“市场流动性”（Market Liquidity）的革命。双边市场的交易量越大、流动性越深，就越容易为用户的每一次搜索意图找到完美契合（Product-Market Fit）的广告。多位谷歌早期员工后来指出，如果 AdWords 从一开始就以完全成熟的竞价拍卖形态问世，在没有足够广告主库存的冷启动阶段极难运转——在流动性匮乏的浅层市场中，根本无法为用户提供高质量、不打扰的广告体验。虽然搜索广告本身谈不上令人“愉悦”，但从经济学角度看，它堪称人类商业史上最“神奇”的交易机制：广告主得以在潜在客户意图最强烈的精确时刻直达对方。这也是为什么谷歌能凭借这一商业模式赚取超过苹果、微软或伯克希尔·哈撒韦的惊人利润。 - 竞争对手 Inktomi 当时对《华尔街日报》评论 AOL 的决定时断言：“他们迟早会发现，谷歌是在夺走你的用户，而不是帮你建设门户资产。”这番预言在日后完全应验，门户网站逐渐式微，而谷歌成了互联网的终极入口。 - 这场押上 1 亿美元保底承诺的豪赌迎来了惊人的回报：仅在 2002 年下半年（合作前六个月），AOL 就分得了 3500 万美元；到了 2003 年，这一数字飙升至 2 亿美元，轻松击穿保底额度。 - 这一战让谷歌在一夜之间从边缘玩家跃升为付费搜索市场的巨头，狠狠击溃了此前独霸该领域的 Overture。 --- ### [02:31:41](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9101s) 雪莉·桑德伯格搭建销售战队与全网流量进攻这场惊天交易的落地，也成为了雪莉·桑德伯格（Sheryl Sandberg）在谷歌大显身手的转折点。彼时她刚加入谷歌不久，正四处探寻适合自己的业务切入点，公司一度给她挂了类似“业务单元经理”的头衔，但当时的谷歌根本没有成型的业务单元架构。奥米德找到她说：“你不是在找一份大展身手的工作吗？我们刚签下了 AOL 这个庞然大物，涌入了成千上万个全新门类的广告主，而我们目前完全不知道该怎么去服务。我们需要一支聪明、适应力极强的队伍来承接这些需求，并且他们必须学会自我规模化，不断向技术部门输出产品需求，从而通过自动化工具提升人效，避免公司陷入无休止的人海战术。”桑德伯格随即招揽了一大批顶尖人才，从零搭建起支撑起整个 AOL 天量业务的 AdWords 销售与运营体系，这也成为她日后前往 Facebook 出任首席运营官之前最核心的战绩。消化完 AOL 交易的红利后，谷歌彻底意识到其核心战略价值：只要能在全互联网的任何角落把谷歌搜索（无论是自然搜索还是付费广告）嵌入用户体验，就必须不惜一切代价去做。这一模式能持续做大广告流动性池子，进而“将整个互联网变现”。 ```mermaid flowchart LR accTitle: 搜索广告网络规模效应正循环 accDescr: 展示广告主规模扩大如何通过提高相关性与点击率推动全网分发扩张 subgraph LiquidPool ["双边流动性池"] A["引入 AOL 等外部天量库存"] --> B["广告主数量与类别急剧膨胀"] B --> C["候选广告池大幅扩充"] end subgraph Optimization ["体验与商业化优化"] C --> D["精准匹配查询意向 (更优质的广告体验)"] D --> E["点击率提高 & 竞价攀升"] end subgraph Expansion ["分发扩张"] E --> F["单次搜索变现收益增加"] F --> G["有实力以极高分成 争夺全网分发渠道"] G -.-> A end ``` --- ### [02:33:38](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9218s) 规模报酬递增法则：改写分发逻辑的商业飞轮搜索本质上是一个赢家通吃的市场，但其底层动力远不止传统消费互联网的横向网络效应或固定成本摊销。在传统的规模经济中，生产 1000 个零件的单件固定成本低于生产 10 个，谷歌同样享有摊销基础设施和工程人力的成本优势。但搜索广告拍卖机制具有一种极其罕见的反常识特性：**在规模扩张的同时，谷歌不仅单位成本在下降，单位搜索的收入（Revenue Per Search）反而在持续上升**。这一“规模报酬递增”（Increasing Returns to Scale）的奥秘源于两重机制： 1. **关键词价格发现与竞价推高**：每个关键词下的竞价广告主越多，微型拍卖市场的价格发现就越充分，胜出者的出价必然水涨船高。 2. **长尾查询的全面变现**：若仅有全球最大的 100 家广告主，系统只能变现头部通用词；而当拥有海量长尾广告主后，冷门罕见的搜索词也能命中广告，使得任意一次搜索转化为商业收入的概率大幅提升。这一逻辑将 Boz 的洞见进一步推导为完整闭环：“更多广告 = 更好广告 = 更好商业模式”。既然每一次搜索的价值更高，意味着每个用户的生命周期价值（LTV）就更高；这赋予了谷歌远超所有竞争对手的资金实力，去以极高代价砸钱抢夺用户获取渠道（CAC）。在这个飞轮下，获取渠道分发 -> 带来更多搜索量 -> 推高关键词竞价 -> 带来更高单次搜索收入 -> 支持出更高的价格采购更多分发。 - 在 2001 至 2002 年间建立起初步领先优势后，谷歌不能仅依赖出色的自然增长与口碑传播“守株待兔”，必须以极具进攻性的姿态迅速扫荡市场，防范对手参透这一逻辑。 - 在具体的分发战术上，谷歌甚至激进到愿意向渠道分发伙伴让渡高达 100% 的广告分成收入，用短期让利换取绝对的市场流动性垄断。 ### [02:37:21](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9441s) 激进的分发分成与不对称竞争优势在早期的渠道分发谈判中，Google 表现出了极其激进的让利姿态。在与 AOL（美国在线）的合作中，Google 甚至愿意将高达 85% 的搜索广告收入直接分给 AOL。这种看似不可思议的分成比例背后，隐藏着一套降维打击的商业算力：由于 Google 单次搜索的变现效率（monetization per search）远高于所有竞争对手，即使在相同的分成比例下（例如双方都给 70%），Google 能带给分发渠道的绝对美元金额也远远超过对手。为了迅速锁定关键渠道，Google 不惜将分成比例推高到 100%，甚至在某些极端案例中给出超过 100% 的分成——自掏腰包补贴渠道，只为抢夺宝贵的流量入口。这种策略使得竞争对手在经济层面上根本无法跟进，任何试图与 Google 进行价格匹配的对手都会迅速耗尽资金。此时仍为私有公司的身份成了 Google 的秘密武器：当时雅虎（Yahoo）等主要搜索引擎已经上市，必须每季度公布稳定可观的财务指标；而未上市的 Google 则可以采取看似“不理性”的策略，承担单季度亏损的风险，以超额成本横扫分发渠道。核心在于，Google 管理层坚信搜索绝非一个仅有 10 亿或 100 亿美元级别的小买卖，而是一个未来年收入达半万亿美元（half a trillion dollars）的超级市场。在这种坚定信念下，只要能建立先发优势并成为规模最大的玩家，在分发端投入任何天文数字的资金都是值得的。 - **超高分成比例**：为拿下 AOL 合作开出 85% 分成，甚至部分渠道分发给出 ≥100% 的返利。 - **单搜索变现溢价**：更高的 ARPU 允许 Google 即使在极端分成比例下依然令竞争对手破产。 - **私有公司优势**：无需受制于上市公司的季度财报考核，敢于承担短期严重亏损换取长期垄断地位。 - **市场终局信念**：在全行业还在怀疑搜索商业回报上限时，Google 已坚定押注半万亿级别的年收入规模。 --- ### [02:39:34](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9574s) 抢占微软浏览器高地：Google Toolbar 的降生拿下了门户网站的分发合作后，Google 面临另一个根本性的分发痛点：用户必须主动在浏览器里输入 `google.com` 才能进行搜索。这种依赖口口相传、手动输入网址或收藏书签的方式极其脆弱。更致命的是，Google 当时并没有自己的浏览器，而全球浏览器市场几乎被微软（Microsoft）的 Internet Explorer 牢牢统治。在整个硅谷，微软是唯一拥有足够财力在搜索领域砸巨资围剿 Google 的庞然大物。如果微软利用 Internet Explorer 封锁流量并力推自己的搜索引擎，Google 将毫无还手之力。 ```mermaid flowchart TD accTitle: Google Toolbar 战略飞轮 accDescr: 展示 Google Toolbar 如何通过提升搜索频次与 ARPU 为渠道买量提供巨额预算并抵御微软浏览器垄断。 subgraph Defense ["防御与入口卡位"] A["微软 IE 垄断浏览器市场"] --> B["2000年12月推出 Google Toolbar"] B --> C["搜索框直接置顶于浏览器内"] end subgraph EconomicLift ["价值提升与买量预算"] C --> D["单用户搜索量暴增 7 倍"] D --> E["普通用户年 ARPU: $2 Toolbar 用户年 ARPU: $10+"] E --> F["获得 ~$8/用户的丰厚套利空间"] end subgraph DistributionFlywheel ["木马分发与生态渗透"] F --> G["付费捆绑预装 Adobe / WinZip / RealNetworks"] F --> H["OEM 厂商合作 Dell PC 出厂预装"] F --> I["Firefox 默认搜索引擎分成"] G & H & I --> C end ``` 2000 年 12 月，在公司成立仅两年半、AdWords v2 尚未出炉之际，Google 提前推出了 Google Toolbar（谷歌工具栏）。当时外界普通用户普遍认为这是“仁慈的 Google 神”免费赐予互联网的一款绝佳小工具，但实际上它是一招兼具进攻与防御的绝妙棋局： - **产品形态**：作为浏览器插件（相当于今天的浏览器扩展），它嵌在 Internet Explorer 书签栏下方的顶部位置，自带 Google 搜索框。用户无需打开网页，在任何页面都能随时发起搜索，这彻底启发了后来将地址栏与搜索框合二为一的现代浏览器设计。 - **核心数据暴增**：安装了 Google Toolbar 的用户，其平均搜索频次猛增至普通用户的 **7 倍**。 - **单用户价值（ARPU）激增**：普通 Google 用户每年的广告变现收入约为 2 美元，而工具栏用户即使保守估计也超过 10 美元。这之间至少 8 美元的单用户增量价值，直接转化成了 Google 大举收购分发渠道的“战争基金”。 --- ### [02:43:25](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9805s) “木马式”捆绑买量与奇招破局：从弹出窗口拦截到 Google Earth 手里握有高达每用户约 8 美元的增量价值空间，而实际平均买量成本远低于此，Google 开始在软件分发领域展开扫荡式的付费预装合作。Google 找到当时 Windows PC 上用户必装的主流软件开发商，支付巨额费用将 Google Toolbar 捆绑进安装包中。当用户下载 Adobe 应用、WinZip 解压软件或 RealNetworks 播放器时，安装程序便会像特洛伊木马（Trojan horse）一样将 Google Toolbar 植入用户的 Internet Explorer 浏览器中。尽管用极端刺耳的词来形容，这种打法类似“广告软件（Adware）”或“特洛伊木马”，但因为该工具栏本身极具实用价值，用户不仅不反感，反而极其喜爱。为了进一步让用户产生无法割舍的黏性，Google 捕捉到了当时网民最大的痛点——泛滥成灾的弹窗广告（pop-up ads）。Google 顺势将“弹窗广告拦截器（Pop-up Blocker）”直接做进了 Google Toolbar 中，使之成为安装工具栏的最强动力。随后，Google 将这种分发打法推向了极致： - **PC OEM 厂商预装**：与戴尔（Dell）达成直销协议，在新出厂的 Dell 电脑中直接预装带有 Google Toolbar 的 Internet Explorer。 - **扶植 Firefox**：在火狐浏览器崭露头角时，Google 斥巨资将其买断为 Firefox 的默认搜索引擎，这笔资金在长达数十年的时间里构成了 Mozilla 基金会最核心的收入支柱。 - **Google Earth 的商业化神操作**：Google 收购以 3D 虚拟地球技术风靡全网的 Google Earth（当时为独立客户端软件）。团队最初曾讨论如何在 3D 地球中植入商业广告，但最终发现软件极度病毒式传播，于是直接将 Google Toolbar 捆绑进 Google Earth 的安装器中。工具栏带来的搜索变现增量，直接完全覆盖并远远超越了 Google Earth 自身的研发和运营成本，彻底省去了在软件内部强推广告的麻烦。随着时间迈入 2000 年代中期，Google 用户的 ARPU 从 10 美元一路狂飙至 20、30 美元乃至更高，而工具栏用户始终保持着更高的留存和搜索频次，为 Google 赢得了持续滚雪球的巨大分发优势。 ### [02:46:38](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=9998s) 分发渠道的战略布局与 Google 商业飞轮 Google 的成功绝不仅仅是因为拥有最出色的产品、飞快的检索速度、顶尖的技术实力以及极具极客色彩的企业文化。人们往往忽略了他们在渠道分发拓展上的极度激进与精准——Google 从来不坐等用户“奇迹般地”自己找上门来。从早期向各类门户争取默认搜索引擎，到如今备受反垄断争议的 Apple Safari 巨额交易（每年向 Apple 支付数十亿美元的流量获取成本 TAC），这条极度重视分发渠道的脉络一脉相承。 ```mermaid flowchart TD accTitle: Google 飞轮效应与生态扩展 accDescr: 展现 Google 搜索广告变现能力支持分发渠道，进而带来海量用户与内容变现的闭环系统 subgraph Core ["搜索与分发核心"] A["高意图搜索 + AdWords 高效变现"] --> B["充沛资金预算投入分发渠道 (如 Toolbar / Safari)"] B --> C["沉淀海量黏性用户并占领流量入口"] end subgraph Expansion ["内容生态延伸"] C --> D["拓展第三方网络内容变现: AdSense 诞生"] D --> E["赋能中小网站 / 创作者分成"] E --> A end ``` 这种严丝合缝的体系让人联想到 Costco 的商业模式：每一个业务模块不仅自洽，而且相互强化，犹如一场精妙绝伦的芭蕾舞，共同铸就了人类商业史上最强大的变现机器。为了牢牢绑定用户，Google 当时尝试了五花八门的桌面客户端软件。除了大获成功的 Google Toolbar（工具栏），他们还推出过名为 Google Desktop 的桌面搜索应用，能抓取本地文件并将其私密整合到网页搜索结果中，宛如早期 Google 企业搜索套件的个人重生版。2004 年，一位新产品经理加入 Google，接手了负责 Google Toolbar 等桌面客户端的应用团队——这位产品经理正是后来的 Alphabet CEO 桑达尔·皮查伊（Sundar Pichai）。 ### [02:48:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10124s) AdSense：从“搜索意图”跨入“网页内容” 在筹划首次公开募股（IPO）之前，Google 迎来了商业模式的又一次重大跃升：推出 AdSense。在此之前，Google 的变现逻辑完全依托于“单次搜索查询（Search Query）”这一基础原子单元。搜索请求具有极高的商业意图，单次价值极为惊人；然而在现实中，用户在互联网上的绝大部分时间其实是在浏览各类第三方网页、阅读文章与消费内容，而不是在 Google 首页上敲下关键词。面对全网广阔却无法直接变现的静态内容，Google 产生了一个绝妙的构想：既然他们拥有成熟的 PageRank 算法与网页内容解析技术，同时掌握着预测点击率（CTR）的广告竞价匹配系统，何不将这套能力逆向运用到发行商的静态网页上？通过扫描页面内容，逆向推导“什么样的搜索关键词会引导用户来到这个页面”，进而直接匹配并展示对应的广告。虽然这属于“内容匹配（Contextual Match）”而非直接的“意图匹配（Intent Match）”，但浏览特定内容的用户往往具备潜在的关联需求。 * **库存与竞价的平衡**：2003 年初，广告主对 Google 广告位的需求极为旺盛，搜索页面的广告位甚至处于严重的“库存短缺”状态；如果供给过剩，竞价拍卖系统就只会收到一两美分的极低出价，保持一定程度的供给约束反而放大了系统的拍卖效率。 * **复用现有资源**：AdSense 无需从零招募广告主，而是直接调取 AdWords 广告池中现成的广告素材与竞价客户，将它们直接复用到第三方页面。 ### [02:51:39](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10299s) 传奇六周：AdSense 的极速诞生与创作者生态雏形 2003 年 2 月，这个构想正式提上日程。传奇工程师杰夫·迪恩（Jeff Dean）仅仅用了六个星期，就凭一己之力硬生生写出了整个 AdSense 系统的初代代码。团队先是在 Google Groups 上进行内部测试，随后决定购买第三方网站的传统展示广告位来进行实地灰度验证——他们买下横幅广告位，但里面呈现的却是由 AdWords 填充的文字链接广告。 ```mermaid flowchart LR accTitle: AdSense 早期内容广告匹配流程 accDescr: 说明 Google 如何解析第三方内容页面并分发文字广告完成变现分成 A["第三方网站静态网页 (如 howstuffworks.com)"] --> B["Google 算法扫描内容 推导对应潜在关键词"] B --> C["AdWords 广告池匹配 逆向呈现对应文字广告"] C --> D["用户点击产生收入"] D --> E["站长获取绝大部分分成 (67% - 80%)"] D --> F["Google 留存佣金利润"] ``` 在测试期间，团队最钟爱的试验田是知名科普网站 `howstuffworks.com`。该网站各页面的主题高度垂直且具有极强的实用性，逆向工程出来的搜索关键词商业价值极高。随后，苏珊·沃西基（Susan Wojcicki）作为产品经理全面接管并推进这一业务。AdSense 虽然需要与发行商分成，利润率不及纯粹的第一方搜索广告，但它给 Google 瞬间带来了数亿美元的新增流水。 * **分成机制**：通常第三方网站站长能够拿走点击广告收入的绝大部分（不同时期约在 67% 至 80% 之间），Google 仅抽取较小比例的服务提成；这与此前 Google 为 AOL 等门户网站提供搜索技术支持并进行收益分成的逻辑如出一辙。 * **初代创作者经济**：在此之前，中小型站长想要变现极具门槛。而 AdSense 只需要站长在网页里嵌入一段 HTML 代码，Google 就会源源不断地向其银行账户汇款。这直接奠定了日后 YouTube 创作者收益分成体系的原型。 * **指数级爆发**：2003 年第一季度末 AdSense 刚刚上线，到 2003 年年底，该业务每天产生的流水就已经突破了 100 万美元。 ### [02:55:01](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10501s) 20% 自由时间、Google News 与算法的进化在走向上市的最后阶段，Google 内部充满了由工程师驱动的创新活力。公司著名的“20% 时间（20% Free Time）”机制在此期间催生了大量令人耳目一新的项目，Google Labs 也成为了向外界展示这些前沿实验创意的集结地。其中，Google News（谷歌新闻）正是某位工程师在 20% 时间里做出来的产物。特别是在“9·11”事件之后，全网用户迫切渴望一种能够汇聚海量媒体、对突发热点事件进行实时更新的新闻聚合工具，Google News 应运而生并迅速捕获了巨大流量，同时也悄然拉开了 Google 与新闻出版业之间数十年版权纠葛的序幕。与产品线扩张同步进行的，是自然搜索排名算法（Organic Search Ranking）的深度演进。Google 不再仅仅单纯依赖最初的 PageRank 和网页超链接锚文本（Anchor Text），而是开始引入越来越复杂的特征变量。到 2007 年，Google 计算单次搜索结果排序时参考的信号维度已经扩展到了 200 多种，搜索技术的壁垒正在以极快的速度向外拓宽。 ### [02:56:14](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10574s) 搜索永无止境：数据飞轮与持续迭代的工程哲学 Google 逐渐建立起强大的数据网络效应：随着使用搜索的人越来越多，算法能够从海量真实流量中获得实时反馈。如果大量用户在搜索某个特定词条后迅速从某个页面跳出（bounce off），系统就会敏锐地识别出该结果并不匹配用户意图，从而调整排名。另一个方向则是“个性化”（Personalization），即系统基于用户的历史行为进行学习推荐——尽管当时还没有 Google 账户体系，个性化能力受到一定限制，但这为未来的生态爆发埋下了种子。很多人误以为 PageRank 是 Google 至今最核心的护城河，但事实上，PageRank 仅仅是最初起跑的引信，它在今天的算法体系中早已不再是决定性的差异化资产。这体现出拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）与其他搜索创业者截然不同的工程哲学。像其他同期的搜索引擎，往往觉得找到了某种技术亮点就大功告成，自以为“已经解决了搜索问题”；但 Google 从未认为搜索问题是可以被一劳永逸解决的。他们深知互联网内容的呈指数级爆炸，其增长速度远超任何爬虫索引或分类策略的处理极限。既然世界是动态变化的，Google 永远无法实现绝对的最优解，唯一的选择就是持续进行巨大的研发投入，不断逼近最优结果。 ```mermaid flowchart TD accTitle: Google搜索的数据飞轮与迭代机制 accDescr: 展示用户查询、行为反馈、基础设施支撑及动态词库如何持续改进搜索算法质量。 subgraph UserFeedback ["用户行为与数据飞轮"] Q["用户发起搜索 Query"] --> R["返回搜索结果并监测行为"] R -->|"跳出率 / 停留时间"| FB["识别相关性缺陷"] R -->|"修改搜索词 (如 cat -> kitten)"| SYN["自动学习同义词匹配"] end subgraph CoreEngine ["核心系统与工程投入"] FB --> ALG["算法持续动态调优 (超越 PageRank)"] SYN --> DIC["实时动态同义词库"] INFRA["自研底层架构 (GFS / MapReduce / 硬件)"] --> ALG end ALG --> Q ``` ### [02:57:52](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10672s) 基础设施演进与精妙的同义词算法除了算法思想的进阶，Google 构筑的第三大壁垒是深不可测的基础设施。早期他们以擅长将廉价的商用硬件（commodity hardware）拼装利用而闻名，带着一种野蛮生长的创业极客色彩；而到了这个时期，这种策略彻底升格为全套自研数据中心、自研底层文件系统（GFS）以及革命性的分布式计算工具（如 Bigtable 和 MapReduce）。除了这些底层系统，Google 在搜索软件细节上也做到了极致，其中极具代表性却常被外界忽略的一项技术是“同义词匹配”（Synonym Matching）。如果用户搜索“猫（cat）”，而网上有一篇极具价值的内容通篇使用的是“小猫（kittens）”，缺乏同义词理解的搜索引擎就永远无法命中。Google 采用了一种极为巧妙的数据洞察机制：他们观察到很多用户在搜索“cool cat pictures”无果后，紧接着会把词条修改成“kitten”；当系统在海量日志中频繁捕捉到这种微小的行为修正时，就能自动推断两者为同义词。Google 借此在后台构建起一个能够实时自我更新的庞大同义词字典。像这样深藏在冰山之下的精巧技术，Google 内部同时运行着上千个。 ### [02:59:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10767s) 跨越资本约束门槛：财务核聚变与 AdSense 的利润构成到了 2003 年底，Google 迎来了历史性的拐点：他们彻底摆脱了资本的约束。在 2002 年年中之前，公司在扩张和探索中还必须谨小慎微地做权衡取舍；但从 2002 年底开始，Google 的账上永远拥有充足的现金去资助他们想尝试的任何疯狂想法。 - **2002 年财务表现**：营收 4.4 亿美元，营业利润达到惊人的 1.85 亿美元。 - **2003 年财务核聚变**：年营收暴增至 15 亿美元，营业利润更是逼近 3.5 亿美元，营业利润在一年内实现翻番。 - **利润率结构的改变**：表面上看，随着 AdSense 的推出，Google 整体利润率似乎有所下滑。这是由于商业模式的本质差异：AdWords 广告费 100% 归 Google 所有，毛利率高达约 85%；而 AdSense 属于外部合作网络，Google 仅保留约 20% 的分成，其余 80% 都要结算给发布商网站。尽管 AdSense 在 2003 年贡献了约 5 亿美元的低毛利收入，但它在战略和生态上带来了极为可观的增量价值。 ### [03:00:49](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=10849s) 为什么 Google 极力抗拒 IPO？时间的车轮推进到 2004 年，迎来了商业史上著名又充满争议的 Google IPO。在今天看来，这是一次无比成功的上市，但在当年，这场 IPO 却饱受传统金融界的抨击，甚至被普遍认为是一场“灾难性的失败”。事实上，纵观整个商业史，除了早期的微软，几乎找不出第二家像 Google 这样完全没有任何理由想要上市的公司——它极其赚钱，自由现金流汹涌澎湃，根本不需要公开市场的融资，甚至上市募资得来的钱他们很可能一分都没动用过。更关键的是，Google 有一个极其致命的理由抗拒公开上市：**微软**。当时 Google 内部弥漫着深重的偏执与焦虑，他们万分提防雷德蒙德的这个巨无霸。微软通过 Internet Explorer 掌控着当时互联网事实上的入口，并且拥有无穷的资本实力。当时微软根本不知道搜索是一门多么暴利的天堂级生意，Google 极力避免让微软看到自己惊世骇俗的财务报表，生怕唤醒这头沉睡的巨兽前来正面绞杀。然而，彼时美国尚未通过《创业企业扶助法》（JOBS Act），根据当时的证券监管规定，一旦一家私人公司的独立股东人数超过 **500 人**，就必须强制公开财务报表，承担与上市公司无异的披露义务。既然必须公之于众，隐瞒利润已无可能。与此同时，尽管红杉资本（Sequoia）和凯鹏华盈（Kleiner Perkins）在董事会没有绝对投票权（拉里、谢尔盖及员工期权池紧握绝大多数控制权），但在刚经历过互联网泡沫破裂的惨重打击后，这些封闭式风险投资基金迫切需要资金清算退出以拯救元气；更重要的是，在 Google 当时的 2,000 名员工中，这场 IPO 直接让其中的半数人一夜之间晋升为百万富翁，内部积累的流动性变现诉求已势不可挡。 ### [03:04:17](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11057s) 华尔街的潜规则与“荷兰式拍卖”的引入到了 2003 年底至 2004 年初，管理层明确意识到股东人数将在 2004 年内突破 500 人上限，不得不开始接洽各大投资银行筹备上市。拉里和谢尔盖本能地厌恶这一套流程：他们反感投行的巨额承销佣金，更对华尔街习以为常的“首日暴涨”（IPO pop）深恶痛绝。所谓的“首日暴涨”表面上看似喜庆，实则是投行包装出来的语言游戏。对于公司和老股东来说，暴涨恰恰证明发行定价被严重压低了——公司多年苦心经营积攒的内在价值，在上市当天被无端让渡给了有特权从投行手中拿到原始配额的核心机构客户，让他们在首日无风险赚取暴利。为了打破这种利益绑架，Google 在接触投行期间，结识了来自旧金山精品投行 WR Hambrecht + Co 的比尔·汉布雷希特（Bill Hambrecht）。不同于纽约那些盘根错节的华尔街大行，汉布雷希特向拉里和谢尔盖推荐了一种非传统的上市定价机制——**荷兰式拍卖（Dutch Auction）**。这种逆向拍卖机制通过先报出高价并逐步降价，由市场竞价确定一个最终能将拟发行股票全部清仓的“出清价格”（clearing price），旨在实现真正透明、高效的价格发现，将主动权从传统投行手中彻底夺回。 ### [03:06:01](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11161s) 荷式拍卖的理想主义与双重股权结构的诞生荷兰式拍卖（Dutch Auction）机制对于拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）来说具有近乎无法抗拒的吸引力。只要全额发售规模在某个竞标价格上全部被认购，定价便水到渠成。在两人看来，这简直太完美了：“这完全就是我们日常的业务逻辑，也是我们最擅长做的事，太令人愉悦了！”埃里克·施密特（Eric Schmidt）后来还提到过一个流传甚广的温情故事：Google 收到了一位老太太的来信，她听说 Google 即将上市，非常渴望能够买到一点股票。创始人希望让普通零售散户也能拥有平等的认购机会，这种情怀深深打动了他们。从理论上讲，传统投资银行家在进行人工询价圈购时，打电话逐个询问大客户“在这个价格你们买不买？要买多少”，本质上也是在以一种松散、人为的方式运行这套撮合算法。然而，除了定价机制和 IPO 发行方式，拉里和谢尔盖还有另一层更深重的焦虑——失去控制权。作为未上市的私有公司时，员工所持有的期权和股权相对闭环锁定；可一旦走向公开市场，员工和早期股东开始变现套现，大量外部公众股东进场，两位创始人很有可能在稀释中失去对公司的绝对控制。一直以来，Google 都坚决否认自己是一家“媒体公司”，但此时他们却将目光投向了传统传媒巨头。新闻出版业历史上早就面临过类似难题：传统报业集团（如《纽约时报》公司 The New York Times Company、道琼斯 Dow Jones 等）为了保护新闻采编的独立性与家族传承，既要吸收公众资本，又绝不允许商业利益干扰编辑立场，因而采用了“双重股权结构”（Dual-Class Share Structure）。在这种架构下，创始人家族持有具备超级投票权（Super-voting shares）的股票，即便其经济权益（所有权比例）被稀释，依然能绝对掌控公司的投票表决权。拉里和谢尔盖随即决定将这套设计搬到 Google 身上。如今，双重股权结构在科技行业已司空见惯——Meta（Facebook）、阿里巴巴、Shopify、Spotify、Coinbase、Airbnb、Zoom、Datadog 等无一例外；后来的 Snap（Snapchat）甚至更进一步，在 IPO 时发售的公众股票完全没有任何投票权，投资者买入后就如同持有绿湾包装工队（Green Bay Packers）的球迷股份一样纯属象征性参与。但在当年，Google 是第一家敢于在公开市场上推行双重股权结构的科技公司。 ```mermaid flowchart TD accTitle: Google 双重股权架构逻辑 accDescr: 展示 Google 借鉴传统报业的双重股权设计，以确保创始人掌控权并隔离公开市场干预。 subgraph CapitalStructure["股权与控制权分离架构"] direction TB ClassB["B 类超级投票股 (创始人与核心高管持有)"] ClassA["A 类普通股 (面向公众市场发售)"] end subgraph Protection["设立初衷与目的"] Control["锁定公司战略绝对控制权"] Insulation["防御公开市场短期利益干预"] Legacy["效仿纽时等传统报业家族采编独立模式"] end ClassB --> Control ClassB --> Insulation ClassA -. 提供资本流通.-> Legacy ``` - **超级投票权的历史突破**：Google 打破了硅谷科技股“同股同权”的传统惯例，成为第一家将媒体家族企业的双重股权治理模式成功引入公开资本市场的科技巨头。 - **行业连锁反应**：自 Google 2004 年敲钟后，几乎所有主流大型科技企业的 IPO 都将双重或多重股权结构作为标准标配。 --- ### [03:09:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11340s) 荷式拍卖的现实受挫与上市后的价值爆发虽然双重股权结构大获全胜，但荷式拍卖在实际执行中却远谈不上顺利。为了这次拍卖，Google 的软件工程师们甚至亲自参与编写了拍卖系统软件，与承销投行展开了一场奇特的联合运营。在询价初期，系统设定的发行指导价格区间高达每股 108 美元至 135 美元；然而到了最终撮合订单、完成清算的时刻，实际发行价却不得不一路腰斩下调至每股 85 美元。以每股 85 美元的价格计算，Google 在 IPO 中募集了 17 亿美元，上市当天的市值达到了 230 亿美元。单从账面上看，这在当时已经是一个天文数字般的估值倍数，很多人甚至认为 85 美元已经把公司的内在价值完全挤干，最初 108 至 135 美元的区间不过是创始人的盲目自大，“既然相信市场机制，那 85 美元应该就是它的公允价值了”。事实却给了这种判断一记响亮的耳光。上市交易首日收盘，Google 股价便蹿升至 100 美元，首日涨幅达 18%；仅仅过了 16 个月（到次年年底），其股价更是较发行价暴涨了接近 5 倍。回顾 Google 的上市历程： 1. **双重股权结构**：被奉为天才神作，此后各家科技巨头争相照搬； 2. **荷式拍卖**：被证明是一场充满争议的实验，在此之后几乎再也没有大型科技企业敢在 IPO 中重蹈覆辙。股价在短短几个月内翻倍，固然为 Google 带来了极具轰动效应的公关声量，让公众对其敬仰倍增，但从 IPO 发行机制本身的目标来看，它彻底失败了——Google 并没有实现精准定价，反而将巨额的账面资金白白“留在了桌面上”。 - **发行数据盘点**：指导发行区间为 108 至 135 美元/股；最终清算发行价为 85 美元/股；融资额 17 亿美元；上市首日市值 230 亿美元。 - **次级市场表现**：挂牌首日即上涨 18% 收于 100 美元；上市后仅数月内实现翻倍，并在 16 个月内录得近 5 倍涨幅。 --- ### [03:11:02](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11462s) 两万亿美元巨头的估值之谜：当下 Alphabet 的市盈率折扣拉长历史的焦距来看，IPO 阶段流失的那点资金不过是宏大叙事中的一个微小注脚。在节目录制时，Google（Alphabet）的市值已经膨胀到了约 2.1 万亿美元，较其 IPO 时的市值暴涨了整整 100 倍左右——这还不算多年来股息分红如果全额再投资所带来的复合增益。然而，眼下 Google 的市场估值却呈现出一幅极其诡异的图景。节目主持人在对谈中揭示，Alphabet 当前的市盈率（P/E Ratio）仅在 20 倍上下徘徊，营收市销率（P/S）约为 6 倍，几乎处于历史最低位。与同处第一梯队的科技巨头横向对比，这种折价显得触目惊心： - **亚马逊（Amazon）**：P/E 约为 35 倍； - **微软（Microsoft）**：P/E 约为 37 倍； - **英伟达（Nvidia）**：P/E 约为 46 倍； - **苹果（Apple）**：P/E 约为 30 倍； - **Meta**：P/E 约为 27 倍； - **Alphabet（Google）**：P/E 仅约 20 倍。更耐人寻味的是，Alphabet 并非业绩停滞——除了势头凶猛的英伟达之外，Alphabet 的营收增速丝毫不逊色于列表中的任何一家公司，甚至跑得更快，且利润率始终维持在高位。显然，资本市场（“市场先生”）对 Google 未来的预期远比对其他巨头悲观得多。讽刺的是，恰恰是 Google 奠定了现代 AI 的基础、率先发表了划时代的 Transformer 论文，然而眼下投资者却深切担忧 AI 会动摇其赖以生存的搜索护城河。 --- ### [03:12:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11570s) 愚人节的重磅炸弹：Gmail 的诞生前奏与保罗·布赫海特在正式开启 Google 历史第二部之前，时间指针被拨回到了 2004 年 4 月。就在 Google 正式提交 S-1 招股说明书的同一个月，他们选择在 4 月 1 日愚人节当天发布了一款震惊全网的新产品。考虑到 Google 一贯热衷于在愚人节发布荒诞恶搞公告的传统，几乎所有人都以为这是一场彻头彻尾的玩笑——因为这款名为 **Gmail** 的免费网页邮箱所承诺的参数听起来实在是好得令人难以置信：**每个用户享有整整 1GB 的免费存储空间**。在当时，主流网络邮箱雅虎邮箱（Yahoo Mail）和微软 Hotmail 为普通用户提供的免费存储额度仅仅只有 2MB 左右，Gmail 一出手便将容量上限拉升了数百倍。不仅如此，Gmail 还将 Google 强大的底层搜索技术无缝内置在邮件系统内，且整套程序百分之百运行在浏览器网页端，速度快到不可思议，彻底颠覆了以往笨拙的 Web 体验。然而从商业逻辑来看，Google 为何要不惜工本地推出这样一款产品？ > “如果能给用户一个极具说服力的理由，让他们始终保持在‘已登录 Google’的状态呢？如果我们能把用户生活中的更多触点与 Google 绑定，让它们成为进入 Google 搜索以及人类有史以来最赚钱的搜索广告业务的超级入口，那该是怎样的一番图景？” Gmail 的幕后缔造者正是硅谷传奇工程师保罗·布赫海特（Paul Buchheit）。他后来成为了 Y Combinator 的核心合伙人，并与布雷特·泰勒（Bret Taylor）共同创办了 FriendFeed，更重要的是，他也是 Google 著名的企业座右铭“不作恶”（Don't be evil）的最早提出者。早在 2001 年，布赫海特便利用公司的“20% 自由时间机制”开始秘密研发这一项目。Gmail 最初的原型粗糙而纯粹：布赫海特只是把工程师内部的 Unix 邮件目录视作一张普通的网页，然后在上面挂了一个 Google 搜索框，让人能够像用 Google 搜网页一样全文检索自己的个人邮件。正是这样一个极度极简的搜索框，悄然孕育出了改写整个互联网应用生态的 Gmail。 ### [03:15:44](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11744s) Gmail 原型意外催生 AdSense 与“城堡”的落成在 Gmail 的早期开发过程中，搜索其实是它的第一个核心功能，其余所有功能都是后来才逐步添加的。当时保罗·布赫海特（Paul Buchheit）一边摆弄这个原型，一边脑海中蹦出了一个灵感：既然 Google 的核心业务是索引自然搜索结果并展示相关广告，那么除了索引和搜索用户邮箱文件夹里的邮件之外，为什么不直接从广告数据库里调取广告，并展示在邮件周围，看看广告与邮件内容的相关性匹配度如何？保罗在公司内部展示了这个原型，拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）看到后瞬间被点醒，立刻反问：“等等，这套逻辑在普通网页上也能行得通吗？”就这样，AdSense 的灵感火花实际上是在 Gmail 的原型设计过程中被点燃的。两位主持人赞叹这个故事的绝妙反转，并感谢保罗·布赫海特向播客分享了这段内幕。随着搜索业务与商业化拼图的闭环完成，第一集关于 Google 搜索帝国奠基的叙事正式画上句号。大卫·罗森塔尔（David Rosenthal）形象地总结道：这一集讲述的是如何“建造城堡”，而下一集将展示他们如何在城堡周围建起整座城市、州乃至一个超级平台或宽阔的护城河。 ### [03:17:29](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=11849s) 商业秘籍（Playbook）：Google 为什么能成功 ```mermaid flowchart TD accTitle: Google 成功的核心要素体系 accDescr: 展示 Google 早期取得非凡成功所依赖的算法、广告拍卖机制、基础设施、人才与飞轮效应。 subgraph CoreEngine["核心产品与商业引擎"] A["顶尖算法与极致体验 （PageRank + 纯粹自然搜索）"] B["完美执行的广告模型 （竞价 + CPC + 质量得分 CTR）"] end subgraph ScaleAndMoat["规模壁垒与飞轮"] C["底层技术与基建优势 （第一性原理与自研分布式架构）"] D["双边市场流动性飞轮 （用户查询越多 -> 广告竞价越激烈 -> 变现效率越高）"] E["自增强数据网络效应 （海量搜索数据反哺算法）"] end subgraph OrgCulture["组织与文化基石"] F["顶级人才密度 （借互联网泡沫破裂吸纳顶尖精英）"] G["极致规模化思维与远大愿景 （整合全球信息）"] end CoreEngine --> ScaleAndMoat OrgCulture --> CoreEngine ScaleAndMoat --> ScaleAndMoat ``` 在复盘 Google 早期崛起的商业秘籍时，本·吉尔伯特（Ben Gilbert）梳理了 Google 取得空前成功的核心驱动要素： - **顶尖的原创算法见解与极致体验**：拥有当时全网最好的自然相关性排序，配合飞快、简洁的界面设计，并且极度克制地杜绝付费收录（Paid Inclusion），长期捍卫了自然搜索的纯洁性与用户信任。 - **搜索广告商业模式的完美执行**：将竞价拍卖、按点击付费（CPC）以及引入点击率（CTR）作为相关性权重的机制融为一体。这不仅促使广告主竞相提升广告相关度，还从数学上最大化了 Google 展示广告的期望收益（Expected Value），构建了一个天衣无缝的良性生态。 - **独创的基础设施优势**：坚持从第一性原理思考问题，在廉价硬件上自研革命性的分布式系统与技术栈。 - **无可匹敌的人才密度**：恰逢 2000 年互联网泡沫破裂，市场上大量世界级技术人才任其挑选。正如保罗·布赫海特所言，身处其中时甚至没有意识到身边聚集的全部是行业最顶尖的头脑。 - **敢想敢干的学术与规模化文化**：大学校园般纯粹、不受世俗经验束缚的雄心，结合顶尖博士的智识力量，加上“必须从第一天就考虑按无限规模扩展”的工程信仰，使团队极具进行高风险、数十亿美元级非线性押注的底气。 - **自增强的数据网络效应**：外界常神话 PageRank 算法本身，却忽略了算法必须依赖源源不断的真实搜索数据喂养才能自我迭代与强化。 - **历久弥新的崇高使命**：“整合全球信息，供大众使用，使人人受益”（Organize the world's information）。既不空泛也不狭隘，外表充满利他色彩，底层却支撑着人类商业史上赚钱效率最高的一门生意。 ### [03:20:26](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12026s) 超越成本递减的双边流动性飞轮大卫特别强调了规模经济在 Google 商业模式中的特殊形态：传统的规模经济主要体现在随着规模扩大带来边际成本的下降；但 Google 的搜索广告平台展现出一种罕见的“双边流动性飞轮”（Liquidity Flywheel）。当用户和搜索查询量持续扩张时，吸引了越来越多的广告主入场竞价；长尾关键词的竞争更加充分，导致每个展示位的变现价值大幅提升。这意味着规模的扩大不仅在摊薄单位基础设施成本，而且还在直接拉高单位流量的收入水平。一旦 Google 洞察到“用户与广告主双边池子越庞大，商业系统整体收益率越高”这一逻辑，他们便有极其充分的经济动机去支付几乎任何代价，以换取更多的流量分发入口与市场规模。 ### [03:21:18](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12078s) 历史不会重复，但押着相同的韵脚：AI 浪潮与时代窗口大卫分享了在策划本期节目时的深层感悟。最初制作 Google 专题是为了理解当前火热的人工智能（AI）浪潮，原本以为讲早期的搜索历史只是“为了吃下营养蔬菜以理解 Transformer 的前奏”。但随着研究的深入，他惊奇地发现：1996 年至 2002 年间搜索产业的发展轨迹，与 2021 年 ChatGPT 诞生以来的 AI 演变竟然完全重合： - **商业化与价值链的投射**：回顾 GoTo/Overture 当年在 TED 大会上推出竞价排名广告时引发的道德争议与舆论哗然，大卫打趣道：如今如果有人做一个大语言模型（LLM）对话机器人，其回答的内容完全取决于哪家赞助商出价最高，公众同样会感到被冒犯，但这恰恰是技术探索商业化必然会碰撞出的试验形态。 - **严丝合缝的时间窗口**：Google 诞生于一个不可复制的黄金时代。如果拉里和谢尔盖早相遇几年（比如 1994 年），当时互联网规模还很小，手工维护的分类目录（如 Yahoo）便足以解决问题，市场不需要复杂的算法；而如果晚相遇几年，互联网爆炸产生的数据量则会让缺乏启动技术与资本支撑的小团队寸步难行。Google 恰好卡在了那个技术变革与内容爆炸交汇的极窄时间窗口之中。 ### [03:24:27](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12267s) 七大力量（7 Powers）剖析与逆向定位播客随后切入战略分析环节，采用汉密尔顿·赫尔默（Hamilton Helmer）在《7 Powers》（七大力量）中定义的框架，探讨 Google 能够持续维持超常利润率的核心壁垒。这七大力量包括：逆向定位（Counter-positioning）、规模经济（Scale Economies）、转换成本（Switching Costs）、网络效应（Network Economies）、流程权力（Process Power）、品牌力量（Branding）以及独占资源（Cornered Resource）。针对第一项“逆向定位”，两位主持人展开了细致讨论。最初有人认为 Google 与老牌玩家之间可能没有强烈的逆向定位，但实际上，面对当时的雅虎（Yahoo）以及 Excite 等老一代搜索引擎时，逆向定位非常显著。当时的门户网站依赖将用户留在自己的站内页面上来赚取横幅广告展示费，而 Google 的核心理念却是“提供最好、最快的搜索结果，让用户尽快离开 Google 跳转到目标网页”。这种对传统门户商业逻辑的颠覆，让当时的巨头因担心破坏自身利润丰厚的门户展示广告模式，而根本无法跟进 Google 纯粹的搜索产品形态。 ### [03:25:15](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12315s) 7 Powers 分析：逆向定位与门户网站的困境在梳理汉密尔顿·赫尔默（Hamilton Helmer）提出的“7 Powers”商业护城河模型时，主播们探讨了谷歌早期的竞争壁垒。首先是“逆向定位”（Counter-Positioning）。有人可能会提出，早期搜索引擎 Excite 等希望用户停留在自己的页面上，而 Google 却反其道而行之，希望用户尽快找到答案并离开页面。但本和戴维认为这并不能算作典型的逆向定位，因为当时搜索还是一个极小的新兴赛道，那些竞争对手只是微不足道的小角色，Google 的胜出纯粹是因为产品“更好”，而仅仅做得更好并不构成逆向定位。逆向定位真正成立的对象是当年的门户巨头（Portals），尤其是雅虎（Yahoo）。尽管门户网站在某个时间节点意识到搜索极其关键，但由于它们商业模式的核心是依赖门户网站上的展示广告收入，根本无法壮士断腕转型去全力发展“把用户尽快送出网站”的纯搜索体验。雅虎虽然是由两位斯坦福大学电子工程博士创办的，却把自己重构为了一家媒体公司。当雅虎后来终于试图迎头赶上时，它先后收购了 Overture 和 Inktomy，并耗费数年时间试图将两者拼凑成一个抗衡 Google 的系统——这就是雅虎著名的折戟之作“巴拿马项目”（Project Panama）。播客中还顺带揭晓了一枚《Acquired》复活节彩蛋：当年扬·库姆（Jan Koum）与布莱恩·艾克顿（Brian Acton）正是因为在雅虎参与该项目相识，后来因极度沮丧而离职，最终携手创立了 WhatsApp。 ```mermaid flowchart TD accTitle: 门户网站在搜索转型中的逆向定位困境 accDescr: 展示雅虎等门户因依赖展示广告而无法有效转向纯搜索，最终导致巴拿马项目失败的逻辑链路。 subgraph YahooDilemma ["雅虎的门户模式困境"] A["依赖门户展示广告收入"] --> B["商业模式依赖用户站内停留时间"] B --> C["无法全力转向纯搜索体验"] end subgraph LateResponse ["迟缓的防守举措"] C --> D["收购 Overture & Inktomi"] D --> E["推行'巴拿马项目' (Project Panama)"] E --> F["整合受挫，难以匹敌 Google 竞价与技术"] end ``` ### [03:26:28](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12388s) 规模经济与竞价拍卖的“超级规模效应” 谈到“规模经济”（Scale Economies），Google 展现出了一种远超传统定义的强大势能。传统的规模经济定义（如 Netflix）是指将某项内容或基础设施的固定采购成本分摊到海量用户身上；Google 确实能够将基础设施、硬件、软件以及获客成本分摊给数十亿用户，但其真正恐怖之处在于：**随着规模扩大，Google 的单用户平均收入（ARPU）反而会水涨船高**。主播们将这种现象戏称为一种崭新的护城河维度——“超级规模经济”（Super-Scale Economies）。这一奇迹的根源在于 Google 的广告采用了微观“实时竞价拍卖”（Auction-Based）机制。竞价市场的流动性越高、竞标者越多，最终出清的价格就被推得越高，Google 从而能够从每一次搜索拍卖中最大化其抽成比例。这与佳士得（Christie's）或苏富比（Sotheby's）拍卖行的逻辑相似：竞拍买家越多，拍品的成交价就越高。这也交织着“网络效应”（Network Economies）：搜索查询量越大，吸引的广告商就越多，进而给广告生态带来更高的变现效率；但对广告商彼此之间而言，由于加剧了竞价压力，反而呈现出一种独特的“负网络效应”。 - **传统规模经济**：在固定研发、服务器和数据中心基建成本上摊薄单用户支出。 - **动态竞价收益**：查询量暴增吸引更多广告主进场，微观拍卖出清价格随之攀升，实现规模越大、单次变现效率越高的良性循环。 ### [03:28:50](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12530s) 转换成本、品牌声誉与核心特质（Quintessence）在 7 Powers 的其余维度中，主播们指出在 2004 年前后，Google 的“转换成本”（Switching Costs）几乎为零——当时用户不需要登录账户，没有个性化推荐，用户随时可以跳到其他引擎，只是当时其他竞品体验太差。关于“品牌力量”（Branding），Google 并没有像爱马仕（Hermès）那样能让广告主因为品牌溢价而支付更高费用的定价权，广告主在广告投放上是极其理性的经济人；但 Google 拥有无与伦比的“雇主品牌”（Employment Brand），在 2002 至 2006 年间，Google 凝聚了极高的极客崇拜，成为全美聪明才智最向往的圣地。至于“独占资源”（Cornered Resource）和“流程权力”（Process Power），在当时也并未形成决定性的壁垒。在提炼 Google 的“企业特质”（Quintessence）时，主播们高度赞赏了 Google 能够尽早洞察到“规模收益递增”逻辑，并敢于长线思维进行极其激进的资本开支投入。此外，戴维补充了史蒂文·利维（Steven Levy）在《In the Plex》一书中记载的一个经典视角：2007 年 6 月 8 日，刚刚离开 Google 的前产品经理贾斯汀·罗森斯坦（Justin Rosenstein）在一封发给老同事的邮件中写道： > “我写这封信是为了传播好消息……Facebook 真的就是**那家公司**（That Company）。哪家公司？就是那种每隔很久才会出现一次、正在改变世界风口浪尖上的公司——就像当年的微软，昨日的 Google。在那样规模尚小、每个员工都能产生巨大影响力的阶段，哪怕别人早已提醒过你它正驶向应许之地，若你没能赶快登上这趟大篷车，三年后你一定会悔恨不已。” 在科技商业史的坐标轴上，先有微软，后有 Google，再有 Facebook，这种承载时代范式转移的“That Company”在整个商业史上都极其罕见。 ### [03:31:59](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12719s) Carveouts：本的剧集推荐进入节目的 Carveouts（个人心头好分享）环节，由于上一期专访史蒂夫·鲍尔默（Steve Ballmer）时略过了该环节，本这次一口气分享了过去两个月中看过的三部高分剧集： 1. **《排练》（The Rehearsal）第二季（HBO）**：由内森·菲尔德（Nathan Fielder）主创。本对内森超乎常人的喜剧投入度（Commitment to the bit）惊叹不已。十年前在《救援老老店》（Nathan for You）中，内森就曾为了给咖啡店拉客，豪掷制作方数百万美元将店面恶搞改造成“傻瓜星巴克”（Dumb Starbucks）；而在《排练》第二季中，内森甚至耗费数年生命，以“减少全球空难事故”为由，深入研究飞行员沟通机制，搭建了极度逼真繁复的机舱与生活布景，雇佣大批群演模拟各种极端体验，试图帮助飞行员消除沟通障碍。其荒诞与深沉的执行力令人叹为观止。 2. **《邻里同谋》（Your Friends and Neighbors，Apple TV+）**：由乔恩·哈姆（Jon Hamm）主演。运镜极其考究，质感类似《继承之战》（Succession）那种围绕上流社会虚构富豪展开的生活图景，但在情节发展中加入了极具意料之外的反转，轻松且极其好看。 3. **《安多》（Andor）第二季（Disney+）**：本认为前三到四集节奏略显缓慢，水准略逊于惊艳的第一季，但后八集的发力堪称星战正史宇宙（Star Wars Canon）中顶级的影视杰作。戴维笑称本不仅是他最好的朋友，更是每次都能精准安利神作的“剧评专家”。 ### [03:35:00](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=12900s) Carve Outs 推荐：Gamecraft 第三季与主机选购大纠结在节目的 Carve Outs 环节中，Ben 和 David 开启了轻松的日常话题互动。Ben 开玩笑说自己就是网络上的“电视剧推荐官”，哪怕明知道 David 家里有两个年幼的孩子、根本没有时间看电视，他依然会乐此不疲地分享好剧，因为这些推荐不仅是给搭档的，更是给广大听众准备的。轮到 David 分享时，他带来了两个精选内容。首先是一个标准的播客推荐：《Gamecraft》第三季。多年以前，Acquired 曾与该节目的两位主创 Mitch Lasky 和 Blake Robbins 做过一次联动对谈。David 称赞他们对游戏行业的探讨极为专业且深度投入，第三季制作得非常出色，只要是对游戏及游戏产业感兴趣的听众，《Gamecraft》绝对是不容错过的行业标杆之作。随后，David 抛出了一个正在发生中的“即时纠结（real-time dilemma）”： - 他原先满心期待 Nintendo Switch 2 的发售，甚至早早在旧金山新开业的任天堂官方旗舰店预约了进店购买名额，幻想着未来等女儿们长大一些可以一起玩游戏； - 然而在 YouTube 上反复研究评测视频的过程中，他被海量关于 Steam Deck 的评测内容深度吸引； - 现在他陷入了选择困难症：究竟是按原计划入手任天堂 Switch 2，还是彻底转向购买 Valve 的 Steam Deck？ - 考虑到节目制作和剪辑发布周期，当这期节目播出时他肯定已经做出了决定，因此他承诺会在下一期《Google Part II》中向听众揭晓最终选择，并提前在 Twitter 上晒图公布。 --- ### [03:37:01](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=13021s) 纽约现场活动预告与本季赞助商鸣谢两位主播借此机会向听众发出邀请，欢迎大家前往纽约参加即将于无线电城音乐厅（Radio City Music Hall）举办的大型线下活动（购票与详情见 acquired.fm/nyc），体验 Acquired 与摩根大通支付（J.P. Morgan Payments）共同筹划的震撼现场。随后，他们对本季度的核心赞助商表达了诚挚感谢： - **J.P. Morgan Payments**：为各种规模的企业提供值得信赖且稳健可靠的底层支付基础设施； - **Anthropic（Claude）**：强大的 AI 助手。Ben 和 David 透露，在准备这期长篇巨制的过程中，他们重度依赖 Claude 协助处理信息，直言它彻底改变了 Acquired 团队的幕后工作流； - **Statsig**：面向产品开发团队的顶级实验分析与特性管理平台； - **Vercel**：提供全方位现代化 Web 开发与部署的完整平台。 --- ### [03:37:51](https://www.youtube.com/watch?v=v1nkN0sM2wc&t=13071s) 致谢重磅嘉宾名单与尾声在每一期超级科技巨头的专题中，大量深度访谈是支撑节目品质的基石。两位主播首先特别感谢了 Worldly Partners 的 Arvin Navaratnam，赞扬他为 Google 撰写的多维度深度商业研究报告（相关链接已附在节目 Show Notes 中）。紧接着，David 宣读了协助本次节目筹备的受访者全明星名单： ```mermaid flowchart TD accTitle: 受访行业领袖与技术专家分布 accDescr: 展示为本期 Google 专题节目接受采访并提供关键洞察的技术元老与高管名单 subgraph Google["Google 早期元老与核心高管"] G1["Craig Silverstein（Google 1号员工）"] G2["Omid Kordestani & Alan Eustace"] G3["Jeff Dean & Urs Hölzle（技术骨干）"] G4["Paul Buchheit & Jen Fitzpatrick"] G5["Anna Patterson & Clay Bavor & Nick Fox"] end subgraph Ecosystem["外部创业者、投资人与行业先驱"] E1["Bill Gross（Idealab / GoTo.com 创始人）"] E2["Bret Taylor（前科技高管 / 创业家）"] E3["Danny Sullivan（搜索引擎观察家 / Google）"] E4["Wesley Chan & Esar Lipkovitz"] end ``` 两位主播笑称，这一长串受访名单简直是一支可以随时颠覆行业的“科技创业梦之队（Dream Team）”。在调研进入尾声时，他们甚至开玩笑说：“我们在做访谈调研时占用了这些大佬太多宝贵时间，恐怕无形中让整个经济体损失了数十亿美元的潜在市值。” 在由衷感谢所有匿名及公开受访者的倾囊相助后，Ben 和 David 与听众约定下期再见，节目在标志性的片尾旋律 “Who got the truth?” 中圆满落下第一部的帷幕。 ## Video description We tell the story of the single greatest business ever created: Google search. From its origins as a Stanford research project called BackRub, Google became the front door to the internet. Today it’s an essential service for over half the world, and one that generates more profit than ANY other US company — more than Apple, Microsoft, or Berkshire Hathaway. But it wasn’t always so obvious. When Larry and Sergey began working on BackRub in 1996, search was a backwater industry in silicon valley. Existing search companies were eking out a living as vendors to the then-dominant “portals” like AOL and Yahoo. Google’s come-from-behind success was the result of three massive step-function leaps forward in algorithms, infrastructure and business model… some invented by Google and some borrowed (and perfected!) by them. Today, things are not so obvious once again for Google. Despite earning more profits than all of its big tech peers, its stock trades at significantly lower multiples — a $1 trillion or more discount to Apple, Microsoft and Nvidia. Investors are concerned that AI will render Google’s beautiful business model obsolete, even though Google also basically invented modern AI and continues to lead on many dimensions. This episode begins a multi-part series where we dive into the full history that led us to this point. Tune in and enjoy! *Links:* - BackRub recreation: https://backrubsanford.neocities.org - Original Google logo: https://en.wikipedia.org/wiki/Google_logo#/media/File:First-google-logo.gif - Jeff Dean’s resume: https://drive.google.com/file/d/1WM65kvbrFMTtrwnkct4XzpCC8hqQe5ZX/view?pli=1 - Worldly Partners’ Multi-Decade Alphabet Study: https://worldlypartners.com/businesshistory - Episode sources: https://docs.google.com/document/d/13C0I8Bg4XLlBDtRBzPwimhlOeojpENhY52YSANrdhTM/edit?usp=sharing *Carve Outs:* - The Rehearsal with Nathan Fielder (Season 2): https://www.hbo.com/the-rehearsal/season-2 - Your Friends and Neighbors: https://tv.apple.com/us/show/your-friends--neighbors/umc.cmc.74o37kzay0yuuub8iumddjsg - Andor Season 2: https://ondisneyplus.disney.com/show/andor - Gamecraft Season 3: https://gamecraftpod.com - Steam Deck vs Switch 2 dilemma: https://store.steampowered.com/steamdeck/ https://www.nintendo.com/us/gaming-systems/switch-2 *More Acquired:* - Get email updates https://www.acquired.fm/email and vote on future episodes! - Join the Slack http://acquired.fm/slack - Check out the latest swag in the ACQ Merch Store https://www.acquired.fm/store! _Note: Acquired hosts and guests may hold assets discussed in this episode. This podcast is not investment advice, and is intended for informational and entertainment purposes only. You should do your own research and make your own independent decisions when considering any financial transactions._ [Source](https://www.youtube.com/watch?v=v1nkN0sM2wc)

Source quotes

[5:03] Larry Page: Probably when I was 12, I knew I was going to start a company eventually. I wanted to make the world better, and in order to do that, you need to do more than just invent things... you need to use business and entrepreneurship to make these things real. [40:14] Andy Bechtolsheim: Great, I'm in. $100,000, [81:40] Michael Moritz: Doug, we've never paid so much for so little. [105:18] to organize the world's information and make it universally accessible and useful. [105:18] We believe that the most effective and ultimately the most profitable way to accomplish our mission is to put the needs of our users first. [111:13] John Wanamaker: Half the money I spend on advertising is wasted. The problem is I just don't know which half. [123:50] Click-through rates are a proxy for relevance. [141:30] Sergey Brin: We could have gone bankrupt. [141:30] Terry Semel: You don't have a business, do you? [141:30] Larry Page: We should be able to monetize the pages. If not, we deserve to go out of business. [148:07] Inktomi: They'll learn over time that Google takes your users. It doesn't help you build your property. [209:56] Justin Rosenstein: Facebook really is that company. [211:59] Ben: some of the best Star Wars canon that exists. [217:17] David: sapping billions of dollars of market cap out of the economy

Original transcript

[0:00] All right, David. Last episode we are [0:02] doing in our studios before Radio City. [0:04] Oh, that's right. How do you feel? Uh, [0:06] we're about to go from like the stage of [0:09] one, very, very small audience of one to [0:12] the very, very big stage where if we [0:14] make a mistake, no one will notice. [0:15] Yeah. And we just re-record it and it's [0:17] like it never happened. We should try [0:18] that at Radio City. Just be like, "Ah, [0:20] strike that. All right, let's take that [0:21] again." Yeah. Hey, this is authentically [0:22] acquired, you guys. This is how we do [0:23] it. You're getting a look at the inside. [0:25] Probably not, though. All right, let's [0:28] do it. Let's do it. Who got the truth? [0:32] Is it you? Is it you? Is it you? Who got [0:34] the truth? Now, [0:37] is it you? Is it you? Is it you? Sit me [0:40] down. Say it straight. Another story on [0:43] the way. Who got the truth? Welcome to [0:46] the summer 2025 season of Acquired, the [0:49] podcast about great companies and the [0:51] stories and playbooks behind them. I'm [0:53] Ben Gilbert. I'm David Rosenthal. And we [0:55] are your hosts. Artificial intelligence [0:58] is the story of our time. It is [1:01] definitively the next trillion dollar [1:03] technology wave after PCs, the internet, [1:05] and mobile. And to understand AI, you [1:09] have to understand the company most [1:11] responsible for its technical foundation [1:13] and the wave that came before it, [1:15] Google. This episode begins our [1:18] multi-art Google saga. Finally, as I'm [1:21] sure many of you out there are saying [1:22] right now, Google has been the front [1:25] door to the entire internet for 25 years [1:27] now, a quarter century. But it wasn't [1:30] always this way. No, it was not. Back in [1:32] 1998 when Google was founded, there were [1:35] a dozen other search engines that [1:38] already existed. And there were a [1:40] variety of different business models, [1:42] most of which were not very interesting. [1:43] Yeah. None of which were very [1:45] interesting. Yes. So today we will try [1:48] to answer the question why did Google [1:50] work and once it did how did it go from [1:53] clever technology and nice product to [1:56] the single greatest business of all [1:58] time. I'm not being facicious listeners [2:01] Google and I should say Alphabet today [2:04] generates more net income or profit than [2:07] any other US company more than Apple, [2:10] Microsoft, Exxon Mobile, JP Morgan [2:13] Chase, Berkshire Hathaway. This is a [2:16] cash gusher. It is a super high gross [2:20] margin, b in a giant market and c [2:24] according to the US government as of [2:25] today they are a monopoly in that market [2:28] with 90% market share. That is three [2:31] enormous numbers multiplied together to [2:35] create that most profitable company in [2:37] the US stat that I threw out earlier. [2:40] Well, I'm glad we don't need to get to [2:41] uh the government and all that until [2:43] much much later in our series. But yeah, [2:46] this is the creation of the most [2:48] beautiful business of all time. And [2:50] Google's market position has been [2:51] seemingly unassalable, at least until [2:54] really this year, until the AI wars [2:56] really heated up. So, of course, it was [2:58] that clean user experience that everyone [3:00] talks about with just a search box on [3:03] the homepage and the focus on the users [3:05] and the highquality fast search that [3:07] spread virally through word of mouth. [3:09] But good product is far from the only [3:11] reason that Google became dominant. So [3:13] today we'll tell the story of why while [3:15] Google was nowhere near the first search [3:18] engine, it was the last. Well, you know, [3:21] uh Microsoft may take a little issue [3:23] with that with Bing, but but only a [3:26] little issue. Little few percentage [3:28] points of issue. Few percentage points [3:29] of issue. Yeah. Well, listeners, if you [3:31] want to know every time an episode [3:32] drops, check out our email list. It is [3:34] also the only place where we will share [3:36] a hint at what our next episode will be. [3:38] Share corrections, updates, little [3:40] tidbits we learned from previous [3:41] episodes. That's acquired.fm/e. [3:44] After this episode, join the Slack to [3:46] talk about this with us and the whole [3:48] acquired community. acquired.fm/slack. [3:52] And if you want more acquired between [3:54] each episode, check out ACQ2, our [3:57] interview show where we talk with [3:58] founders and CEOs building businesses [4:00] and areas we've covered on the show. The [4:02] most recent was with Jesse Cole from the [4:04] Savannah Bananas. David, that was the [4:07] most fun I've ever had recording [4:08] basically any podcast episode. It was [4:10] bananas. It was awesome. Yes. Well, [4:12] before we dive in, we want to briefly [4:14] thank our presenting partner, JP Morgan [4:16] Payments. Yes. And as you all know, JP [4:18] Morgan is our partner for our big Radio [4:20] City show coming up on July 15th. We've [4:23] been hard at work preparing. It's going [4:25] to be an amazing night. The show is [4:27] basically sold out. I think there like [4:29] less than 100 tickets remaining right [4:30] now out of the 6,000 total. So, if you [4:33] haven't grabbed yours yet, go do it like [4:34] right now or they will all be gone. JP [4:37] Morgan will be showing off all their [4:38] latest and greatest payments tech there. [4:40] We'll have exclusive merch, a fan meetup [4:42] before the show, an afterparty, and an [4:44] encore event the next day at the New [4:46] York Stock Exchange. So, details on all [4:48] of that are in the acquired Slack. We're [4:50] so pumped. Yes, we cannot wait to see [4:52] you there. So, with that, this show is [4:54] not investment advice. David and I may [4:56] have investments in the companies we [4:57] discuss and this show is forformational [4:59] and entertainment purposes only. David [5:03] Rosenthal, when is the first time any [5:05] human being searched for anything ever? [5:10] Uh, I have no idea. That's a great [5:13] question. Not that far back. Where do we [5:14] start? Yeah. Well, you know what? We [5:16] should ask Google. But to do that, we [5:19] need to tell the story of Google. And [5:21] that story starts in March of 1973 in [5:24] Lancing, Michigan, where Larry Paige is [5:28] born as the second child and second son [5:30] to Carl and Gloria Page. And Larry, of [5:33] course, grows up in Lancing because his [5:35] dad, Carl Paige, Senior, is a professor [5:38] of computer science in nearby East [5:40] Lancing at Michigan State University. [5:43] Now, before my dad, who went to MSU and [5:45] is an MSU alum, gets too excited here, I [5:48] regret to inform him, my dad, and you, [5:51] Ben, that Carl got his PhD from [5:54] Michigan. I'm sorry about that. And, uh, [5:57] would send his son there as well. Yeah. [5:59] Yeah. Both of his sons. And [6:01] unfortunately, Larry's mom also went to [6:04] Michigan, also got a CS degree there, [6:07] and also teaches programming as a [6:09] programming instructor at MSU. Yeah, [6:11] pretty Michigan heavy. pretty Michigan [6:13] heavy, but uh pretty amazing childhood [6:16] in, you know, the early mid70s here for [6:19] Larry and his older brother. I mean, [6:21] maybe not unique. I'm sure there were a [6:24] few other households in America, in the [6:26] world that grew up with both of their [6:28] parents steeped in computers as computer [6:31] science professors, but really pretty [6:34] unique. Incredibly unique. Are you [6:36] kidding me? Larry Paige grew up with two [6:38] computer science academics as parents in [6:41] the 70s which would have meant that his [6:43] parents would have needed to start in [6:44] the ' 50s. Very very few households like [6:47] right at the same time as the PC era is [6:51] coming online and Microsoft and to just [6:54] have that be your air your daily [6:56] existence growing up as a kid like how [6:58] incredible is that amazing. So even more [7:02] so for future Google to come for the [7:05] 1979 to 1980 academic year when Larry is [7:08] six and seven years old. His dad does a [7:11] sbatical year at Stanford. So the whole [7:13] family goes out and lives in PaloAlto [7:16] and Stamford and early Silicon Valley [7:20] there makes a big impression on young [7:22] Larry. And then he continues to kind of [7:24] be influenced by this because I [7:26] mentioned his older brother Carl Jr. [7:28] who's 9 years older than him. Carl goes [7:31] to Michigan, majors in CS just like [7:33] Larry would, and then when he graduates, [7:35] he goes out to the West Coast, actually, [7:36] to the Pacific Northwest, and he works [7:38] fairly early at Microsoft and then [7:40] Mentor Graphics down in Oregon. Carl Jr. [7:43] would ultimately also come down to [7:45] Silicon Valley and uh play a little role [7:47] in this story as we will see in a little [7:49] bit. But back to Larry here. Part of the [7:52] reason I say all this, and I want to [7:55] include Sergey too in what I'm about to [7:56] say, even though we haven't introduced [7:58] him yet in the story, I think there's [8:00] this perception today that Larry and [8:02] Sergey were these like bumbling academic [8:04] guys who weren't really businessminded [8:07] and Google was a research project and [8:09] this all sort of happened by accident [8:11] that they built the best business of all [8:12] time. Like, absolutely freaking not. We [8:17] want to dispel that notion right now. [8:19] You know, one of the things we heard [8:20] over and over again talking to people in [8:22] research is how hugely ambitious the two [8:26] of them were, not just for the products [8:28] they were building, but for Google, for [8:30] the business. Larry's a different [8:32] generation and a very different [8:34] personality than Mark Zuckerberg, but [8:37] you should think about his ambition and [8:39] his desire to build a huge world [8:43] changing company at the same level as [8:45] him. Yes. To your point, Google did not [8:48] happen by accident. Or another [8:49] reasonable comparison, the generation [8:51] before Bill Gates. I think they publicly [8:54] Larry and Sergey don't get the same sort [8:57] of ethos that those two get, but the [8:59] same fire was there. Larry would say [9:02] later, this is a quote from him, [9:03] probably when I was 12, I knew I was [9:06] going to start a company eventually. I [9:07] wanted to make the world better, and in [9:09] order to do that, you need to do more [9:11] than just invent things. And he would [9:13] say another time later, you need to use [9:15] business and entrepreneurship to make [9:17] these things real. It's not enough just [9:19] to invent them. Yeah. And he's alluding [9:20] to two things there. One is companies [9:23] are the vehicles by which you bring [9:25] ideas to the masses. And two is in a [9:27] capitalist society, a company is the [9:30] vehicle that can accumulate profits, [9:32] which then you can reinvest to build [9:34] something of your great scale and [9:36] ambition. Yes. And Larry got this all [9:38] along and Sergey did too. Okay. Okay, so [9:41] as we all know, Larry goes to Michigan [9:43] undergrad. He graduates in 1995 and then [9:45] he goes off to Stanford to get his PhD [9:47] in computer science where has a fateful [9:50] meeting with his business partner, [9:53] friend, soulmate, Sergey Britain. And [9:56] isn't the way this all went down that [9:58] Larry was visiting and Sergey was [10:01] already in the program? I think Sergey [10:03] was leading some kind of tour to try to [10:05] sell Larry on joining the program. And [10:08] Larry is, even though he's like the new [10:10] guy there, he's like challenging Sergey [10:13] at every little corner. He's bringing [10:15] up, "Wouldn't a better system be this?" [10:17] And they're talking about cities and [10:19] transportation and civic design. And [10:21] they're almost sort of like bickering [10:23] back and forth in this verbal sparring [10:25] of who's smarter even though they just [10:26] met. That's sort of like the story that [10:28] I read and I don't know what did you and [10:29] I read probably six or seven books [10:31] between us on the history of Google. [10:32] Yeah. Many, many versions of this story [10:34] out there. As I was doing the research, [10:37] I talked to one of our good friends, [10:39] Anna Patterson, who was been in the [10:41] Google orbit for a very long time. She [10:43] was an early employee, left, started a [10:45] company, Google, reacquired, was a VP of [10:47] engineering there for a long time. She [10:49] told me another little bit of this story [10:51] that has never, I think, before been [10:53] told publicly. Who? Well, it turns out [10:56] Anna was a posttock at Stanford at this [10:58] time and she's the one who organized [11:00] this new student's weekend. So she [11:02] organized Larry and Sergey meeting and [11:04] she told me that the first night of the [11:05] weekend, so like the first event, the [11:08] first time that they actually met was at [11:10] drinks at the British Bankers Club in [11:12] Menllo Park and this sort of magical [11:15] friendship, bickering back and forth, [11:17] but you know, real partnership was [11:19] started there and it was already going [11:20] that night. So Larry and Sergey, they [11:22] end up shutting the bar down that night [11:24] and another famous local Stamford alum [11:28] picks up the tab for the group. [11:32] Can you guess who that person was? [11:36] Uh, I don't know. Lay it on me. Charles [11:39] Schwab. No way. Yep. Apparently lived [11:42] locally in the area, went to the British [11:44] Bankers Club all the time, and he'd do [11:45] this. He'd just see Stanford students [11:46] there and be like, "All right, I got you [11:48] guys." That's awesome. So Charles Schwab [11:50] funded the first date of Larry and [11:51] Sergey. Yeah. Amazing. And Charles [11:54] Schwab accounts would go on to hold [11:56] billions and billions and billions of [11:58] dollars worth of Google stock because of [11:59] that. [12:00] amazing. But yes, I think the spirit of [12:02] all these stories is true. It was a [12:04] instant electric friendship partnership [12:08] between the two of them and that would [12:10] carry on forever. I mean, they shared an [12:12] office at Google. I don't think we've [12:14] really covered founders before that were [12:17] true partners in the way that Larry and [12:19] Sergey are. Like, you know, reminds me a [12:21] little bit of you and me on a very [12:22] different scale. Yes. On a very [12:24] different scale. But that's interesting. [12:26] Equal co-founders. I'm racking my brain. [12:28] Maybe Bill Gates and Paul Allen, but [12:30] even then it became very clear very [12:32] quickly Bill Gates is the guy. Yeah, [12:35] that was reflected in their equity [12:36] ownership. Exactly. I'm sure we've [12:38] covered other companies where there were [12:40] equal equity ownership amongst founders, [12:42] but where it was a real true [12:44] partnership. One plus one equaled like a [12:47] hundred. It's fascinating. Yeah, maybe [12:50] Jensen, Curtis, and Chris and Nvidia, [12:53] but over time, you know, it kind of [12:55] became Jensen. Yeah, you're right. This [12:57] is sort of unique for us to be covering [13:00] true founder partners that made it [13:03] decades together. I mean, even Warren [13:05] and Charlie, Charlie was never full-time [13:08] at Berkshire Hathaway, right? And [13:10] certainly owned way less. The closest I [13:12] can think of is maybe Capital Cities [13:14] with Tom Murphy and Dan Burke. H, it's [13:18] really interesting. Okay, so Sergey, [13:20] what's his story? Sergey was also born [13:22] in 1973 uh a few months later in August [13:24] in some place even colder than Michigan [13:27] Moscow which of course then was part of [13:30] the Soviet Union and Sergey's family was [13:32] Jewish. Soviet Union was not exactly a [13:35] great time and place to be Jewish or you [13:37] know probably to really be anything [13:39] there. His family lived in a three- room [13:42] apartment in Moscow in what I assume was [13:45] state allocated housing. They shared it [13:48] with his paternal grandmother, but his [13:50] dad was a extremely talented [13:52] mathematician. And so when Sergey is [13:55] four years old, his dad attends an [13:57] international mathematics conference. [14:00] This is like 1977, 1978, and realizes, [14:04] oh, I got to get my family to the west. [14:06] We got to get out of here. So it takes [14:08] him two years to be able to immigrate [14:09] out of the Soviet Union, but they [14:10] eventually come to the US. His dad [14:12] becomes a math professor at the [14:14] University of Maryland. His mom becomes [14:16] a researcher at NASA's Gddard Space [14:18] Flight Center. Okay. So, Larry's parents [14:20] are both computer science professors. [14:22] Sergey's dad is a math professor and his [14:26] mom works for NASA. Yes. [14:30] The pool is small of people with [14:32] backgrounds like that in the 80s. Yes. [14:35] As you would imagine, Sergey is very [14:36] precocious. He graduates high school at [14:38] age 16, goes to the University of [14:41] Maryland, gets his undergrad degree in [14:43] both math and computer science in three [14:46] years, graduates at age 19, and then of [14:49] course gets into Stanford for his PhD. [14:52] Do you know what he did in the summer [14:53] before coming to Stanford? Oo, I have no [14:56] idea. He interned at Wolfrram Research. [14:59] No way. Really? Stephen Wolfrram. Oh, [15:02] shout out to friend of the show, [15:03] Stephen. I know. I know. developers of [15:05] Mathematica that well from Alpha I guess [15:07] eventually another search engine. So [15:10] yes, Sergey every bit Larry's [15:13] intellectual equal, every bit his [15:16] sparring partner and maybe a little more [15:18] zany too. More of a love for [15:20] rollerblading than Larry, let's put it [15:22] that way. Yes. And frankly, it kind of [15:24] pencils now that you're starting to get [15:26] the picture of these two over time. You [15:27] know, in the far future, closer to [15:29] today, Sergey is the one doing stuff [15:31] like Google Glass and skydiving videos. [15:35] The zany parts of Google are sort of [15:37] Sergey's DNA and the really honed [15:41] products and products that make the [15:43] business work are a little bit more [15:45] Larry's DNA. But to be honest, there's [15:47] incredible overlap everywhere between [15:49] the two of them. Yeah. Okay. So, fall of [15:52] 1995, Larry arrives at Stanford. [15:54] Sergey's already there. Terry Winterrad [15:57] is his PhD adviser and Larry and Sergey [16:00] are already building this great [16:01] friendship. Larry that academic year [16:04] presents a dissertation topic to Terry [16:07] to his adviser in collaboration with [16:09] Sergey. And the idea they have is this [16:12] worldwide web thing seems to be becoming [16:16] a thing. You know, here we're 1995. [16:19] We're what, a year after Netscape was [16:21] started, a couple years after Mosaic. I [16:23] have a stat for you on this era in the [16:25] internet. Oh yeah, light on me. This is [16:27] from John Battel's book, The Search, [16:29] which is excellent. From 1993 to 1996, [16:33] the web grew from 130 sites to more than [16:36] 600,000. [16:38] And if you compute that rate of growth [16:41] over that 4-year period, it is 723% [16:44] growth year-over-year for four years. [16:47] That is exponential. Yeah, this is the [16:50] version of the Jeff Bezos realization [16:51] where he's like, I got to leave Dean and [16:53] I got to build Amazon. Like nothing like [16:54] this has ever happened before, right? [16:57] The internet is a phenomenon like no [16:59] other. And if it keeps going, I mean, [17:01] it's amazing for something to grow 700% [17:03] year-over-year at all, but it happens. [17:06] But for that to keep happening year over [17:08] year over year for half a decade, that [17:10] is worth quitting your job, dropping [17:12] everything, changing your whole life. [17:14] Yep. And just a couple years before, two [17:18] other PhD students at Stanford had [17:20] started this thing called Yahoo. And so [17:23] I think perhaps inspired by what Yahoo [17:26] was doing, Larry proposes that they're [17:30] gonna work on this idea of a system that [17:33] will allow people to make annotations [17:36] and notes directly on websites instead [17:38] of in a centralized directory like Yahoo [17:41] because Yahoo was human handcurated [17:44] commentary, you know, directory of [17:45] websites. And so the idea is like, oh, [17:47] this could be a decentralized [17:49] annotation system where anybody can say [17:52] what's interesting about a website. Oh, [17:54] that's funny. I thought I knew the whole [17:55] history of Google. I somehow missed [17:57] this. So Terry, Larry's adviser, is [17:59] like, "Okay, you know, I like the uh [18:02] problem space, shall we say, of the web [18:04] for, you know, your dissertation here, [18:06] Larry, but why don't you go refine this [18:09] idea a little more and come back to me?" [18:11] So Larry goes off and of course he's [18:13] collaborating with Sergey on this and as [18:16] they think about it they realize that uh [18:18] actually there's sort of a fundamental [18:19] flaw in what they were planning for this [18:22] annotation system which is that for a [18:24] big site like say the New York Times or [18:26] something it's just going to get overrun [18:28] with hundreds thousands millions of [18:32] users commenting and you need a way to [18:35] separate the wheat from the chaff so to [18:37] speak of the comments. You need to be [18:40] able to have the good ones rise to the [18:41] top. You need a a way to rank them, you [18:44] might say. And so Larry has a quote [18:46] here. It wasn't that we intended to [18:48] build a search engine. We built a [18:50] ranking system to deal with annotations. [18:52] We wanted to annotate the web. Build a [18:54] system so that after you'd viewed a [18:56] page, you could click and see what smart [18:58] comments other people had about it. But [19:00] how do you decide who gets to annotate a [19:02] big site like Yahoo? We needed to figure [19:04] out how to choose which annotations [19:06] people should look at, which meant we [19:08] needed to figure out which other sites [19:10] contained comments that we should [19:12] classify as authoritative. [19:14] Hence, page rank. Ah, yes, we should say [19:18] page. It is ironic that the things that [19:22] were being ranked were web pages cuz the [19:24] actual page and page rank is named for a [19:26] Larry page, not for the pages they would [19:29] rank. Right. Exactly. So, uh, Larry goes [19:32] back to Terry and he's like, "Okay, this [19:34] ranking idea, this seems like a really [19:37] interesting computer science problem. [19:39] The annotation thing seems messy. Why [19:41] don't you just focus on rankings?" So, [19:45] Larry goes back and ultimately has the [19:47] breakthrough leap. Oh, we should apply [19:50] rankings to web pages themselves. Larry [19:53] says, "Wow, the big problem here is not [19:56] annotation. We should use it not for [19:58] ranking annotations but for ranking [20:00] searches. Ding ding ding ding ding. And [20:02] thus at least the germ of the idea for [20:05] page rank as we all know it today is [20:07] born. So essentially just the mechanics [20:10] of what the idea is is try to rank [20:14] websites based on how authoritative they [20:17] are based on how credible they are. And [20:20] this is something that has been done [20:22] somewhere before the web very close to [20:25] home for all these Stanford folks. [20:27] Academia of course. Yes. How important [20:30] is a research paper? Well, that depends [20:32] how many other people cited the research [20:34] paper. And in particular, not just how [20:37] many raw number of other papers cite a [20:41] research paper. How many important [20:43] papers site your research paper? If [20:46] you're in an important journal, what do [20:48] those papers site? And this becomes the [20:51] inspiration for how they're going to do [20:52] the ranking of web pages. And this was [20:54] actually a research field before the [20:56] web, the study of academic citations. [20:58] There's already sort of a body of work [21:00] around how to do this. Well, oh, [21:03] interesting. I didn't know that [21:04] actually. I mean, it makes sense. It's [21:06] kind of like how Hollywood loves making [21:07] movies about Hollywood. Academia loves [21:10] doing papers about papers. Yeah. So [21:13] there are some examples to look at of [21:15] how might one use references or [21:18] citations to weight importance. Right? [21:22] So we're almost all the way there to the [21:26] huge leap that would become page rank [21:28] and then back rub and ultimately Google. [21:31] But there's still one missing piece. [21:33] They've got the theory of how to do [21:35] this. But what's a citation on the web? [21:40] Well, they realize it's a link. A [21:42] hyperlink is the exact same model as an [21:45] academic citation. And not only is it [21:50] the same as a citation, it's even better [21:53] because there's this metadata embedded [21:56] within the link, which is the anchor [21:59] text. Anytime you click a link, you [22:00] know, anybody who's creating a link can [22:03] make anchor text for it or if you're [22:04] like uh write whatever they want and [22:06] then just today command K on your [22:08] keyboard and then you can make that text [22:10] into a link. Well, that's pretty easy to [22:13] identify as metadata on an HTML page. [22:16] And so you get not only a citation of [22:19] the link, but a few words of what the [22:21] author of that link thought about it. [22:23] Yes. When someone is linking to you, [22:26] they often do a better job of describing [22:29] your website than you do on the page [22:31] yourself. Anybody that's just sort of [22:33] looking at your website to try to figure [22:35] out what's this about, the actual words [22:37] on your website tend not to do as good a [22:39] job as everyone who links to you in [22:42] aggregate. What words did they use to [22:44] describe your website? So, this whole [22:46] thing is a genius idea and we're going [22:50] to talk about all the work that they had [22:51] to do to implement it. It basically [22:52] works right away. The notion of hey, [22:55] what is the output if we try to create a [22:57] system that ranks all websites for [23:00] authoritiveness based on how many other [23:02] reputable websites are linking to it and [23:05] then later on we can use the anchor [23:08] text. But right now, just this ranking [23:10] system, it spits out a list that's [23:12] sorted exactly as you would hope. It is [23:15] the most authoritative websites first [23:17] and all the crap all the way at the [23:18] bottom. Yep. So Terry's like, "Yeah, do [23:23] this for your dissertation." Like, [23:25] "Great. Let's do this project." [23:26] Reputation on the web. That's going to [23:28] be valuable. So there's one more thing [23:30] to making this brilliant page rank idea [23:33] work, which is a little problem, which [23:36] is that the way the web is architected. [23:39] Any given web page only shows outgoing [23:43] links. There's no way to query a page [23:46] and say, "Oh, who links to me?" You can [23:49] only query a page and say who do you [23:51] link to? Right. It's kind of like it's [23:53] easy for me to answer the question who's [23:55] in my phone book in my phone today. It's [23:58] hard for me to answer the question whose [24:00] phone books am I in? Exactly. And so the [24:03] only way you could figure that out is if [24:05] you somehow went and got a copy of [24:07] everybody's phone book, right? And then [24:11] could back trace all the links. Well, [24:13] that's what they do. Google could not [24:16] have been built at any other time in [24:18] history. Yes. Because the web was [24:20] actually small enough that you could go [24:22] suck it all up then. Exactly. It was [24:24] small enough that as a research project, [24:27] it was not totally insane. Just a little [24:31] bit insane. It's still pretty insane to [24:33] say, "Oh, I'm going to go crawl the [24:37] entire internet, make a copy of every [24:39] web page out there, store it in [24:42] something, which we'll get to, [24:44] and then trace back all the links, and [24:46] then reverse compute all the links to [24:48] answer that one seemingly simple [24:50] question of what web pages link to me." [24:52] If you had tried to undertake this as a [24:54] brand new project, even just a year or a [24:58] couple years later, it would have been [25:00] impossible because the web would have [25:02] already gotten so big that to just start [25:04] denovo and create a index copy like [25:08] this, like a full copy even one year [25:11] later would have been tens of millions [25:12] of dollars and within a couple years [25:13] would have been hundreds of millions of [25:14] dollars. Yeah. Prohibitively expensive. [25:18] So, we're in what? 96 97 here. Yep. [25:20] We're in 96, kind of the back half of [25:22] that first academic year of Larry at [25:25] Stanford. Great. So, with Terry's [25:27] encouraging Larry and Sergey go forth to [25:31] undertake this ambitious project, they [25:33] set up a page on the Stamford internet. [25:37] They decide that they're going to call [25:39] the project back since it uses back [25:43] links for ranking web pages. So, they [25:47] spin up back.stanford stanford.edu [25:52] and Larry writes the first [25:54] implementation of page rank and the [25:56] crawler to go do this. He writes it in [25:58] Java and it's like super buggy and [26:01] basically doesn't work. So they ask one [26:04] of their friends there at Stanford, guy [26:06] named Scott Hassan to help them. He [26:09] Scott's a better coder than Larry and [26:10] Sergey is. So he codes it up in Python [26:13] and it actually sort of works. Now, [26:17] Scott, not an employee of Google, never [26:19] would become an employee of Google [26:21] because it's not a company yet. It's a [26:23] research project, right? And they hadn't [26:24] even come up with the name Google. I [26:26] mean, nothing about this. It's not a [26:27] search engine. Exactly. So, if you go to [26:30] that project homepage and you can find [26:32] cached versions of this on the internet [26:34] or images. I think there's even a [26:35] recreation of it out there. Yes. With [26:38] the really weird black and white picture [26:40] of a back rub with the red text over it. [26:42] Yeah. I think that picture actually was [26:44] on back rub.stanford.edu. edu. It looks [26:46] like somebody's back. They certainly [26:48] didn't search Google images for it. I [26:50] know that. Yeah. Exactly. Exactly. So, [26:52] the text on the page says, "Back rub is [26:55] a quote unquote web crawler which is [26:57] designed to traverse the web. Currently, [26:59] we are developing techniques to improve [27:02] web search engines." So, yeah, Ben, [27:03] they're not thinking of Back Rub as a [27:05] search engine yet. They're thinking of [27:07] back rub as just a implementation of a [27:09] crawler and the page rank algorithm. Oh, [27:12] that's interesting. But they have the [27:14] insight that this method of ranking, if [27:18] it turns out to be better, could [27:20] contribute to better search engines. [27:22] Yep. Because again, even though Larry [27:24] really wants to start a company, he's [27:26] thinking he's going to get his PhD and [27:28] then go forth here. Kind of like Mark [27:29] Zuckerberg got years into Facebook [27:31] before realizing, oh, this is my [27:32] company. So Larry and Sergey are [27:34] building Back Rub here. And I mentioned [27:37] Scott Hassan, their friend who helps [27:39] code it up in Python, doesn't end up [27:41] joining Google. Well, what happened to [27:43] Scott? Scott leaves while all this is [27:46] going on and starts a company because [27:50] it's 1996, 1997. You're here in Silicon [27:53] Valley, right? The bubble is inflating. [27:55] Like, what do you do? You go start a [27:56] company. It's your obligation to go take [27:58] a smash this piñata. So, Scott leaves, [28:02] starts a company called egroups. That [28:06] company a couple years later ends up [28:07] getting acquired by Yahoo, becomes Yahoo [28:09] Groups for about $400 million. [28:12] Do you know who co-founded egroups with [28:15] Scott? Man, you're stumping me today. [28:18] No. Larry's older brother, Carl Paige. [28:21] Oh, that's this company. Yes. Wow. So, [28:25] here we are. We've got Larry and Sergey [28:27] doing this research project to improve [28:31] search engines. Meanwhile, their buddy [28:34] who helped code that and Larry's older [28:35] brother, they just went and they started [28:37] a company and then they raised money. [28:39] Sequoa and Mike Merittz would end up [28:41] funding egroups and then they sell it [28:43] for $400 million to the leading web [28:45] company at the time. That seems like a [28:47] good idea. Yeah. Larry and Sergey start [28:49] thinking, "Ooh, wait, maybe we should do [28:53] something commercial around this. This [28:54] seems like it would have value." [28:57] So that leads them in the spring of 1997 [29:02] before school is out to start shopping [29:05] this back rub technology around to the [29:07] other existing search engines at the [29:10] time. They're not yet thinking that this [29:12] could be a company. They're thinking [29:13] we're going to sell this technology to [29:16] another search engine. They're going to [29:17] pay us a lot of money for it. We'll go [29:20] come help implement it and then we'll go [29:21] back and finish our PhDs. because they [29:24] effectively have it working at this [29:25] point even though it's not you know [29:28] hardened. They have a crawler that has [29:30] run on a small number of websites. They [29:33] have a very modest index that has been [29:36] created. It's not all efficient and [29:39] everything but they sort of have the [29:40] proof of hey look this ranking is [29:42] actually a good ranking of how [29:44] authoritative these websites are. But [29:46] other than some demo proof of concepts, [29:48] they haven't yet built the consumer [29:50] version of, hey, you can go query this [29:52] thing and anybody can use it, right? So [29:55] they're shopping it around that spring, [29:57] that summer. They get a bunch of [29:58] meetings. They meet with infosek, lyos, [30:00] all the existing search engines. And we [30:02] should say there was a large list of [30:05] other search engines, portals, internet [30:08] properties, search engine like things. [30:11] Yes. That existed. It had traffic. I [30:14] mean Archie, Gopher, Alta Vista, Hotbot, [30:17] Inktoy, Los, Yahoo, Excite, Infosek. The [30:22] list goes on and on and on. Yes. So the [30:25] closest they get that spring and summer [30:27] is with Excite. The story is amazing. So [30:31] supposedly this is according to In the [30:32] Plex, Steven Levy's book, great book [30:34] that we used as a source for the [30:35] episode. They end up getting a meeting [30:38] with Venode Kosla, a legendary founder [30:41] of Sun Micro Systemystems. By this point [30:43] in time, he's one of the top VCs in the [30:46] valley. He's at Kleiner Perkins [30:48] alongside John Door. The two of them are [30:50] running the firm and Venode is on the [30:52] board of Excite. And so somehow Larry [30:56] and Sergey and I think they bring Scott [30:58] along get a meeting with Venode and they [31:00] hammer out a deal that Excite is going [31:02] to license this back rub search [31:05] technology from the two of them for [31:07] about a million dollars. Part of that's [31:09] in cash, part of that's in Excite stock. [31:11] and Larry and Sergey are going to come [31:14] work get excite that summer implement [31:17] back rub for their search, you know, [31:20] basically make excite into Google and [31:22] then they're going to leave and they're [31:23] going to go back to Stanford in the [31:24] fall. And they get so far that they run [31:28] a test, a sideby-side test of Excites [31:32] search results, the original, you know, [31:34] algorithm and then the back rub [31:36] algorithm. And the legend goes that [31:39] they're demoing this test to excite CEO [31:44] as like a final step to finalizing this [31:47] deal. And the results are so relevant [31:50] with back rub. You know, you get exactly [31:52] what you search for, exactly what you [31:54] want. It's right there. You click, you [31:56] go to it. And the usual excite search is [31:59] bad. You have to like click around, you [32:01] go forward, you come back. You spend a [32:03] lot of time on the site. And the CEO is [32:06] like, "Why on earth would we move to [32:09] your algorithm? I want people to stay on [32:12] my site. I make money when people stay [32:15] on my site. I don't want them to leave [32:17] my site. You guys are crazy. Get out of [32:19] here. I'm killing the whole deal." It is [32:21] amazing that as early as 19967, [32:25] this time period, this very important [32:29] conflict of interests is teased out. [32:32] This is basically why Google beat Yahoo. [32:35] I mean, there's a lot more to it, but [32:37] the portals all had this mentality of we [32:40] want to build more and more and keep [32:43] people on our site in our ecosystem, [32:45] continue to look at our banner ads. And [32:47] Google from this point, basically [32:49] forever, was how quickly can we deliver [32:52] someone something relevant so they can [32:54] leave Google and have had a good [32:56] experience finding what they really [32:57] wanted. Now all this is laughable in [33:00] hindsight but made total sense at the [33:03] time because what was the business [33:06] paradigm for all these sites? What was [33:08] the model? It was banner ads. It was CPM [33:11] cost per thousand views. You wanted page [33:15] views and impressions on your site. And [33:18] here what Backrub is doing is they're [33:20] going to kneecap your page views, right? [33:22] They're going to dramatically reduce the [33:23] number of page views and allocate them [33:25] to other properties on the web so they [33:27] can leave your property. This deal was [33:29] never going to happen with Excite or [33:30] anybody else because it broke the [33:31] business model, right? It was not [33:33] strategic for them. It was a conflict of [33:35] interest to implement this type of [33:38] better ranking, better search, faster [33:40] technology. So, back to the lab, right? [33:43] No deal. Yeah, exactly. That fall, 1997, [33:47] Larry and Sergey go back to school. All [33:49] these get-richqu deals have fallen [33:51] apart. Nobody wants back rub. And [33:54] Larry's just like, "Well, f it. All [33:57] right. This is how I believe search [33:59] should work. We're going to build this [34:01] thing ourselves here at Stanford." Quote [34:03] from him. We couldn't get anyone [34:04] interested in buying back rub. We did [34:06] get offers, but they weren't for much [34:07] money. So, we said whatever. We went [34:09] back to Stanford to work on it some [34:11] more. These companies weren't going to [34:12] focus on search. They were becoming [34:14] portals, i.e. they wanted eyeballs, page [34:16] views. They didn't understand search and [34:18] they weren't technology people. Yeah. [34:20] And I mean when he says we talked to [34:22] others, they tried to sell page rank to [34:24] Yahoo for $1 million and were rejected. [34:28] And there's going to be chapters of this [34:31] story that you can mark by the different [34:32] times that Yahoo discussed buying Google [34:35] and didn't. But this is the very first. [34:37] They showed the tech to Infosek. Also [34:40] didn't happen. Infosek was bought by [34:41] Disney and later shut down. They showed [34:43] the tech to LOS. They were shopping it [34:45] all around town. So they get back to [34:47] campus. They're like, "All right, we're [34:49] going to build a real search engine [34:50] ourselves. First order of business, the [34:54] name [34:56] back rub. Uh, [34:58] you're probably not going to fly." Yeah. [35:00] Describes the technical underpinnings, [35:02] but doesn't really describe the [35:04] searching. Yeah. So, they're casting [35:07] about trying to find the right name to [35:08] encapsulate what they're doing. This [35:10] sort of new good form of search. And the [35:13] story is that Larry's dormate suggests [35:16] the term Google. Oh, wait. Do you know [35:19] the name before this? Oh no. Do I know [35:22] something you don't? Oh, this is great. [35:23] Oh, go for it. The name is what box? [35:27] What box? What box? It rolls right off [35:30] the tongue. You know, it's a box that [35:32] you type stuff into. Sort of a question. [35:35] I got it. But they decided that it [35:36] sounded too close to a porn site, so [35:39] they decided not to go with it. Well, [35:41] hey, I guess if Facebook can be a [35:44] product name and a company. Yeah, maybe [35:46] we could all be what boxing things. I [35:48] mean, we are all WhatsAppapping, so to [35:50] be fair. Yeah, it wasn't that crazy. It [35:53] wasn't that crazy. But yes, what box out [35:56] next name? How did Google come about? [35:58] Larry's dorm mate suggests that they [36:01] might want to use the term Google G O L [36:07] which is the mathematical term for one [36:10] followed by 100 zeros 10 to the 100th [36:13] power. Yes. And the legend is that Larry [36:17] loves the name. Sergey likes it. They go [36:20] to register the domain name and Larry [36:23] misspells it and thought that Google was [36:25] spelled G O O G L E. misspells. Really? [36:28] I thought go.com [36:30] was taken. Oh, maybe that's it. Maybe [36:33] that's it. Like anything here, you know, [36:35] there's a lot of legends floating [36:36] around. Yes. But the misspelling is [36:38] actually great because you kind of [36:40] should spell it the way that other [36:43] people are most likely to spell it. [36:45] Right. Exactly. So, this is before uh [36:48] you've got the Google did you mean in [36:50] the search box. Yeah. Yes. Spelling was [36:52] important. So Sergey designs the [36:56] homepage and makes the first logo using [36:59] the open source drawing program [37:01] You ever use back in the day? Oh, [37:02] and you can tell it is drawn using [37:05] A lot of people probably can think of [37:07] the earliest Google logo you've ever [37:09] seen. Even the real nerds out there are [37:11] like, "Oh, yeah, I know about that one [37:13] that was real colorful before the drop [37:15] shadow thing." There's even one before [37:17] that that's like completely illeible, [37:20] and this is the the one that we're [37:22] talking about. But it was rainbow [37:23] colored. Yes, it was. We'll link to it [37:25] in the show notes and on social media. [37:27] It's fun to look at. Yep. But basically [37:31] that homepage design of a colorful logo [37:36] and a search box that was it then that's [37:39] it today 97 onward. So that 97 98 [37:43] academic year is when they're building [37:45] back rub into Google by spring quarter [37:49] of that year Google.com [37:52] is doing 10,000 queries a day. This has [37:56] started to spread virally first on the [37:58] Stanford campus and then to other [38:00] academic universities and communities [38:03] out there. get wind of what they're [38:04] doing and then it starts spreading into [38:06] Silicon Valley and it's like bringing [38:08] the Stanford network to its knees with [38:10] all the traffic that is happening on [38:12] google.com out of Stanford. So at one [38:15] point they actually did bring down the [38:17] Stanford network. This is how fast it [38:19] all happened. It's all during this [38:21] academic calendar where they're taking [38:24] back rub they're working in a search [38:27] box. So there's now a keyword that we're [38:29] ranking things for, not just arbitrarily [38:32] ranking them. That keyword relies [38:34] heavily on the anchor text description. [38:35] So there's sort of these two early key [38:37] innovations. There's waiting results [38:40] based on backlinks and there's [38:42] description from anchor text. And it [38:45] really did just kind of work. The [38:47] technical underpinnings are extremely [38:48] difficult. They're having to do things [38:50] like steal computers from other research [38:53] projects. Do you know about the loading [38:55] dock stuff, David? Oh, yeah. Yeah. Yeah. [38:56] There's these famous stories of other [38:58] researchers that have ordered computers, [39:00] but they actually aren't going to start [39:01] the project for a few months. So Larry [39:03] and Sergey would go grab them off the [39:05] loading dock, spin them up, use them for [39:07] Google just for a few months until they [39:10] need to go and hand them over for the [39:12] other research projects. David, to your [39:14] point, they bring down the Stanford [39:16] network because there's so much traffic. [39:18] There's so much demand for something [39:19] that is just clearly a better way of [39:21] ranking websites than what everybody [39:23] else was doing. you know, everyone else [39:24] is basically just using keywords on [39:27] pages and saying, "Well, what pages [39:30] exist out there with the word dog?" And [39:32] if they have a whole bunch of instances [39:34] of dog, then that's going to be the top [39:37] of your results for dog, no matter how [39:38] authoritative they are. And so, [39:40] obviously, that's a problem. So, this is [39:42] just a better way to do search. And [39:45] they're really starting to soop up [39:47] Stanford's network bandwidth. At one [39:49] point they're using about half the [39:51] bandwidth of the entire university to be [39:54] serving out google.com pages. And David, [39:58] to your point, it's not a heavy website. [40:00] I mean, it's a white page with an image, [40:02] a search box, and then when you go to [40:04] the results page, there's no images. And [40:06] so, if they're consuming an incredible [40:07] amount of bandwidth for something that's [40:09] so asset light, people are using the [40:12] crap out of this thing. Yep. So, here we [40:14] are, end of the 1998 academic year. It's [40:17] clear this is going to be a company. [40:19] This has to be a company. Stanford's [40:21] about to tip over if it stays as a [40:24] project anymore. Stanford has been very [40:26] kind to say we're going to keep housing [40:28] all the infrastructure for this thing, [40:30] but at some point this needs to be a [40:31] company so that you can get it off of [40:33] our network and fund it on your own. And [40:35] nobody should shed a tear for Stanford [40:37] here because as part of the tech [40:38] transfer to spin it out of the [40:41] university, they end up getting like 1% [40:43] of the company or something like that. [40:44] Stanford did very very well for their [40:47] large S here. Yes. So Larry and Sergey [40:51] go to a professor in the CS department [40:53] at Stanford named Dave Sheridan. And [40:57] Dave had started an Ethernet company [41:00] called Granite Systems with Andy [41:02] Bealshime from Sun while also staying as [41:05] a professor at Stanford at the same [41:07] time. He was a founder of Granite [41:09] Systems and had stayed as a professor. [41:11] Cisco had just acquired Granite for 220 [41:15] million. And so Larry and Sergey are [41:17] like, "Oh, okay. Dave, he's one of our [41:18] professors. He knows how to do this [41:20] stuff." And he's like, "Well, why don't [41:23] you talk to Andy about how we could spin [41:25] this out and make it into a company?" So [41:28] Dave emails Andy that evening. Andy [41:31] replies right away. He's like, "Sure, [41:32] I'm kind of busy tomorrow, but how about [41:35] we meet at your house at 8 a.m. in the [41:36] morning? I'll come by on my way to the [41:38] office." And thus begins the story of [41:41] Google's legendary [41:44] seed financing round and the crazy cast [41:47] of characters involved in it. But before [41:49] we tell that story, now is a great time [41:52] to talk about our presenting partner JP [41:54] Morgan Payments. So listeners, you've [41:56] heard us talk about specific aspects [41:58] over the last few episodes like [41:59] biometric payments, their developer [42:01] platform, and their supply chain [42:02] financing. But today we wanted to pop up [42:05] a level and tell you about the acquired [42:07] and JP Morgan partnership itself. Well, [42:10] first is the team. The folks there at JP [42:13] Morgan are a absolutely world class and [42:16] b really get acquired and what makes [42:18] this community special. Just look at the [42:20] Chase Center show we did. They took our [42:22] vision and said, "What if we did all [42:23] this times 10?" And now we're about to [42:25] go do it all over again at Radio City. [42:27] And deeper than that, they're just an [42:28] insanely trusted brand. JP Morgan [42:31] Payments is the world's largest payment [42:33] franchise. They power 18 of the top 20 [42:35] corporations in the world and most [42:38] companies we've covered on the show. In [42:40] fact, over 90% of the Fortune 500 [42:42] companies do business with them. A [42:44] couple years ago, we were worried that, [42:46] oh, JP Morgan Payments might only be for [42:49] big companies, but it's not. We've seen [42:51] startups that heard about them when [42:52] acquired become customers. And that's [42:54] our goal in picking partners is to find [42:56] the very best companies that create [42:57] value for listeners and will scale with [42:59] your success and be around forever. That [43:01] is JP Morgan Payments. They literally do [43:04] $10 trillion in payment volume a day. [43:08] Think about how insane that is. And with [43:10] JP Morgan processing over 50% of all US [43:13] e-commerce transactions, their software [43:15] and payment rails basically underpin our [43:18] entire global financial system. Yes. And [43:21] lastly, every single one of your [43:22] companies needs payments. JP Morgan [43:24] thinks about payments as a lever for [43:26] growth, not just vanilla operational [43:28] stuff. They've been investing heavily [43:30] with products now for fraud prevention, [43:32] FX, working capital, and more. All of [43:34] course built enterprisegrade and with [43:36] developer tools and APIs. You can learn [43:39] more at jporggan.com/acquired, [43:41] which itself is a cool custom site [43:43] they've built that has details on the [43:45] products we've been talking about all [43:46] season, plus a little behind-the-scenes [43:48] video of acquired live at Chase Center [43:51] from last year. When you get in touch, [43:52] just tell them that Ben and David sent [43:54] you or shoot us a message in Slack, and [43:56] we'll get you connected with their team. [43:58] All right, David, the Google seed round. [44:02] Yes. Here we go. So 8 am the next [44:05] morning, Larry and Sergey rouse [44:07] themselves out of bed over on the [44:09] Stafford campus, head on over to [44:11] downtown Palo Alto at Dave's house, and [44:14] Andy drives up. He's like, "All right, [44:16] I'm in a hurry. Show me what you got." [44:18] They demo Google for him. Andy loves it. [44:21] He's like, "Great, I'm in. $100,000." [44:24] And Larry and Sergey are like, "But we [44:26] weren't talking about raising money. We [44:28] just wanted some advice to start a [44:30] company." Andy's like, "Great. I'll go [44:31] get the check from my car. [44:34] He writes a check to Larry and Sergey [44:36] made out to Google Inc. for $100,000. [44:40] Basically just throws it at them, hops [44:42] in his car, and takes off. [44:44] Google Inc. does not exist yet. [44:48] This is actually true. This actually [44:50] happened. Andy's like, "You guys figure [44:51] this out. That's your problem, not mine. [44:53] I'm good for the money." No investment [44:55] documents, no valuation. Just here's [44:58] $100,000. I assume I will get something [45:01] for my investment. Yep. Exactly. And [45:04] this was the forcing function for [45:06] Google, Inc. to get founded. So Larry [45:10] and Sergey need to be able to like spin [45:12] up an entity, have that own the [45:15] intellectual property from Stanford and [45:19] set up a bank account for that entity [45:21] such that they can deposit this check [45:22] before it expires. It takes a a couple [45:25] months to get all this done. Yes. which [45:27] depending on who you ask is either very [45:30] good or very bad because in the [45:32] intervening months Dave himself decides [45:35] to throw in another $100,000 to the [45:38] funding here and Larry and Sergey meet a [45:41] former Netscape guy named Rahm Shriram [45:45] who started advising them on starting [45:48] this company spinning things out and [45:49] longtime acquired listeners will [45:51] recognize this name from our Amazon [45:53] episode. Oh yes, Rahm had left Netscape [45:57] and joined a startup called Jungly that [46:00] Amazon.com then acquired. And so Rahm [46:04] had a little bit of liquidity. He throws [46:07] in for $250,000 [46:09] into the round. And he's like, "Hey, do [46:12] you guys want to meet Jeff?" Jeff Jeff [46:14] Bezos, [46:16] who at this point, it's kind of [46:17] interesting. You think about Amazon and [46:18] Google as kind of equally old companies. [46:21] Jeff is sort of the elder statesman of [46:23] the internet. His company was started in [46:25] '94. This is '98. Amazon just went [46:27] public the year before. I mean, it's [46:29] kind of crazy that at this point in [46:31] time, it was little old Lar and Serge [46:33] grad students meeting public CEO Jeff [46:36] Bezos. Yeah. Yeah. It's almost like when [46:38] your six-month old baby is hanging out [46:40] with a 14-month-old, you're like, "Oh my [46:43] god, they're so different." But like [46:44] they're going to be classmates in a [46:45] couple years. They're basically the same [46:47] age. Exactly. So Rahm arranges a meeting [46:50] and the next time that Jeff is in [46:52] Silicon Valley, they all meet at Rom's [46:54] house and similar to Andy Bealshime, [46:58] Jeff's like, "Great Rom, what are you in [47:00] for? 250? I'm in for 250, too." [47:05] And so all in, they end up raising a [47:07] million dollars at a $10 million post [47:09] money valuation. And yes, Jeff Bezos [47:13] does a quarter of Google's seed round. [47:16] It's so crazy. You know, we knew about [47:18] this in the past because we talked about [47:19] on the Amazon episode, but whenever I [47:21] heard someone reference Jeff Bezos was [47:23] an angel investor in Google, I always [47:24] thought, well, yeah, but you look at any [47:27] of these startup cap tables and there's [47:28] 50 founder friends in addition to the [47:31] main VC of that's not surprising at all. [47:34] But Jeff is a quarter of the money in [47:35] the seed round. Yeah. One of four [47:38] investors. Is that right? That's right. [47:40] Yep. It's Andy, Dave, Rom, and Jeff. [47:42] Jeff has never said whether he sold any [47:45] of his Google shares along the way. But [47:47] by my math, if he didn't, that stake is [47:51] worth about 20 billion dollars today. [47:53] And even if he did sell at IPO, he [47:55] turned that 250K into something like 200 [47:58] million at IPO. Right. Right. Right. [48:01] Nice returns. And Amazon stock was [48:04] probably in the dumpster when Google [48:06] went public. So, you know, could have [48:08] used the money. Anyway, Google is now an [48:11] official company. They've got a million [48:13] dollars in cash from their crazy seed [48:15] round. It seems like they might burn [48:17] through that kind of fast because of [48:19] their uh business model of trying to [48:21] store the entire internet on uh their [48:23] servers. But, you know, you got a [48:26] million bucks. You're no longer a [48:27] Stanford project. You're spun out. You [48:30] got investors. What's next? Well, first [48:33] office space. [48:35] So famously they go find space in a [48:38] Menllo Park garage, a house owned by one [48:42] Susan Wajiski who was a manager at Intel [48:46] and soon would become herself an early [48:49] Google employee and then eventually CEO [48:51] of YouTube. Yeah, exactly. But besides [48:53] office space, it's time to put your [48:56] product out into the world. And I think [48:58] it's worth drilling in here. What was [49:02] the state of search in 1998 when Google [49:06] becomes a company? Yep. I think it's [49:08] worth saying a little bit more about two [49:10] other players to set the context. The [49:13] first is Alta Vista. Folks who are old [49:16] enough might remember. Alta Vista pre [49:19] Google was like pretty good. It was [49:22] pretty good. Alta Vista has a [49:24] fascinating history. This is wild. I [49:27] didn't know this until doing research [49:28] for this episode. Do you know where [49:29] Ultimista came from? I do, but most [49:32] people don't. Deck Digital Equipment [49:35] Corporation. Deck Digital Equipment [49:38] Corporation's Western Research [49:40] Laboratory, which was their PaloAlto [49:43] Research Lab, their equivalent of Bell [49:45] Labs. But Deck, I mean, the company we [49:47] talked all about in our Microsoft series [49:49] where Dave Cutler came from, who wrote [49:52] Windows NT, you know, legendary, [49:54] legendary company, right? hardcore [49:57] enterprise big hardware computing [50:01] company. The minicomp computer company [50:03] Judy Faulner wrote Epic on a deck mini [50:06] computer, right? Yep. This is where [50:07] Altivista came from. And the big insight [50:11] that they had was you can go crawl the [50:15] web to build the index in parallel. So [50:18] before Alta Vista, all the other web [50:21] crawlers out there that were building [50:22] search engines, they were just like [50:24] singlethreaded processes. You go crawl [50:27] one page and then you go crawl another [50:29] page and then you go crawl another page. [50:31] And the internet was small enough back [50:32] then that people didn't really think to [50:34] do it any other way because remember [50:36] it's all growing so fast. Oh, [50:38] fascinating. And it's a super [50:39] parallelizable process. Why not? Oh, and [50:42] this makes sense because Deck hardware [50:43] would be pretty well suited for this. [50:45] They've got a big enterprise appliance [50:47] that they need applications for. [50:50] Exactly. This is why it was a research [50:52] project at Deck. It was a way to show [50:55] off the power of their latest enterprise [50:58] class servers that they were hawking. [51:00] So, if you think about what are the [51:03] competitive vectors of search, what [51:05] makes one search engine better than [51:07] others, it's not just the ranking. Today [51:11] people think about page rank and Google [51:12] and the innovation and the ranking and [51:14] the relevancy that was the most [51:16] important thing. There are actually two [51:18] other vectors that are critical. Y one [51:20] is speed. How fast are you going to [51:22] return the results? Which we take for [51:23] granted today but not too long before [51:25] Google's founding. Search was a thing [51:28] where you'd kick off a query and then go [51:30] do something else and wait for it to [51:32] come back. Right. It was like AI today. [51:34] You're doing deep research. You're just [51:35] like, "Okay, great. Send a query, you [51:36] know, go get some lunch. Come back." [51:38] Exactly. We'll get more into speed in a [51:40] minute. [51:40] But the other attribute that's super [51:42] important is the index. How big is the [51:47] index of pages that you're searching [51:49] across? And before Alt Vista and [51:53] parallelization of crawling, all the [51:55] indexes of all the other search engines [51:57] were super small. Maybe a million pages [51:59] was like the biggest. So you could have [52:01] the best search engine in the world, but [52:03] if you're only getting a small [52:04] percentage of the actual sites out there [52:07] that it's searching, it's not going to [52:09] be that useful. And the funny thing [52:11] about this period of time too is very [52:13] rarely were people actually updating [52:14] their index. So when something would say [52:17] like a million pages crawled, that was [52:19] cumulative and they just kept adding new [52:22] sites to it and assuming you weren't [52:23] doing many updates. Right. Right. Right. [52:26] Oh man, how times have changed. So when [52:28] Alta Vista spins out of deck and [52:30] launches as a commercial company, an [52:33] entity in and of itself, its big claim [52:35] to fame is its index. It has 16 million [52:38] pages in its index versus the [52:40] competitors. Alt Vista still kind of [52:42] sucked at relevancy and speed. So the [52:46] information retrieval algorithms that [52:47] Alt Vista and others were using was [52:50] highly based on how many times a given [52:53] query word appeared on the page. So if [52:57] you wanted to rank highly for dog food, [53:00] you just spam dog food in invisible text [53:03] all over your page or even not [53:05] invisible, you just want to make the dog [53:07] foodiest dog food page on the web. [53:09] Exactly. And the other thing that you [53:11] know I guess Alta Vista was probably [53:13] fine at this but not good was speed. And [53:17] that was a particular problem. I mean [53:19] Alta Vista had great hardware from Deck [53:21] and really expensive hardware too. [53:23] That's a main thing to underscore here [53:25] is that yeah, they're doing this cool [53:26] parallelization thing which leads to a [53:28] bigger index, but if they had to be [53:30] their own company, it is an extremely [53:32] expensive company to run to have all [53:34] that deck hardware for search, which we [53:37] should say doesn't have a great business [53:39] model yet. The business model is just [53:41] banner ads, low price, not very [53:44] targeted. That would all come later. And [53:47] so the whole search market, why would [53:48] anyone take it seriously? cuz to date it [53:51] doesn't feel like there's a good [53:52] business there and to do a really good [53:54] job at it, it would be very expensive to [53:56] run. Yes, it is funny. As you can [53:59] imagine, the powers that be at Deck are [54:02] pushing really hard on can we sell more [54:05] of these boxes. Do they really want to [54:08] be the ones developing the best search [54:10] engine or really what they want is for [54:13] other people to be developing stuff like [54:15] this, see it as a proof of concept and [54:17] then start their own companies to buy [54:19] more and more deck hardware? Yes, [54:21] exactly. Ben, search as a industry was [54:24] not interesting. One because the [54:26] economic upside was capped and two also [54:31] people love directories and portals and [54:35] Yahoo. like Yahoo was the big player, [54:37] not Alta Vista or Excite or Loss or [54:40] Infosek or any of these others. So Yahoo [54:43] was the site that was taking off like [54:45] wildfire. They'd gone public in 1996 at [54:48] a billion dollar market cap. By 1998 [54:51] when google.com is launching as a [54:53] company, Yahoo is a 20 billion public [54:56] stock. This is the juggernaut. And what [55:00] was so great about Yahoo, like yes, it [55:01] was started by Jerry Yang and David Pho, [55:03] you know, two other Stanford PhDs, but [55:06] it wasn't technologydriven. It started [55:09] as Dave and Jerry's guide to the [55:11] internet. [55:12] It didn't start as their like academic [55:14] research. What Yahoo was was exactly [55:18] that. It was a handcurated [55:21] guide directory to the internet. It was [55:24] kind of like the yellow pages with even [55:27] better annotations to why you would want [55:30] to look at a particular given site. And [55:32] that's what people thought like, oh, [55:34] technology search engines will never be [55:36] able to replace human curation and human [55:39] thought about what the most interesting [55:41] sites on the web are. Hell, this is why [55:43] Larry's original idea was this [55:45] annotation idea. It was humans who are [55:47] going to rank things. And for the size [55:49] that the web was at the time, Yahoo was [55:52] correct. Yes, when you have a small [55:55] number of total websites, curating them [55:57] is interesting. But when you have 10,000 [56:00] times more, [56:02] more niches that people are interested [56:04] in, directory is not going to be an [56:07] efficient way to surface what people are [56:09] looking for. If you believed that the [56:12] internet was going to get as big as it [56:13] did, search became a more interesting [56:16] front door. But for this period of time, [56:18] directory was an amazing front door to [56:20] the internet. Yep. Now here we are in [56:23] 1998. The internet is already big enough [56:25] that yes, it's clear there are a lot [56:27] more interesting web pages out there. So [56:29] Yahoo has search to address that. [56:31] Exactly. So the model was hybrid. All [56:34] these portals and Yahoo included went [56:36] hybrid of when you search on Yahoo, the [56:40] results you get at the top of the page [56:41] are their handcurated directorydriven [56:44] results and then they backfill with a [56:47] search engine. And so they would partner [56:49] with these search engines to provide [56:51] backfill results. And people thought [56:53] that this was the ideal solution. So [56:55] interesting. It is the epitome of kind [56:58] of just good enough technology. It was [57:00] so different than Google. Google wants [57:02] to be the very best technology solution [57:04] for a problem, the most elegant. And I [57:07] think the Yahoo solution was very, yeah, [57:09] yeah, search just has to be good enough. [57:11] The curation is sort of the thing that [57:13] matters. We're a media company. you [57:14] having enough human editors to cover all [57:16] the big categories, but the business is [57:19] showing banner ads. They may or may not [57:21] be relevant to whatever page you happen [57:23] to be looking at right now, and we're [57:25] effectively a media company that has [57:26] search just in case. Yes. Okay. So, [57:29] that's what's going on at Yahoo and all [57:31] the portals. The opinion of Larry and [57:34] Sergey [57:35] is a we don't want to do a homepage like [57:39] that. We don't want to clutter it up. [57:40] Our whole point is to help people find [57:43] what they want. Which of course raises [57:44] the question, well, what's the business [57:46] then? Because if you're not keeping [57:49] people on site to see your banner ads, [57:50] the only moment that you really have is [57:53] on the search box page and on the search [57:55] results page. And they were extremely [57:58] against, well, really ads generally. [58:00] They didn't think it was good for users, [58:02] but they were just against especially [58:04] banner ads. And there's this sort of [58:06] scary thing, which doesn't seem scary [58:08] now because we know how it played out, [58:09] but just imagine trying to evaluate this [58:11] company. It's growing like wildfire. [58:14] Everyone's using it. There's one known [58:17] business model for this entire sector. [58:19] It's not a great one, but it is known. [58:21] And these guys are dead set against [58:22] using it. But on the other hand, let me [58:25] pitch it to you a different way. These [58:27] guys are building the front door to the [58:31] internet, which was just growing 700% [58:33] year-over-year. So, isn't that going to [58:35] be really valuable? Yes, but we don't [58:37] know how yet. But the problem is [58:40] actually even more dire than what you're [58:42] saying. As usage is growing, they need [58:45] more infrastructure, but they're not [58:47] making any money, right? And for each [58:50] piece of this, you need more [58:51] infrastructure. You need the crawler to [58:54] go and crawl the whole web and store, [58:57] you know, not entire web pages, but [58:59] little pieces of web pages that you can [59:02] reference from your index. You need the [59:03] index itself. You need to serve the web [59:06] pages up for when people are doing the [59:08] searches. There's a bunch of components [59:10] of this infrastructure that all need to [59:12] scale and they all need to scale [59:13] differently. Yep. Which brings us to [59:16] really what is the second big reason why [59:20] Google worked so well and became the [59:22] Google we all know today. One is [59:24] accurate, relevant, fast search results [59:26] and page rank and everything we've been [59:27] covering. Two though is the [59:30] infrastructure to actually make this [59:32] whole thing work and scale efficiently. [59:34] Yes. So right after they raise the angel [59:36] round, Larry and Sergey go out and they [59:39] recruit just like unbelievable top tier [59:44] engineers and computer scientists to [59:46] come rewrite the code and work on this [59:48] infrastructure problem. So pretty [59:50] quickly they get Ers Hoa and then Jeff [59:52] Dean who are just these absolute [59:54] legends. They are both still at Google [59:56] today. ERS is now a fellow, but he ran [59:59] all of Google's infrastructure from 1999 [60:01] until 2023. Before joining, he'd done [60:04] his PhD at Stanford and he was a [60:05] professor at UCSB. He'd also written the [60:10] primary Java virtual machine that Sun [60:13] used as like the official Java virtual [60:15] machine. Oh wow. And Larry and Sergey [60:18] recruit him out of academia to come join [60:21] as employee number eight. and his [60:22] initial job title was search engine [60:25] mechanic because quote everything was [60:27] broken. [60:29] So that's and he builds all this [60:32] incredible infrastructure. Jeff Dean who [60:34] they also recruit around the same time [60:36] from a senior engineer at deck. Yes. [60:39] Yes. And Jeff is basically like Google's [60:41] Dave Cutler. So today Jeff runs AI at [60:45] Google. He also implemented the first [60:47] version of AdWords, built AdSense, [60:50] rewrote the core search pipeline five [60:52] times, co-invented and implemented Big [60:54] Table, Map Produce, TensorFlow, and [60:56] Gemini. [60:59] He actually keeps his resume up to date [61:00] online. We'll link to it in the show [61:02] notes. It's incredible. We're bearing a [61:04] little bit of a lead here. We we spoke [61:05] with Jeff to prep for this episode and I [61:06] watched a handful of talks he's given. [61:08] Uh delightful human and God, what a [61:12] great engineer. Just generational [61:14] talent. But this is like that early [61:16] nucleus of engineers that Google [61:18] recruited. It's amazing that they [61:21] attracted them because prospects were [61:23] not good that all of this would work and [61:24] scale. And it was only because of these [61:26] guys that it did. Well, and here's the [61:28] crazy thing. Later, there's an easy [61:30] point to make, which is Google got to [61:32] hoover up all the best talent because [61:34] they were a solid business after the dot [61:37] crash. But this in 989, we're in the go- [61:41] go times. the dot bubble hadn't burst [61:44] yet and Larry and Sergey managed to [61:46] recruit this talent. I think this is [61:48] like a history turns on a knife point or [61:50] like a make or break the company thing. [61:52] The fact that they were able to get [61:53] these guys in a hot talent market really [61:57] speaks to Larry and Sergey's vision, the [62:00] excitement around the idea, how novel [62:03] their approach was, everything. Yep. And [62:06] part of the reason why this talent was [62:09] attracted to Google, sure some of it was [62:12] like, "Oh, the product's really good and [62:13] people are using it and so that makes [62:14] the company interesting." The other part [62:16] of it though was that the technical [62:18] challenges and the architecture coming [62:21] out of Stanford was super unique and [62:25] novel. This was a really really [62:27] interesting thing to work on. And why [62:29] was that? So the Google index that they [62:33] needed to build and operate on for the [62:36] search engine for page rank to work was [62:39] so much bigger than any other index out [62:42] there. Google needed like the entire [62:44] page to compute all the rankings and [62:47] find the links, find the back links. [62:49] They needed to architect Google with [62:51] this huge distributed computing system. [62:55] So the index was so big that it wouldn't [62:58] fit on a single machine or a single [63:01] server no matter how big or how [63:03] expensive. So what they do to store the [63:06] index and to operate on it with this [63:08] distributed file system is they break [63:11] the giant index into tons and tons and [63:14] tons of little chunks they're called of [63:17] individual 64 megabyte files. Small [63:20] files, tractable files, and they get [63:24] stored on lots and lots of different [63:26] discs and lots of different machines and [63:27] lots of different servers and ultimately [63:29] different data centers all over the [63:31] world. And then there's a separate [63:34] server that keeps a master mapping of [63:36] all the chunks like where the chunks [63:38] physically are. And so when a query [63:40] comes in and needs to operate on the [63:43] index data, the master server just [63:46] returns only the chunks that it needs, [63:48] not the whole index. And that makes the [63:50] whole thing possible. So basically that [63:52] one server you're talking about can kind [63:55] of just say, "Oh, all the chunks are [63:57] here on all these different machines [63:58] that are distributed throughout my data [64:00] center. just look at those chunks and [64:02] that way it can kind of just pull and in [64:04] a parallel way pull from all those [64:06] different chunks concurrently. Yeah. And [64:09] I think that's even abstracted like from [64:11] a compute perspective they see the [64:13] master map they feel like they have [64:15] access to the whole file but then what's [64:19] actually getting returned to them to [64:20] operate on is only just the chunk data [64:22] that they need. Hm. So, Google was sort [64:25] of forced to do distributed computing [64:27] because their index file was too large [64:28] to store on any one machine, no matter [64:31] how big or fancy it could be. Yep, I [64:33] think that's right. Which sort of [64:34] enables the whole thing in the first [64:36] place and is technically extremely [64:37] interesting. But now the physical [64:39] infrastructure side, all right, because [64:42] you have all these chunks and they can [64:44] live anywhere and Larry and Sergey [64:48] already had to grab commodity hardware, [64:50] you know, hard drives and motherboards [64:52] directly back at Stanford. Well, Z comes [64:54] in and he's like, whoa, we can just keep [64:56] going with this. Let's keep using cheap [64:59] commodity components and hardware. And [65:02] yeah, they'll suck and they'll fail a [65:04] lot and things will burn out, but that's [65:07] okay. because we've got this distributed [65:09] file system, we'll just replicate [65:11] everything like three or five times, [65:13] right? And we can cleverly design [65:15] software to account for the fact that we [65:18] have commodity hardware or commodity [65:19] RAM. These systems that were not [65:22] assembled with the notion of being [65:24] enterprisegrade, we can sort of design [65:27] Google with the idea to take into [65:29] account the fact that the hardware is [65:31] not enterprise grade. And that means we [65:32] can get cheaper hardware and run in a [65:35] distributed computing way. And frankly, [65:38] I think this makes it interesting to a [65:40] lot of engineers who kind of want to [65:41] work on hard problems. How do I design a [65:44] system when I can't count on a whole [65:46] bunch of stuff from the underlying [65:47] hardware that I would get to count on if [65:50] it was a fancy deck server? So, I read [65:54] that industry average server hardware [65:57] failure rate at the time was around like [65:59] 3 to 4% per year. Google's hardware [66:03] failure was over 10% per year. [66:06] But the whole system was designed that [66:08] it didn't matter. It was all just [66:09] replicated. Super interesting. So this [66:12] keeps scaling up and up and up over the [66:14] years with Google pretty quickly. Maybe [66:19] even while Google is still a private [66:20] company, they technically become the [66:23] world's largest computer manufacturer. [66:26] Oh wow. Because they're not buying fully [66:29] baked servers. They're just buying [66:31] components and assembling them into this [66:33] sea of components. proto data center in [66:36] their data center and then data centers [66:39] and their early data centers they're [66:40] never really putting them in PC housing [66:44] right yes so these early you know quote [66:46] unquote machines they're building [66:47] they're not even putting PC cases on [66:50] them they just mount the motherboards [66:53] directly on corkboard and then they put [66:56] like the RAM in there and they put hard [66:57] drives in there and then they just stuff [66:59] them in their data center racks the [67:01] photos of these early Google quote [67:04] unquote Server racks are crazy because [67:07] the way that their agreements worked in [67:10] the colllocated data center facility is [67:12] they would lease by square footage, not [67:15] by energy consumed, not by number of [67:18] mach by square footage. And so when you [67:21] give a computer scientist a constraint, [67:22] they will optimize for it. And the goal [67:24] is how much of Google can I power in [67:28] this square footage? And the way that [67:30] you optimize around that is well [67:32] incredible density of hardware. So we're [67:35] not putting cases on these computers. [67:36] We're putting corkboards in. And imagine [67:39] just a sheet of cork which is an [67:42] insulator. So you know that these [67:44] electrical components that you don't [67:45] want to conduct between each other are [67:47] not going to conduct between each other. [67:49] And you just stuff a server rack full of [67:51] corkboards with all this commodity [67:54] hardware sort of strewn about it. And it [67:56] looks unbelievably messy. It's extremely [67:59] economical and then you just kind of [68:01] handle it all in software. And the net [68:04] of this is that Google can scale period, [68:08] but also can scale way more cheaply as [68:11] search traffic rises and as the index [68:13] keeps growing and getting bigger than [68:15] anyone else out there on the market. So [68:17] once the business model kicks in, this [68:20] is why Google search has like a 87% [68:23] gross margin on it. Yep. There's this [68:26] incredible story of Google's first data [68:29] center was a collocation data center [68:32] facility kind of a shared uh physical [68:35] space in Santa Clara called Exodus and [68:37] the data center cage the space that [68:39] Google had allocated was right next to [68:43] the cage for in which was a competing [68:46] search engine out there that we'll talk [68:47] about in a minute and Google folks talk [68:50] about the in cage had all these gleaming [68:53] sun machines and lots space and lots of [68:56] airflow and all this incredible cable [68:59] management and then you had this like [69:01] Frankenstein Google thing next to it. [69:04] And to your point, Ben, they were only [69:06] paying by square footage. They weren't [69:08] paying for power. So, they were sucking [69:11] up all the power of the data center. And [69:14] I heard a story that they actually at [69:15] one point may or may not have stolen a [69:17] power circuit from the uh Inktomy cage [69:20] next door. Borrowed. Borrowed. Borrowed. [69:22] Borrowed. Yes. So one fun illustration [69:25] of what does it mean to be on commodity [69:26] hardware versus enterprisegrade [69:28] hardware. Jeff Dean shared a fun story [69:30] with us that on enterprisegrade hardware [69:32] you would have something in RAM which is [69:34] called a par bit. And in consumer grade [69:37] hardware you don't and what is a parody [69:39] bit? A par bit adds one extra bit to [69:43] memory that is basically for error [69:47] checking. It looks at the rest of the [69:48] data in the bite and if it's even it'll [69:51] set the par bit to one and if it's odd [69:53] it'll set the par bit to zero. The con [69:55] of this is now you need an extra bit. So [69:58] it makes the overall machine more [69:59] expensive because you're losing 1/8 of [70:02] the RAM to this par checking. But the [70:05] benefit is you know that nothing ever [70:07] got corrupted because the likelihood [70:09] that one of your bits got flipped and it [70:12] also flipped your parody bit is very low [70:15] because you can kind of check and see [70:17] wait it's supposed to be an odd number [70:19] according to the par bit but it's an [70:20] even number so you know likely something [70:22] went wrong and when I say something went [70:24] wrong this is from like random radiation [70:27] that is just flying around the universe [70:29] at any given time stuff goes wrong all [70:30] the time well because Google is using [70:32] this commodity hardware they then have [70:35] to do these crazy things in software and [70:38] build all these layers themselves to say [70:40] we're running on crap hardware. We don't [70:42] know for sure that the value in memory [70:45] is correct. Can we have a second way of [70:47] verifying that it's correct? So it's [70:49] that sort of I don't know cool software [70:51] engineering but also another layer of [70:54] systems that you have to build when [70:55] you're on commodity hardware. So the net [70:58] of these constraints and the incredible [71:00] technical team Google has is that they [71:02] design everything from the ground up. [71:05] The computing systems, the file systems, [71:07] the data centers, the racks, the [71:08] hardware, everything. And they built [71:12] stuff like GFS, the Google file system, [71:14] map reduce. Yahoo would eventually feel [71:17] like they needed to copy map produce to [71:19] be competitive and they would open [71:20] source that as Hadoop. So if people know [71:22] Apache Hadoop, that is a Yahoo copy of [71:24] Google's map produce. Yep. Of course, [71:27] shepherded and stewarded mostly by [71:29] people outside of Yahoo eventually, but [71:30] that's where it came from. Yep. And then [71:33] ultimately, because Google is building [71:34] their own hardware, racks, data centers, [71:38] and they can do it cheaply, they put [71:40] data centers all over the world. And [71:41] then that means they can deliver search [71:44] results and ad results instantly to [71:47] users all over the globe. Yeah. So this [71:50] speed really starts to become a bragging [71:52] point for Google where whenever you do a [71:54] query, it'll show you how long the query [71:56] took. It's usually like a quarter [71:58] second. And they used to brag about the [72:00] index size. Now they say I'm returning a [72:02] gajillion results to you. But this was a [72:04] really big flex for a long time is we [72:07] searched a huge index, billions of [72:09] pages. We found a huge number of results [72:12] and we did it really fast and we're [72:14] going to show all those numbers to you [72:15] because a we're engineers and we're [72:17] awesome, but b we know it's the best [72:20] stats that anyone out there could report [72:22] to you. Yep. It's a stake in the ground. [72:24] So all this grows out of the constraints [72:26] of like they don't have any money first [72:29] at Stanford and then this little angel [72:31] around they raised. They don't have a [72:32] way to generate any money and yeah they [72:35] don't have any way to generate any [72:36] money. So we cannot overstate how [72:39] important Google's infrastructure [72:41] innovations were. All this comes out of [72:44] these constraints that Google the [72:46] company has. But none of this would have [72:49] mattered if they didn't figure out the [72:51] business model. But before we tell the [72:53] story of the building of the best [72:57] business model of all time, now is the [72:59] perfect time to thank a new friend of [73:02] the show, Anthropic, and their AI [73:04] assistant, Claude. Yes. So, as we were [73:07] doing our research for this episode, [73:08] looking through decades of interviews, [73:11] SEC filings, technical papers, Claude [73:14] was an awesome research partner. I [73:16] actually had this crazy thing happen. [73:17] I've never had this happen before. I was [73:19] asking Claude a bunch of questions about [73:20] early Google financials, and it kept [73:22] giving me numbers that I thought were [73:24] wrong, but the numbers it was giving me [73:26] were consistent. So, I went back and I [73:27] checked against the numbers in the book [73:29] I'd been using. It turns out that Google [73:32] actually restated their early [73:33] financials, and Claude was all over it. [73:35] So Claude actually saved me from making [73:37] a somewhat meaningful error later in [73:39] this episode. I've never had AI do that [73:40] before. Yeah. Now I wish we had Claude [73:42] for all what 200 of our other episodes. [73:45] Yeah. I also played around with Claude's [73:48] extended thinking mode in my research, [73:50] especially analyzing Google's S1, which [73:52] gave me a bunch of great insights that [73:54] we'll share later in the episode. Yep. [73:56] This type of research is where Claude [73:58] really shines compared to well [73:59] traditional web search. Claude can [74:01] analyze hundreds of pages of documents [74:03] simultaneously. And now with its [74:05] connection to Google Workspace, it can [74:06] pull insights from files in Google [74:08] Drive, check Gmail, and search the [74:10] latest information, all while [74:11] maintaining context about what you're [74:13] actually trying to understand. And [74:15] Claude is built by Anthropic, so you [74:17] know it has a focus on being helpful, [74:18] harmless, and honest. When you're doing [74:20] serious research, like we do here at [74:22] Acquired, you need an AI you can trust [74:25] to be accurate and not just impressive. [74:27] Yep. Plus, Claude's new Opus 4 is widely [74:30] regarded as the world's best coding [74:32] model. So, as you can imagine, we talked [74:33] to a lot of engineers doing what we do, [74:35] and we keep hearing over and over again [74:37] from developers that Claude is their [74:39] favorite model. Acquired listeners can [74:40] get halfp price on Claude Pro for 3 [74:42] months using our link at [74:44] claude.ai/acquired. [74:46] That's cl aud.ai/acquired [74:49] or click the link in the show notes. [74:51] Yep, Claude thinks with you, not for [74:53] you. And trust us, when you're trying to [74:55] untangle 25 years of Google history, [74:57] that makes all the difference. Our [74:59] thanks to Claude and Anthropic. All [75:00] right, David. So going from having no [75:03] business to the greatest business model [75:06] humankind has ever discovered. Not [75:09] exactly a straight line, huh? Uh, no, [75:12] not at all. So how does it all start? So [75:15] early 1999, even despite the incredible [75:19] infrastructure work and being able to [75:21] scale cheaply, Google's still running [75:24] out of money from the angel round. and [75:27] they hire a recent Stanford undergrad [75:30] named Salar Commonar who joins Google as [75:33] employee number nine. And like many [75:37] Stanford students at the time, he [75:40] started using the search product while [75:41] he was an undergrad, was blown away, and [75:44] he's like, I got to go work for this [75:45] company. He basically bangs down [75:47] Google's door, tries to get hired, and [75:49] finally they're like, okay, okay, come [75:50] on in. And so the first thing that Larry [75:52] and Sergey give to him to do is kind of, [75:56] you might argue, the most important [75:57] thing in the company. They're like, [75:58] "Well, we're running out of money, so we [76:00] need to go raise venture capital. We [76:02] don't want to write the business plan in [76:04] the pitch deck. You write the business [76:05] plan in the pitch deck." [76:08] So Sally goes off and he writes the [76:11] pitch deck for the Google Series A, a [76:13] collaboration with Larry and Sergey, and [76:15] they come up with a threepronged [76:18] business model that they're going to [76:20] present to VCs, three ways that they're [76:22] going to make revenue, and it's handwavy [76:24] as hell. Yes. All right. So number one, [76:27] the biggest revenue driver projected [76:30] going forward, you know, the main Google [76:32] business innovative new business model [76:34] that they're going to pursue here. They [76:36] are going to sell Google search [76:39] technology to enterprises [76:43] so that companies can use the same [76:47] amazing consumer Google search [76:49] technology to search their own documents [76:53] and internets. [76:56] This is the business plan for Google for [76:58] the series A. You know what this kind of [77:00] feels like to me? It feels like they're [77:01] saying Google.com is so precious and [77:04] amazing and special. We don't want to [77:06] risk it by having to make money on it. [77:09] So, can we make money doing something [77:11] else with our technology? Yes. That will [77:13] fund what we really want to do, which is [77:16] Google.com. Now, it's pretty funny to [77:19] talk about this. Now, in retrospect, [77:21] they did have a couple reasons why they [77:24] thought this might work. Number one was [77:28] all the way back at Stamford, Back Rub, [77:31] and then Google had actually been used [77:33] for this use case. you could use Google [77:35] to search internal Stanford internet [77:38] stuff and people did. It was like a [77:39] great experience for Stanford students. [77:42] Also, before the series A somehow Larry [77:45] and Sergey had managed to actually sell [77:48] one of these deals to the company Red [77:50] Hat. That's right. That was their first [77:51] revenue, right? Was Red Hat. Yeah. The [77:53] open source Linux company. They sold [77:55] this enterprise search deal to Red Hat [77:58] for $20,000. They're like, "Oh, great. [78:01] There's a market here." So that was [78:02] going to be the main business driver. [78:06] And then there were going to be two [78:07] other business lines too in the company. [78:10] One was going to be, well, sure, okay, [78:12] VCs, you know, you make us, we'll sell [78:15] ads, CPM banner ads, the same way [78:17] everybody else does. We're not going to [78:19] like it, but we'll put it in the [78:20] business plan. And it seems like they [78:22] didn't think through it any more than [78:23] that, cuz I couldn't find anything about [78:26] is that going to appear on the search [78:27] results page? Is that going to appear on [78:29] google.com next to the search box? It [78:32] seems like it was never real enough to [78:34] actually have a plan for it. And indeed, [78:37] the series A pitch deck and business [78:39] plan like was intentionally vague. I [78:41] think part of the reason Larry and [78:42] Sergey were like, "Okay, seller, new [78:43] guy, you go do this." Is like they [78:45] didn't actually want to tell VCs that [78:47] much. Yeah. So, that was number two. And [78:49] then number three was that they were [78:51] going to license Google organic search [78:54] results to portals and directories as [78:57] essentially OEM search to backfill [79:00] results like we were talking about with [79:01] Yahoo. Other search engines were doing [79:03] this. Inkto had gotten started at this [79:06] point in time. Inctomy was the next door [79:07] neighbor at the data center Exodus in [79:10] Santa Clara. And they built a sizable [79:12] business selling white labeled organic [79:14] search results to other portals. [79:16] Exactly. This was Intim's whole business [79:18] is they just sold organic search results [79:20] white label to other portals. But [79:22] there's five big customers and 50 total [79:25] customers out there for this business. [79:27] Yep. So this is the business plan. This [79:28] is the pitch deck. They go out. Remember [79:31] we're in spring 1999 here. So even [79:35] though this is a little hairrained, it's [79:36] still the dot bubble. Money is still [79:39] flowing. There's a great Michael Moritz [79:41] quote in Steven Levy's book talking [79:43] about this particular moment in time. He [79:44] says, "Nobody's feet were on the [79:46] ground." This is a very Sir Michael way [79:48] of putting things. Deliciously, Michael [79:50] quote. Yes. And Google's got all this [79:53] usage and engagement and growth numbers [79:55] that, hey, internet companies trade on [79:57] eyeballs, so of course this is going to [79:58] be a hot deal. Famously, Kleiner and [80:01] Sequoia end up splitting the deal and [80:04] Michael Morates and John Door, the two [80:07] most legendary VCs in the world, they [80:09] team up, they join forces, they split [80:11] the deal and they both join the board of [80:13] Google at the Series A, which is unheard [80:16] of. The fact that Larry and Sergey were [80:18] able to say, "You both only get 12 and a [80:22] half% of this company and you have to do [80:23] it together." They both must have [80:26] really, really, really wanted to do the [80:27] deal. That is true. And I love that even [80:30] today, even you have that opinion. We [80:34] heard from folks in the research that [80:36] this really was like a Google PR master [80:39] stroke to seed this narrative. Oh, [80:43] well, they held a press conference in [80:44] person with both Sir Michael and John [80:47] Dur. There it's the first Google press [80:48] conference. Larry and Sergey are there [80:50] in Google branded shirts. Yes, they made [80:53] a big deal about this. The reality is [80:56] Sequoia and Kleiner split tons of deals. [80:59] This was not the first one. It may have [81:01] been the first one that Michael and John [81:04] split together, but like they had been [81:06] on boards together before. Maybe they've [81:07] done one round or the other and Sequoa [81:09] and Kleiner split deals all the time. [81:10] But hey, it was the dot era and [81:12] everybody needed a PR strategy and that [81:15] one worked well. Fascinating. But the [81:18] point is this was a hot deal. There's [81:20] all sorts of stories out there about [81:21] other investors coming into the round or [81:23] trying to get in or trying to get in [81:25] later. Yeah, I think in the middle of [81:27] negotiations they got another term sheet [81:29] at $150 million valuation instead of a [81:32] hundred. But I think they were sort of [81:34] already pot committed to Sequoa and [81:36] Kleiner Perkins. And I actually don't [81:38] know who the 150 came from, but I do [81:40] know it was someone who Rahm Sriram set [81:43] up the meeting with. Interesting. And [81:46] remember, Larry's older brother had [81:48] co-founded e-groups, which Meritz had [81:51] funded at this point. So, like, they [81:53] knew they wanted to go with Sequoia and [81:55] Kleiner. That was the goal all along. [81:56] Yep. Regardless, the round gets done. [82:00] $25 million total raise at a $100 [82:04] million [82:06] post money valuation. [82:08] A $und00 million valuation was genuinely [82:11] wild for the time. Yeah. Even in the [82:14] heyday of the.com craziness for a series [82:17] A at a 100 post that was newsworthy. [82:21] So funny thinking about this today. I [82:23] know today it's so quaint. It's so [82:25] quaint. The comp in today's world is [82:27] imagine reading a headline that a series [82:30] A after a company just had a few angel [82:32] investors got done at a multi-billion [82:34] dollar valuation. That's kind of the way [82:37] it would have felt in tech at the time. [82:39] Yes, totally. But despite that and [82:42] despite all the hype around the series [82:43] A, there's still kind of an urgent [82:46] imperative to make revenue. Yes, there's [82:49] 25 million in the bank now, but you've [82:50] got VCs involved. And the playbook back [82:53] in the do days was invest in the [82:56] company, get quick revenue, go public. [82:59] Yep. So, there's a fire lit under Google [83:01] to figure out the business model. Right [83:03] around the same time as the series A [83:05] happens, Larry and Sergey meet a guy [83:08] from Netscape named Omid Cordistani. And [83:12] the first time they meet, I think Omid [83:14] is thinking about it in the context of, [83:16] oh, I'm wearing my Netscape hat. I'll [83:19] evaluate, is there some partnership [83:21] here? And I think he quickly gets the [83:23] sense maybe, but Netscape just got [83:27] bought by AOL. It's getting a lot less [83:30] fun here. And Omid was the VP of sales [83:32] in Bisdev at Netscape. Yes. So he's very [83:34] familiar with building an internet-based [83:36] business. And what these Google guys are [83:39] doing is very interesting. Yeah. And of [83:41] course, Larry and Sergey as the great [83:43] recruiters that they are like, "Hey, why [83:45] don't you come work here?" So Omid joins [83:47] Google essentially as chief revenue [83:50] officer and he's tasked with, "Okay, [83:52] take this business plan, take these [83:54] three areas and make them a reality." [83:56] And we should say too, Omid is an [83:58] awesome guy. We talked to him in [83:59] research. Yeah. So he goes out and of [84:02] course the first the innovative business [84:03] model that they pitch the VCs you know [84:05] number one the enterprise search [84:06] business model he goes out and starts [84:08] trying to sell it but as you might [84:11] imagine especially at the time there [84:13] wasn't a lot of uh customer pull shall [84:15] we say for this. Yes. So for the first [84:19] six plus months of the company, kind of [84:21] the rest of 1999 after the venture [84:25] funding, things are not looking good on [84:28] the revenue front. David, do you [84:31] remember, this is way back in Acquired [84:32] History, what Doug Leone told us in [84:36] February of 2020 when we were recording [84:38] our episode with him? Yes. Yes, I know [84:40] exactly what you're going to say and [84:41] I've got some more flavor on that quote. [84:43] You give the quote. Okay. So the quote [84:46] is from Sir Michael Moritz. He comes to [84:49] Doug and they're running Sequoia Capital [84:51] together at the time and he says, "Doug, [84:54] we've never paid so much for so little." [84:59] And uh that's the lore. I got a little [85:02] more behind the scenes flavor on the [85:04] quote. [85:06] Apparently it was not Sir Michael who [85:08] said it first. He might have just been [85:09] repeating it back. Apparently was it [85:11] John Door? It was Venode. Oh, in the [85:14] Kleiner partnership. That's what I [85:15] heard. Oh, this is like the hot potato [85:18] of quotes. Nobody wants to actually take [85:19] credit for this. Either way, though, the [85:21] sentiment is right. You can understand [85:23] why Kleiner and Sequoia would feel this [85:26] way. They kind of have egg on their [85:28] faces. They just paid a $100 million [85:31] post for a series A company with no [85:34] revenue. The revenue is not [85:36] materializing. We're now into the year [85:38] 2000. The bubble is starting to burst [85:41] and Google still basically has no [85:44] business. Yeah, no business. But growing [85:48] market share, fervent loving fandom [85:52] among the people using it, providing [85:54] real value to people. There's got to be [85:56] something here. Yes. Exactly. So the [85:59] revenue imperative is becoming well [86:02] imperative. More of an imperative, shall [86:03] we say? And Omid's a smart guy. He's [86:05] like, I'm not going to just keep banging [86:07] my head against enterprises here. Like, [86:09] we're going to pursue the other two [86:10] business lines that are obvious. You [86:12] know, who knows how big they'll be, but [86:14] at least we'll make some money. So, set [86:15] up just regular ads, same way as [86:17] everybody else does it. So, Omid goes [86:20] and hires Tim Armstrong in New York to [86:22] set up an ad sales force. And Google [86:24] does start selling ads at the top of [86:27] search result pages. And what do these [86:29] ads look like, David? So importantly, [86:32] Larry and Sergey insisted that, okay, if [86:35] we have to have ads on here, they need [86:38] to be text only. We can't serve images [86:41] and banner ads like everybody else does [86:43] because that'll slow down the page. Yes. [86:45] It's like they always talk about it for [86:47] taste, which is true, but yes, it's a [86:49] performance thing. Yes, it's a page [86:51] performance thing. So, there these great [86:53] stories about Tim in New York and Omid [86:56] and the ad salesforce that they're [86:57] building up at Google. They're going to [86:59] ad agencies. They're going to [87:01] advertisers directly. They're trying to [87:02] sell these ads and like um no images, [87:05] just text. Trust us, it's going to work. [87:07] Not very exciting to these Madison [87:09] Avenue guys. How are they paying for [87:12] them right now? Still CPM. So, you buy a [87:18] keyword [87:20] and then you're promised that you're [87:22] going to be an ad that appears on the [87:25] page whenever that keyword is searched [87:28] and you're going to pay per thousand [87:30] impressions of that keyword. Is that [87:33] right? Yes. Okay. Not self-s serve. No [87:37] web tools for this. This is like [87:39] negotiated over the phone and then they [87:41] manually hand it into Google that when [87:45] this keyword is searched, you need to [87:47] display a text ad for this person and [87:50] track how many times you display it [87:51] because we then need to invoice them for [87:55] how many times the page loads. Yes. Not [87:58] even on the phone. That would be really [88:00] uh technologically advanced for Madison [88:02] Avenue at the time. All ad insertions [88:05] were done by fax [88:07] at this time. So Google had to install [88:09] fax machines at its headquarters to take [88:11] these insertion orders for these ads [88:13] that they were selling. Awesome. Now [88:16] what was the pitch though to Madison [88:18] Avenue about why this would work? Intent [88:20] baby. Exactly. [88:23] So the very first project that Jeff Dean [88:25] did when he came over from deck is Larry [88:28] and Sergey and Z told him, "All right, [88:31] the VCs say we got to sell ads. go [88:33] figure out the tech to serve ads on [88:37] google.com, but don't do anything to [88:39] degrade search or user experience. And [88:41] so Jeff works with Marissa Mayer, who [88:43] just joined from Stanford undergrad, and [88:44] they're like, "Okay, like fine, this is [88:47] going to have to just be text. What can [88:49] we do as a test to see if we can [88:52] engineer something that'll work with [88:54] text ads? We could scale, run it against [88:57] a bunch of queries." Well, what about [88:59] Amazon affiliate links? [89:01] We know a guy at Amazon. We know a guy [89:04] at Amazon who happens to own a good [89:06] chunk of this company. So, Google goes [89:09] and signs up as an Amazon affiliate, [89:12] which how crazy is this? The Google [89:14] business model was validated like doing [89:17] customer development, you know, startup [89:19] idea validation using Amazon affiliates [89:22] as the mechanism. Yes, Jeff Dean codes [89:25] it up so that dynamically as users are [89:29] searching, if there's a query that is [89:31] related at all to any book in the Amazon [89:35] library catalog, Google will dynamically [89:38] generate a text ad saying, "Go buy this [89:41] book at Amazon, insert an Amazon [89:43] affiliate link and drive traffic over to [89:46] Amazon." And the amazing thing is, of [89:49] course, it actually works. Now, whatever [89:52] the Amazon affiliate commission, you [89:54] know, if it's 4% or 5% of the revenue of [89:56] a $10 book, obviously is not going to [89:58] change Google's fortune in the amount of [90:01] money that they'll generate from this. [90:03] But it's a test. It's a test and it's [90:05] proof that they can then take to [90:07] advertisers, we can capture intent and [90:10] we can send highly monetizing traffic to [90:14] you based on the keywords that you are [90:16] buying. And so, if you think about the [90:17] funnel steps, there's the impression of [90:19] an ad. Hey, they saw the ad. Then two, [90:22] there's the click. They click through [90:23] the page. Then three, there's the onpage [90:25] conversion. It would be one thing to [90:28] just test click-through rate, which [90:30] would have shown a great result here [90:33] because click-through rates on search [90:34] ads are higher than ads that are just [90:36] randomly around the internet because [90:39] someone is intending to buy something. [90:41] They have high intent. That's a great [90:43] place to show an ad. The click-through [90:46] rate is going to be higher. But what [90:47] they also know because it was an Amazon [90:49] affiliate links is they know the [90:51] downfunnel number too. They know [90:53] conversion is actually higher from this [90:55] traffic because we got to know how many [90:58] books on Amazon were sold from the [91:01] number of impressions that we served. So [91:02] they can say this intentbased ad system [91:06] has high click-through rate and high [91:08] conversion. Yeah, they're just text ads, [91:11] but you're going to like the numbers. [91:12] This is ultimately a math problem. Yeah, [91:15] it was an absolutely [91:17] brilliant [91:19] test and bootstrap of the first step of [91:21] the Google ad model. Yep. So, that's [91:25] sort of gen one of the Google Ads [91:27] business. And they get it set up, it's [91:29] going, it's making some money, they're [91:32] winning some clients on Madison Avenue, [91:34] like good, great. The more interesting [91:38] piece for the next year, year and a half [91:41] is the OEM search portal deals. [91:44] And there's some big deals to come. [91:46] Netscape, Yahoo, AOL, this is white [91:50] labeling Google search to be the search [91:53] powering other places search activity [91:56] that has tons of traffic, which oh by [91:59] the way is also going to train huge [92:01] portions of the internet to use Google [92:04] search. Yes, it is. And specifically, [92:07] what were the portal deals at this [92:09] period of time before Google had a real [92:13] functioning paid search business? What a [92:16] portal deal represented was just letting [92:19] a portal use Google search to power [92:21] their organic search and in exchange [92:23] just getting paid a fee for that. Yep. [92:25] It was the Inctomy business model. It [92:27] was exactly that. We're selling our [92:29] search results for you to use on a [92:31] thirdparty page and it's effectively [92:33] like a B2B supplier. They're a vendor to [92:36] a portal more or less. Yes. It turns out [92:39] though that we're still in the era of [92:41] the internet where portals are pretty [92:43] big. Ton of traffic. And they're not [92:46] just any old vendor. They're a vendor [92:48] that at the bottom of every page says [92:50] powered by Google. Yes. So pretty [92:53] quickly after Omid joins from Netscape [92:56] back in 1999, [92:58] he goes back to his old colleagues at [93:00] Netscape AOL and gets essentially like a [93:04] proof of concept deal done with them. [93:07] And that's for Google to backfill [93:09] organic search results on Netscape's own [93:12] directory service that they just [93:14] launched to compete with Yahoo. So, it's [93:16] interesting. We always think about [93:17] Netscape, the browser, but this was [93:19] presumably like the Netscape homepage. [93:21] Whenever you opened up, you know, it [93:23] would go to Netscape.com or something [93:25] and there'd be all these Netscape [93:26] services there available, one of which [93:28] was search. Yep. And I presume having [93:31] been acquired by AOL, this was now more [93:33] of a strategic priority for Netscape [93:35] because this is AOL's like whole [93:37] business model at this point besides [93:38] the, you know, monthly dialup fees, [93:40] right? Worth remembering AOL way, way [93:42] bigger at this point. Tens of millions [93:45] of people using AOL, way fewer going [93:49] through netscape.com to deliver traffic [93:51] to this Google search. This is sort of [93:53] the small part of the organization [93:55] they're working with right now. Yep. [93:56] Exactly. But relative to how small [94:01] Google actually is at this point in [94:03] time, it's huge. still big enough that [94:06] when they flip the switch and Google on [94:09] Netscape goes live with Google powering, [94:11] you know, organic search results, [94:13] there's so much traffic that it blows [94:16] out Google's infrastructure. Well, it [94:18] comes close. Basically, they're watching [94:20] the analytics like a hawk. OM gets an [94:23] urgent call from Sergey saying the [94:24] traffic is about to tip over. And this [94:27] is potentially company killing cuz this [94:29] is their big strategic priority. If they [94:32] prove that they are untrustworthy to [94:33] Netscape and can't deliver, then how are [94:36] they going to keep Netscape's business, [94:37] let alone get any other portal deals or [94:40] be able to go sell to enterprises that [94:42] they're thinking they're going to do at [94:43] this point in time, right? So, they [94:44] cannot tip over. And so, they have this [94:47] pretty tough decision to make. But [94:49] actually, it's not even a decision at [94:50] all. This is obvious. We are shutting [94:53] off google.com for today and we are [94:56] going to prioritize all traffic from [94:58] Netscape for our servers until we can [95:01] stand up more machines. Yeah, just think [95:03] about this for a minute. Think about [95:05] everything that Google is today. Never [95:08] goes down. Universally available [95:10] basically everywhere in the world on [95:12] every device. It's freaking Google. In [95:14] 1999, [95:16] they shut Google down so that they could [95:18] serve Netscape's users. Yes. I mean, [95:22] look, the revenue is very material [95:24] coming from this and the reputational [95:26] impact is very material. Like I said, it [95:28] sounds like a hard decision. It's [95:29] actually not a decision at all. It also [95:31] ends up being the right strategic [95:33] decision because of what you mentioned a [95:36] minute ago of the powered by Google logo [95:38] at the bottom. Sure, you shut off [95:39] google.com for a day and your own Google [95:41] users of course don't like that, but [95:44] you're training millions of new Google [95:46] users who are going to see powered by [95:47] Google at the bottom. And they got [95:49] trained. I mean, so the Netscape deal [95:51] brought in 3 million total searchers per [95:53] day. And at first, Google's sitting [95:56] there begging these early portals, hey, [95:58] please put Powered by Google on and [96:00] really trying to get that inserted in [96:01] the deal. Later on, Google got so well [96:04] known for having quality fast search [96:06] results on a big index that it was a [96:09] value proposition to show your users, [96:11] oh, our search is powered by Google. It [96:13] becomes the Intel inside of search. Yes. [96:16] It's the ingredient brand. That's [96:18] exactly right. So now they've got some [96:19] distribution, millions of users, but [96:21] still very little revenue in June 2009. [96:25] Yep. So over the next year or so, [96:27] obviously they're working on the other [96:29] business models too, but OMIde keeps [96:31] signing up, you know, some smaller [96:32] portals, some international portals on [96:34] the success of the Netscape deal, [96:36] getting more of these OEM portal deals [96:39] for Google. And then they start working [96:42] on the big kahuna, Yahoo. Yes. But [96:48] before we tell that story, now is a [96:50] great time to thank one of our favorite [96:52] companies, Statzig. So David, it's [96:55] funny. In a way, Google was the first [96:58] modern software company in the way we [97:00] think about them today with super cushy [97:02] campuses, an engineering first culture, [97:04] 20% time, which was copied by fast [97:07] growing startups who wanted to be like [97:09] Google. Yep. Google also pioneered [97:12] something else we take for granted, [97:13] though, which is great data tools for [97:15] product and engineering teams. Google [97:17] was really the first company to make [97:19] analytics dashboards, AB testing, [97:21] feature controls, and other tools [97:23] accessible across thousands of engineers [97:25] and data scientists that they'd [97:26] ultimately scale to. These tools enabled [97:28] a fast-moving bottoms up approach to [97:30] product development. So, just like [97:31] Google's office and culture, this [97:33] practice also became the norm for future [97:36] tech companies. And for the past 15 [97:38] years or so, every major tech company [97:40] has had entire teams of people [97:42] rebuilding this stack of tools [97:44] internally for themselves. But something [97:46] interesting is happening with the latest [97:48] generation of tech giants. Rather than [97:50] building these tools themselves, [97:51] companies like OpenAI, Figma, Atlassian, [97:53] Brex, Notion, Anthropic, they're just [97:55] using Statsig. Yes, Statsig has rebuilt [97:59] this entire suite of data tools that was [98:01] available at just 10 or 15 big giant [98:05] companies as a standalone company [98:07] itself. This is experimentation with [98:09] proper statistical analysis, feature [98:11] flags for safe deployments, session [98:13] replays, analytics, and more. All backed [98:16] by a single set of product data. And [98:18] using Statsig is not just about saving [98:20] engineering time. It's about getting [98:22] worldclass infrastructure from day one. [98:24] Rather than arguing about metric [98:26] definitions or troubleshooting broken [98:27] tools, your team can just focus on [98:29] building a great product. And since they [98:31] process an enormous amount of data, [98:34] Statsig does, trillions of events per [98:36] day, they also scale with you. And if [98:38] you already have your own set of product [98:40] data, they make it easy to extend into [98:42] their tools. Statig is warehouse native, [98:44] so they can plug directly into your [98:46] existing product data, whether it's in [98:47] Snowflake, BigQuery, whatever. Yep. So, [98:50] if you're interested in giving your [98:51] product team access to incredible data [98:53] tools, go to statsig.com/acquired. [98:56] They have a generous free tier with a [98:58] $50,000 startup program and affordable [99:01] enterprise plans. That's stats [99:03] satsig.com/acquired [99:06] and just tell them that Ben and David [99:07] sent you. All right, David. So, how did [99:11] Google get big? In June of 2000, they [99:15] sign a deal with Yahoo right as the [99:18] whole world is falling apart. The dot [99:20] bubble is bursting. The peak of the [99:22] NASDAQ was March 2000. In June of 2000, [99:26] Google signs the deal with Yahoo that [99:28] they are going to take over all organic [99:32] search result back fills on Yahoo.com [99:35] with the powered by Google branding. And [99:38] Yahoo is going to invest $10 million in [99:41] Google as part of this deal. [99:43] What a deal. This totally saves Google. [99:46] Between the revenue that they got from [99:49] Yahoo for this and the $10 million [99:52] investment, it keeps the company going [99:54] through the next couple years of the dot [99:57] winter until they figure out the AdWords [99:59] business model. Yep. So traffic doubled [100:02] to 14 million searchers per day on day [100:06] one of this deal, June of 2000. We're [100:08] now a year later than the Netscape deal. [100:11] So started to get a material portion of [100:14] web traffic here with 14 million [100:16] searchers per day. Yep. And the next [100:19] year in 2001, the first full year of [100:21] this Yahoo portal search deal, Yahoo [100:25] pays Google $7.2 million for organic [100:29] search results. So it's material. And [100:32] again to underscore between the $10 [100:34] million investment, this revenue, the [100:36] other portal deals, Netscape, others, [100:38] and then others that they're able to get [100:40] on the back of Yahoo, [100:42] this revenue really bridges the company [100:44] through the dot winter. That's such a [100:46] good point and something that's often [100:48] pretty overlooked that there was no [100:50] potential for Google to raise more money [100:53] here. the venture capital gravy train [100:55] was over. And so we're sitting here [100:58] saying, "Oh, they really need to make [101:00] money and when are they going to turn [101:01] the revenue switch on?" And we're sort [101:03] of like hand ringing over here. We're [101:05] only 2 years into the company's life. [101:07] Think about startups today. You don't [101:09] have expectations of profitability. [101:11] Yeah. Exactly. Within a couple years of [101:14] founding, but Google's got a very [101:16] expensive business to run between the [101:19] people and the infrastructure. and [101:22] there's no more ability to finance it. [101:24] So revenue really was the only option. [101:27] So in the midst of all of this, as [101:30] Yahoo's coming online, the board is also [101:33] pushing Larry and Sergey to hire a CEO. [101:37] Yes. So as part of the series A process, [101:40] John Door had very begrudgingly [101:44] extracted a promise from Larry and [101:46] Sergey to hire a quote unquote [101:49] professional CEO. Larry was CEO for the [101:52] series A and then CEO for the next [101:55] couple years after the series A. They [101:58] really didn't want to do it. They were [101:59] dragging their feet. It took 16 months [102:02] to find a CEO. I don't think that was [102:05] entirely because it was hard to find [102:07] someone. I think some of that was h [102:09] let's see how long we can get away [102:10] without one. Yes. My favorite story from [102:14] the whole Google CEO hiring process was [102:20] sort of the standard playbook here that [102:22] John and Mike and Sequo and Kleiner [102:24] would run with founders when convincing [102:26] them to hire a CEO is take them around [102:28] the valley, take them on the tour, have [102:30] them meet the CEOs of the great [102:32] companies in the valley of the public [102:33] companies and say, "Look, see what a [102:35] great CEO can do for your business." [102:38] So they do this with Larry and Sergey. [102:40] They go around, they meet everybody in [102:42] the valley. They're unimpressed. They [102:44] don't like any of them. And finally, [102:46] after months of this, they come back and [102:48] they tell Kleiner and Sequoa, "All [102:51] right, there's one person that we met in [102:54] this whole process who we think meets [102:56] our bar who we would be willing to come [102:59] in and hire as our CEO here at Google." [103:03] Oh god. Who is it? Steve Jobs. Really? [103:07] Yes. Yes. Yes. Cuz wasn't he like an [103:09] idol of theirs? Yeah. Well, and he had [103:11] just come back to Apple from Next. [103:14] Whether they really meant it or not, I'm [103:16] sure if Steve had been willing to come [103:18] be CEO, they probably would have said [103:20] yes, of course. But I think it was more [103:22] like a hey, a little thumbming their [103:24] nose at the VCs like we're keeping our [103:27] buyer high. It's Steve Jobs or nothing. [103:29] Wow. [103:31] So great. Also deeply ironic given what [103:35] was to come between Apple and Google. 10 [103:38] years later. Yeah, but that is for the [103:40] next episode. So anyway, it was a pretty [103:42] contentious process through all of it. [103:45] You know, 16, 17, 18 months in. Finally, [103:48] Eric Schmidt emerges as probably the [103:51] only viable candidate out there. And I [103:53] think Eric was acceptable to both sides, [103:56] both because he was an actual engineer [103:59] and had been at Sun, he was CEO of [104:01] Noville. He was a business person, too. [104:03] who he'd been a CEO, been a CEO of [104:05] public company and you know famously he [104:07] hit the ven diagram of everything. He [104:09] also went to Burning Man as did Larry [104:11] and Sergey, right? And so Eric joins in [104:14] March of 2001 [104:16] and again I think Larry and Sergey were [104:18] still kind of resentful of the process. [104:21] I think it did come to work pretty well [104:24] and they and Larry especially realized, [104:27] hey, there's parts of being a CEO, [104:29] especially as we're getting bigger, that [104:31] like I don't really like and Eric can do [104:35] those things. I don't really want to run [104:37] a finance org. I can have Eric do those [104:39] things. And it ended up working really [104:42] well at a critical moment for the [104:44] company where they needed revenue, they [104:45] needed to build a business, and they [104:47] needed to scale. Yeah. And the three of [104:49] them kind of ran the company together. I [104:50] think they had a daily standing meeting. [104:53] So, it wasn't like there was a CEO that [104:57] took over and put the founders out to [104:58] pasture. It was CEO and then Larry was [105:01] president of products and Sergey was [105:03] president of technology. But really, it [105:05] was like there are three people running [105:07] this company together. Yep. And a [105:10] trusted relationship between three [105:12] people is just more manpower than two [105:15] people. Yeah. And the organization at [105:18] this point was so we haven't talked [105:20] about googliness yet. Yes. Let's talk [105:22] about googiness. Uniquely googly that [105:25] there would have been organ rejection if [105:27] Eric tried to take a heavier hand. I [105:29] mean he really kind of came in with a [105:33] lens toward learning and understanding. [105:35] There was somebody who decided very [105:37] early in his tenure, maybe even on his [105:39] first day to move into his office with [105:41] him cuz there wasn't enough space [105:42] anywhere else. And so he was like camped [105:44] out like an engineer at Google. had an [105:46] office mate for many months with an [105:48] engineer and you know it's sort of this [105:51] googliness. It was a little bit of like [105:53] an acid test for him. Yeah. Right. But [105:56] this giant worldview, let's solve big [106:00] problems together. Can we think bigger? [106:02] Uh no matter how crazy the solution, if [106:05] it sounds like a good idea, it's worth [106:06] running down. Googliness is kind of [106:09] utopian in a way that makes all other [106:12] companies look almost like an an evil [106:14] empire. It feels like a university in a [106:17] lot of ways. Yeah, that was the culture [106:18] there. And they wanted the mentality of [106:20] a campus too where they want [106:22] inexperienced people who don't know what [106:24] they don't know. So they try novel [106:26] approaches to problems. They collaborate [106:28] more than they otherwise would have. [106:30] Yeah. But who are really high [106:31] horsepower. Yeah. Everyone there was [106:33] ludicrously high IQ from the very [106:35] beginning. But I think that sort of [106:38] collaborative utopian thing went along [106:40] with the IQ. The phrase that I heard a [106:43] lot in the research from talking to [106:45] folks who are early Google was a healthy [106:47] disregard for the impossible was like [106:50] the modus operendi there. And it's this [106:52] culture that comes up with a mission [106:54] statement to organize the world's [106:56] information and make it universally [106:58] accessible and useful. I mean this was [107:00] in 1999 in their very first press [107:03] release after the financing that has [107:05] been the mission statement. It's also [107:07] amazing how much that mission statement [107:10] scaled. Yes, I was going to save this [107:12] for way later in analysis, but organize [107:16] the world's information. Not too broad, [107:19] not too narrow, in many ways altruistic [107:22] to attract the right type of talent that [107:24] you want, but also one that lends itself [107:27] to tremendous monetization. If you're [107:29] going to organize the world's [107:30] information and you have a bunch of [107:32] smart people, you are going to be able [107:33] to create a money printing machine based [107:36] on organizing the world's information. [107:38] They actually have a great quote in [107:39] their IPO perspectus. We believe that [107:41] the most effective and ultimately the [107:44] most profitable way to accomplish our [107:45] mission is to put the needs of our users [107:47] first. So there's this almost like [107:50] trifecta of wonderfully altruistic [107:54] sounding mission A that B lends itself [107:56] to this incredible monetization model. C [108:00] as long as we're putting the needs of [108:02] our users first. Yep. And I think Eric [108:05] really bought into this because it was a [108:08] risk. Even though he joined Google in [108:11] 2001 and a lot of these portal deals [108:13] were already underway. The Yahoo portal [108:15] deal had already happened. It wasn't [108:18] clear that Google was going to be like a [108:20] smash hit home run. I think it was clear [108:23] that it was going to survive and they [108:25] had enough revenue and they could be [108:26] profitable. But we're talking about [108:28] somebody who's the CEO of a public [108:30] company, Noville, and taking a risk to [108:33] come back to a private company that yes, [108:35] had a lot of usage, but hey, startups [108:37] are out of favor now. And I think it was [108:40] really him making a bet too of like, no, [108:43] this is what I want and I'm going to buy [108:44] into this. Yep. Totally agree. All [108:47] right. So, what was Eric walking into [108:49] here with Google and call it spring of [108:52] 2001? We're now through getting the [108:56] Yahoo portal deal done. Basically [108:59] stabilized the ship, saved the company. [109:01] Google's going to survive the dot crash [109:03] between the $10 million investment from [109:05] Yahoo plus the revenue from that portal [109:08] deal. Eric hasn't started yet, but Larry [109:11] and Sergey now turn their attention back [109:14] to ads, and they're really not happy. [109:16] like the current state of play with ads, [109:18] even though it's working to a certain [109:20] extent and advertisers are happy, there [109:23] are a bunch of problems with it. One, [109:26] it's all still hands sold on Madison [109:28] Avenue. So, like the market of the pool [109:30] of potential advertisers is nowhere near [109:34] as big as the pool of potential [109:35] searchers and intent that's happening on [109:37] Google, right? They can't really scale [109:38] this business. And so, it would require [109:40] getting an enormous amount of spend from [109:43] each of the small number of customers [109:44] they already have. Yep. And then scale [109:47] too. Another reason it's not going to [109:48] scale well is that it's all sold by [109:50] hand. So as you scale the business and [109:52] you scale the number of advertisers, [109:53] you're going to need to scale the number [109:54] of people you need to sell by hand. Like [109:56] that sucks. Then you end up looking just [109:58] like Yahoo. Yeah. So that's on the [110:00] scaling side. Then on the experience [110:05] side, the user experience side, there's [110:08] no notion of ad quality here. Google as [110:11] a value proposition to its users is we [110:14] give you the highest quality, most [110:17] efficient, best search results possible. [110:21] We help your needs the best and the ads [110:24] aren't really lining up with that. [110:25] There's no way to make sure that they're [110:27] good. Yes, [110:30] exactly. So, that's a problem. And then [110:32] four, Google's just flat out leaving [110:34] money on the table. They're giving [110:36] advertisers this great product of, hey, [110:39] we have intent of people searching for [110:41] these keywords, but Google's just [110:42] getting paid on a straight CPM basis for [110:46] what they're selling. They're not [110:47] participating in the economic value, and [110:49] they're pricing a little bit kind of [110:51] finger in the air on what the price of [110:53] any given keyword should be. Exactly. [110:56] So, now fall of 2000, they're like, [110:57] "Okay, let's address this." Yes. So all [111:01] four of those issues are things that [111:03] Google's going to address in this next [111:05] evolution of AdWords. But there's a [111:08] whole part of the world that heavily [111:10] inspired [111:13] AdWords v2. Yeah. Adwords v2 uh you [111:17] might say is Google's Instagram stories [111:19] moment. Yes. There was an innovator in [111:21] the space called Overture or its [111:24] original name goto.com. We should tell [111:27] you that story now. So go to Bill Gross [111:31] started the company out of his startup [111:33] incubator Idea Lab and he did it with [111:35] quite a bit of flare coming to the world [111:37] from the TED conference in February of [111:40] 1998. So same time as Google's about to [111:43] launch right and at the time existing [111:45] search engines as you'll remember had a [111:47] problem. This is the same exact problem [111:49] that Larry and Sergey recognized. [111:51] Quality was going down. In the old world [111:54] keyword matching algorithms were fine. [111:57] There was no one gaming the algorithms. [111:59] There wasn't a lot of real commercial [112:01] activity yet. And search engines weren't [112:03] well understood yet. And so the old, [112:06] hey, go search for dogs and the most [112:10] relevant website is probably the one [112:11] that says dogs the most. That still kind [112:13] of worked. Yeah. So now you're starting [112:15] to get in 1998 all this stuff like [112:17] keyword stuffing, white text on a white [112:19] background, people getting porn sites to [112:22] appear in search results no matter what [112:23] you're searching for, hijacking traffic, [112:26] all that sort of stuff. So, Bill had [112:28] this very radical idea. The best search [112:30] results should be determined by the free [112:33] market with dollars. Whoever is willing [112:36] to pay the most is probably the very [112:38] best search result for your given query. [112:41] And spammers who aren't relevant to your [112:43] search can't afford to pay because [112:44] there's not going to be super high [112:46] conversion. But super legitimate [112:48] businesses that would actually solve the [112:49] pain point that you're searching for [112:51] could. Just like how yellow pages in the [112:53] phone book had paid inclusion as a [112:54] philosophy that would lead to only the [112:57] most relevant listings for any given [112:59] category, right? And at the time, this [113:01] was like a completely crazy idea, but [113:04] when you think about it, it actually [113:06] does make sense. If I have a product or [113:09] service that can solve the need you're [113:12] expressing for through your your intent [113:13] in the search, I should be willing to [113:15] pay more than anybody else to meet your [113:17] needs. Absolutely. It's just a different [113:19] way of solving ranking and relevance [113:21] than Larry did. Larry and Sergey sort of [113:24] figured it out on the organic side and [113:26] Bill sort of figured it out on the paid [113:28] side. So, Bill went so far as to uh go [113:32] to didn't actually develop any organic [113:34] search technology on their own. They [113:35] relied it was all paid. Yes. Only on [113:38] paid listings. And if you kept [113:40] scrolling, they actually did show [113:42] organic results, but they would license [113:43] them from Inc. to me and others, David, [113:45] like you were saying, as a backfill. All [113:47] right. So the net of all this on the TED [113:48] stage, Bill gets wildly criticized for [113:50] this. Some people even booed the idea [113:53] when he was on stage at TED, but crazily [113:56] Bill's idea was basically right and it [113:58] had a ton of ideas that would become a [114:00] part of Google that we'll talk about [114:02] here in a minute. So here's how it [114:03] worked. When you searched, Goto would [114:05] show you a list of the paid results [114:08] exactly in order of who paid the most [114:10] with no fanciness at all beyond that. [114:12] and they would show you the price that [114:15] someone was willing to pay for your [114:17] click. So right there on the page you [114:20] could see 21 cents, 23 cents, 24 cents. [114:22] Yeah, it was fully transparent. Yes. So [114:25] insight number one, paid ads on keywords [114:28] auctioned off to the highest bidder [114:29] showing up first. Insight number two, [114:31] and again this is way back in 98, was [114:34] that this whole cost per thousand [114:36] impressions thing was wrong and that [114:38] eventually he thought the whole world [114:40] was going to move beyond this to a cost [114:42] per click or pay-per-click pricing. And [114:44] so he thought, why not just do it today? [114:46] And so Goto Advertisers only had to pay [114:49] when a user actually clicked. And the [114:51] origin of this is since Bill had a bunch [114:53] of companies at Ideal Lab, he could [114:55] uniquely feel this painoint. He sort of [114:57] hated the fact that he was getting build [114:59] for all these impressions at his [115:00] companies when he just wanted to pay for [115:02] the actual clicks. I mean, this is how [115:05] advertising worked throughout all of [115:07] human history to this point. You know, [115:08] there's that famous John Womaker quote [115:10] of half the money I spend on advertising [115:12] is wasted. The problem is I just don't [115:14] know which half. This new model of [115:17] performance-based advertising wasn't [115:19] possible until the internet when you [115:20] could track clicks and conversions. But [115:23] now all of a sudden as an advertiser you [115:26] don't have to worry anymore about what's [115:28] wasted. You know it's all performing. [115:30] And the nuance is CPM actually works [115:33] fine in brand building situations but on [115:36] conversion you actually care about the [115:38] click. So basically with cost per click [115:41] you're getting free exposure every time [115:44] your ad shows up but nobody clicks on [115:46] it. But in a high intent environment [115:47] you're not trying to get exposure. [115:49] You're trying to actually capture the [115:50] clicks. So, it's kind of reasonable the [115:51] way that it shook out that a lot of [115:54] brand-based advertising is still CPM [115:56] based, but on search engines, it totally [115:59] should be CPC. So, how did it go? Well, [116:02] it worked insanely well out of the gate. [116:05] Go to did a 100red million in revenue in [116:08] one year. [116:10] Way more than Google. Yes, this is a [116:13] good business model. They have found by [116:16] the way this is also selfserve. [116:19] There is a website where you as an [116:22] advertiser can log in and place a bid. [116:25] There is an auction that happens, you [116:28] know, a real-time auction where the [116:30] person with the highest bid, again, [116:31] placed through the website is on the [116:34] very top. Does this sound familiar to [116:35] anyone who's used Google's advertising [116:37] tools? So, the 100 million happened in [116:39] year 1. By mid1 1999, they had 8,000 [116:42] advertisers. [116:44] Compare that against what AdWords had [116:46] when they launched in October of 2000, [116:48] which was 350 advertisers in the beta [116:51] program. To your comment about scale, [116:53] David, this can just scale to so many [116:55] more advertisers. Go to goes public [116:58] within a year. Yeah. Isn't this crazy? [117:00] You're probably sitting there thinking [117:01] like, how are they not the dominant [117:02] player? So, one thing that did not [117:05] happen was patents. [117:08] They did not patent the idea of the [117:10] auction or of pay-per-click. And I got [117:13] the chance to talk to Bill when we were [117:14] prepping for this episode. He's very [117:17] direct about all this, very reflective, [117:19] also a brilliant guy. He just thought [117:21] they were obvious. He just thought this [117:23] is the way it should be done. Of course, [117:24] it should be build perclick. Of course, [117:26] there should be an auction and the [117:27] highest bidder uh is the one that that [117:30] wins. So, the nuts and the bolts of it [117:32] are that right before going public, the [117:35] lawyers flagged, hey, you really should [117:36] patent some of this. but they were just [117:39] outside the window of what was [117:41] patentable because he shared them more [117:43] than a year ago on stage at TED. So the [117:46] ideas were no longer eligible. The TED [117:49] conference. That's amazing. There's some [117:52] really interesting background to all [117:54] this too. You would think, of course, [117:56] Bill was right. This stuff is obvious. [117:58] Why had nobody tried this until 1998? [118:01] Somebody actually had tried this [118:03] earlier. There was a search engine [118:05] called Openext that did try paid search [118:09] results in 1996, [118:12] but the internet was still enough of [118:15] sort of a utopian uh community like [118:18] small enough and sort of an outgrowth of [118:19] academia that I mean people booed Bill [118:22] Gross and Goto on stage at TED in 1998. [118:25] In 1996 when OpenEx tried to do this, it [118:29] was like they got kneecapped right away. [118:31] Heresy. Yeah. Yeah. It was heresy. And [118:34] because that happened, everybody else [118:36] had a hangover from it of like, oh, [118:37] that's like a third rail. You can't [118:39] touch that. Internet users will never [118:41] tolerate paid search, right? So, it's [118:44] funny. Maybe they wouldn't have gotten [118:45] the patents anyway since Open Text was [118:47] doing it before, but that was the ethos [118:49] of that early web is how dare you litter [118:51] our organic results with your paid [118:54] inclusion, putting these ads front and [118:56] center. Look, originally Larry and [118:58] Sergey were thinking this too, right? [118:59] Yeah. The great irony is all this [119:02] criticism. We're going to flash forward [119:04] for a second. When Google does launch [119:06] AdWords V2, there's a sidebar with a [119:10] separate color. It looks super [119:12] different. The word sponsored is very [119:14] clear. You are very aware that you're [119:17] looking at like a whole separate pane [119:19] over there. That's the paid icky world [119:22] relative to my beautiful clean Google [119:23] search organic results. Anyone who's [119:26] used Google in the last few years knows [119:28] the world basically ended up exactly the [119:30] way Bill Gross envisioned. It's one [119:32] column of results. The first few are [119:34] sponsored. In Google's case, they label [119:37] them even less than Bill was labeling [119:39] them at goto and then it's followed by [119:41] the organic results after that. So what [119:44] was once criticized as absolutely [119:46] heretical has come to become basically [119:48] the dominant model of search and search [119:49] monetization today. Yeah. But the [119:52] interesting thing is the timing was not [119:53] right in the mid '9s for this. Yes. By [119:56] the time the bubble was sort of fully [119:58] inflated, the internet had become [120:01] commercial enough that hey, it was okay [120:04] for Bill to try this. And then he and Go [120:08] to Overture set the example of like, oh, [120:10] oh, this is how you're going to monetize [120:13] search. This is really how you're going [120:15] to monetize the internet. And then [120:17] Google can look and see, oh, maybe we [120:19] should do that, too. [120:21] Yes. So, a couple of quick things. [120:23] Overth did file some smaller patents on [120:26] the self-s served tools. Google did [120:28] eventually end up owing them $360 [120:31] million for infringing. But these big [120:33] ideas, CPC, auction, those are now out [120:35] there given to the world for free. So, [120:38] within the next two years, they realize [120:40] that they can take this paid search [120:42] model they have and bring it to portals [120:44] too. So just like Google started doing [120:46] organic portal deals, GoTo starts doing [120:49] paid portal deals. This goes so well [120:51] they become a B2B company. They rebrand. [120:53] This is when they switch from Goto to [120:55] Overture. They start powering the ads [120:57] for Dog Pile, Metaarch. Then they get to [121:00] the big boys with AOL and MSN and [121:02] eventually they get Yahoo. Yahoo alone [121:05] was a hund00 million deal. You know, [121:07] Google's playing over here in like fun [121:09] pennies on the ground land where they're [121:12] please sir give me some money for the [121:13] organic results and meanwhile overure [121:16] has it figured out these paid results we [121:18] are doing massive massive white label of [121:21] share deals. Once they get to Yahoo some [121:24] huge percentage of Yahoo's overall [121:27] company revenue becomes paid search ads [121:30] powered by overture. Right. Yes. I think [121:32] it's like 75%. So, let's just flash all [121:35] the way forward to this. Yahoo ends up [121:37] buying Overture for 1.6 billion. There's [121:40] a little bit of a bidding war back and [121:41] forth with Microsoft, but that's the [121:43] final price. Yahoo basically says, "We [121:45] have to own this thing." I mean, it is [121:47] our revenue. And Yahoo Market Cap had [121:49] gotten decimated when the bubble popped. [121:52] So, this was a large portion of Yahoo's [121:54] market cap that they spent for overture. [121:56] Yep. But what choice did they have? They [121:57] were over a barrel. It was the majority [121:59] of their revenue was coming from this [122:01] vendor who was rev sharing with them. [122:03] Yep. Okay. So, David, to end the [122:06] overture story before we go over to what [122:08] did Google learn from all this and start [122:10] implementing some fun trivia. Did you [122:13] know that Goto tried to acquire Google? [122:17] I did not know that. So, here's how it [122:19] went down. I asked Bill about this. Bill [122:21] thought it was a match made in heaven. [122:23] So, Google's got the best way to bring [122:25] relevant organic search with page rank. [122:29] Really amazing for informational [122:31] non-commercial searches. And GoTo has [122:33] this amazing paid system for the [122:35] commercial searches. You should totally [122:37] have one system that marries [122:39] informational queries and commercial [122:40] queries together. It's got the two best [122:43] ways to surface relevant things to you, [122:45] one paid, one organic. And Larry and [122:48] Sergey before they raised the Sequoia [122:50] and Kleiner round came to Bill and said, [122:53] "What about 200 million?" Wow. [122:57] Bill thinks, "Actually, seems fine. This [123:00] seems fair. You guys are really on to [123:02] something." They were had a chance of [123:04] getting acquired for 2x the valuation of [123:07] that extreme fund raise, right? Wow. Was [123:11] overture already public at this point? [123:13] Yes. So then Bill goes to the rest of [123:15] the Overture board and Overture at the [123:18] time is worth $2 billion and the board [123:21] their conclusion is basically how could [123:23] we give up 10% of our very important [123:28] valuable revenue generating company to [123:31] this little company with zero revenue. [123:34] It' be a dilutive transaction [123:37] and so no deal. Well, there's almost [123:40] zero chance that Google becomes Google [123:42] if that deal had happened. So, yeah, [123:44] exactly. That's the thing with these [123:46] like what would have happened otherwise [123:47] acquisitions. Yeah, that's amazing. [123:50] Well, okay. So, back to fall of 2000 [123:52] when Larry and Sergey and Google can now [123:55] finally focus on improving their ads [123:57] product. You I think we heard this from [123:59] folks in the research. Once they saw how [124:02] well the goto and overture model was [124:05] working, I think Larry and Sergey were [124:08] kicking themselves of like, ah, we [124:09] should have just done this from the [124:11] beginning. Like, why did we waste time [124:14] doing this the other way? And like, [124:16] yeah, it's going to take a lot of [124:17] technology to build this out and like, [124:18] yeah, we're going to have to focus on [124:19] it, but it's obviously the better [124:22] business. Obviously, Larry and Sergey [124:24] geniuses from their childhood through [124:27] their undergrad research projects. the [124:29] way that they conceptualize the original [124:31] page rank algorithm, everything truly [124:34] geniuses. But the second superpower on [124:37] top of that is it doesn't always need to [124:39] be their idea. They're very good at [124:42] hearing the best idea, whether it's from [124:45] outside of Google or someone else inside [124:46] Google and adopting that and making that [124:49] the thing that they run with. So October [124:52] 2000, they put Salar on the project to [124:55] improve AdWords. The first obvious thing [124:58] that they need to do is they need to [125:00] build a self-s served system. As long as [125:03] Google's still taking manual orders for [125:05] ads, they're not going to be able to [125:07] implement any of the technology to bill [125:09] people per click or anything like that [125:11] or let in smaller advertisers and expand [125:14] the pool. Exactly. Which clearly [125:16] Overture had shown there was a market [125:17] for this if they had 8,000 advertisers [125:19] against Google's, you know, couple [125:21] hundred. Yes. That they were selling by [125:22] hand. Famously, by the way, Tim Draper [125:26] was looking to invest in Overture and [125:28] eventually did lead their round. And to [125:30] test it out, he actually opened up his [125:32] computer and he bid on the keyword VC [125:35] when they were pitching him. [125:38] That's amazing. That's the like [125:39] historical proof that I have that [125:40] Overture had self-s serve and then he [125:43] got outbid. It was like one penny, two [125:44] penny and then he he started a bidding [125:46] war over the term VC. [125:50] Ah, that's such a great story. I love [125:52] it. [125:54] Okay, so obviously they need self-s [125:55] serve. That's the first thing to work [125:56] on. But Salar and Larry behind the [125:59] scenes too are clearly thinking like, [126:01] okay, how do we do this in a googly way? [126:05] Yes, we're going to borrow a lot from [126:06] overture, but I think they had a Spidey [126:09] sense already that like a that wasn't [126:11] googly, you know, we're so pure over [126:13] here, but also that it wasn't quite [126:16] right that Overture had gotten like [126:19] threequarters of the way there on [126:20] cracking the business model. Yep. And so [126:23] the thing that Salar and the team really [126:26] start noodling on is we've got this [126:30] beautiful algorithm in page rank that [126:33] can deliver highly relevant organic [126:37] results. Is there a way that we could [126:40] incorporate something like that into our [126:42] ad system as well and ensure ad quality? [126:46] Like yes, the paid system in it in and [126:50] of itself goes a long way towards [126:52] ensuring ad quality, Ben, as you were [126:53] talking about earlier, but there's [126:55] there's still potential for abuse here. [126:57] Y what can we do to really make sure [126:59] that these things are good? Well, okay, [127:02] if we're an online self-served system, [127:04] we're measuring clicks, we're going to [127:07] ultimately switch to pay-per-click, we [127:10] could track those click-through rates. [127:12] And what if we made that a signal to the [127:16] ranking of how we show the ads? I mean, [127:19] it's not just like fully paytoplay where [127:22] if you pay the most, you get placed at [127:24] the top, but actually we incorporate as [127:26] part of our ad ranking system how [127:29] effective your ads are at click-through [127:31] rate. That might solve the problem. Such [127:34] is the birth of ad rank. You've got page [127:37] rank that uses all the clever things we [127:39] talked about earlier with number of [127:41] people linking to you and how [127:42] authoritative those sources are for the [127:44] organic results to make sure that the [127:46] most relevant results are being surfaced [127:48] to you. Now we have a way over in the [127:50] paid side of the house with ad rank to [127:52] take all the great stuff that we just [127:54] talked about with overure, the self-s [127:55] serve model, the auction, the cost per [127:58] clickbased system, and we add in [128:02] click-through rate, and we feed it back [128:05] into the algorithm, creating ad rank, [128:07] which is really the main two things [128:09] going into where is your ad going to be [128:12] positioned in the ranking. It's both how [128:15] much you're willing to bid and it's how [128:17] often are users actually clicking [128:19] through so they know that it's the right [128:22] ad to be showing at that right moment. [128:25] Click-through rates are a proxy for [128:27] relevance. Yes. And by the way, as a [128:32] really nice side benefit of that, if [128:37] your formula for placing ads is a [128:40] combination of the price that an [128:43] advertiser is willing to pay per click [128:46] and the click-through rate of the ad, [128:50] well, that's actually the mathematically [128:53] optimal formula for maximizing your own [128:55] revenue as Google. H. Oh, that's [129:00] interesting. Highest price paying per [129:03] click and then the highest likelihood to [129:07] click. That is the ad that you should [129:09] show to maximize your own revenue. Oh, [129:11] it's basically an expected value [129:12] calculation. Exactly. Oh, that's funny. [129:15] But it's also perfect for advertisers [129:17] because it means that if you're a better [129:19] advertiser for that keyword, then you [129:22] actually get to pay a lower price. If [129:23] people are more natural to click through [129:25] to your service and transact on your [129:27] product, you get the privilege of [129:30] bidding lower prices and still winning [129:32] the auction. All incentives are aligned [129:35] for the user. Oh, and for the user [129:38] because then it means that the user is [129:41] only ever seeing products that are the [129:43] most relevant. Yep. So, it's funny how [129:46] all of this gets rolled out. It's fall [129:48] of 2000. They start working on this. The [129:50] first version they launch includes [129:54] self-s serve and includes ad quality, [129:59] but it doesn't yet include CPC or the [130:03] auction, which is funny. It's sort of [130:05] like they did the hardest technical [130:06] stuff first. Interesting. So this first [130:10] rev in the fall of 2000 attracts a ton [130:13] of advertisers. You've now opened the [130:14] floodgates to the longtale of [130:16] advertisers and you've introduced this [130:19] click-through rate element, this ad [130:21] quality element, you know, ad rank to [130:24] how ads are going to get served. [130:26] Advertisers pretty quickly figure out [130:29] they're still paying on a CPM basis, not [130:32] on a perclick basis. They figure out [130:34] that they can game the system by [130:36] clicking on their own ads. So, because [130:39] that'll boost the click-through rate and [130:41] then their ads will get shown more. [130:42] That's so funny. And they're not paying [130:44] per click. So, they're not costing [130:45] themselves money when they're clicking [130:47] on the ads. Right. It's actually an [130:48] efficient use of impressions to use them [130:51] internally to boost click-through rate. [130:53] Exactly. So, it becomes like for a set [130:55] of months, the greatest arbitrage in the [130:57] history of the internet was to click on [130:59] your own ads. That's so funny. On [131:02] Google. But it proves that like, okay, [131:04] this is going to work. So then they [131:06] fully borrow after that. We're now into [131:08] 2001. They borrow the rest of the model [131:11] from Overture with the cost per click [131:13] payment basis and the auction model. [131:16] Now, interestingly, they do one up [131:19] Overture on the auction model. They go [131:22] to the second price auction. This is [131:25] such a genius mechanic. There's like [131:27] eight genius mechanics we've talked [131:28] about so far in the episode, but this [131:30] one really sings. If you're the winner [131:32] of an auction, they make it so you never [131:35] have to pay anything more than one penny [131:39] above second place. So, let's say I bid [131:42] 20 cents, you bid 30 cents, and then [131:45] some other guy bids 50 cents. Well, that [131:48] other guy is going to win, but he's only [131:51] going to have to pay 31. And you might [131:53] say, well, that's silly for Google. [131:55] Like, they're leaving money on the [131:57] table. But what Google is thinking is [131:59] with a much longer lens and saying, [132:01] "Well, we'd rather have our advertisers [132:03] a trust us and feel like we're not [132:05] gouging and b not feel like they have to [132:08] constantly like check and fiddle and [132:11] look to see who the other biders are and [132:13] if they want to adjust their price, it's [132:15] actually a long-term value maximizing [132:18] thing to do. Even though in the short [132:20] run, of course, you're leaving pennies [132:22] on the table each click." It's a version [132:24] of um uh Ben, you have this theory that [132:26] you and I have been talking about that [132:28] every great company has like a stored [132:31] potential energy of value maximization [132:34] that it doesn't fully maximize. Like [132:36] Costco is the the extreme example of [132:38] this, but Google has this too. That's a [132:40] great point. Yeah, the second price [132:42] auction is storing potential energy in a [132:44] way. So, there's a fun story around [132:46] this. As you can imagine, this is a [132:50] little hard to explain to advertisers [132:52] when they roll it out, like how this [132:54] works, why it's going to be good for [132:56] them, etc. They used to just write one [132:59] check and send a fax and that check [133:01] bought them a big batch of impressions [133:03] and now you have this confusing morass, [133:06] right? So, there's a fun little story [133:08] about how to educate advertisers around [133:10] how this model works. All of this, the [133:12] sort of full version of AdWords that we [133:14] know today had launched at the very [133:16] beginning of 2002. And Cheryl Samberg [133:19] had joined the company right around that [133:20] same time, too. And she was working on [133:22] ads. You know, part of her job is to [133:23] sort of pitch to advertisers what this [133:26] new model is and explain it to them. [133:27] She's banging her head against the wall. [133:28] It's so hard to do. And so, she calls up [133:30] her mentor, Larry Summers, who had [133:32] previously been the US Treasury [133:34] Secretary and Cheryl had been his chief [133:35] of staff there. She's like, "Larry, I'm [133:38] having a hard time explaining this to [133:40] advertisers. you know how this model [133:41] works, the second price auction. It's [133:43] weird. And Larry's like, "Oh, this is [133:46] what's called a victory second bid [133:47] auction. There's a lot of economic [133:49] literature about this. This is the [133:51] optimal way to do auctions and this is [133:53] actually how the Federal Reserve sells [133:55] its Treasury bonds. You should just tell [133:57] advertisers that." [134:00] So, she does it. You know, I don't think [134:02] it sinks in right away, but eventually [134:04] advertisers get the message. [134:07] That's funny. I do know this was very [134:10] painful for Google to do to transition [134:12] all their advertisers over to this new [134:14] model. They actually called it project [134:16] sunset where they had to sunset them off [134:18] the old model, bring them on to the new [134:20] pricing. Even though it was better for [134:21] everyone, it's miserable along the way. [134:23] I have a funny kod for you to the whole [134:26] overure thing. Great. Go for it. So, I [134:30] think it is correct that Google and [134:32] particularly Salar led the charge on the [134:35] insight that click-through rate is [134:37] really important and factoring it back [134:38] in is relevant. Overure did also figure [134:42] it out earlier. The problem was that [134:46] after the team implemented it, [134:48] advertisers, as you would imagine, no [134:51] longer knew how much to bid. And [134:53] Overture's whole thing was we have this [134:56] transparent thing on the page where we [134:58] show the prices and whenever you load up [135:01] a page now the prices were out of order [135:03] and you're like well what am I supposed [135:04] to get to bid to get the top spot and [135:07] also like it makes it plain and clear [135:10] you guys are supposed to be transparent [135:11] now you're this like confusing black [135:13] box. Overture did not take the pain of [135:16] transitioning over to this new model and [135:19] they just abandoned it and said, "Uh, [135:21] we're not going to mess with this [135:22] click-through rate thing," which ended [135:23] up being crucially important to the [135:25] model. This also highlights something [135:27] that we would be remiss not to say as [135:29] well. The technical infrastructure to [135:32] dynamically execute second bid auctions [135:36] every single time a user is making a [135:38] search query was incredible, especially [135:42] at Google scale. in 2001 with the [135:45] technology available then. Yes. Yeah. I [135:48] totally believe that Overture did try to [135:50] implement it had the same idea and also [135:52] that Overture had really good technology [135:53] too. Like I I think that is true. But [135:56] again back to Google's infrastructure [135:58] and the commodity hardware and the [135:59] scaling out data centers and the [136:01] distributed file system and distributed [136:03] computing to really scale this you [136:05] needed special infrastructure that only [136:08] they had. So, okay, we're talking about [136:10] the pain of transitioning the Google ad [136:13] model and ultimately the whole business [136:15] over to this new beautiful AdWords model [136:18] and and how hard that was to put some [136:19] numbers on it. So, in 2001, the year [136:22] that the Yahoo deal saved the company, [136:25] Google for the year ultimately did $86 [136:28] million in revenue that year and $10 [136:30] million in profit. So, turned profitable [136:33] in 2001. like great numbers by any [136:36] metric for a startup and especially in [136:38] the middle of the dot winter, but almost [136:43] all of that revenue was the combination [136:44] of these portal deals and the old ad [136:48] system that was in place. So, Project [136:51] Sunset and transitioning over to the new [136:53] ad system put a large portion of that [136:56] revenue at risk. Oh, yeah. It was not [137:00] necessarily an easy decision. I mean, it [137:02] was an easy decision because like the [137:04] performance was so clearly better and [137:06] like eventually economic incentives [137:08] would kick in, but it was still a little [137:12] tenuous there in Google land for a [137:13] while. The other thing that's happening [137:15] at this exact same time is Eric Schmidt [137:17] arrives and discovers that 50% of [137:19] Google's searches are outside the US, [137:21] but they have no international ad sales. [137:25] There's no international business. So he [137:27] almost jokingly tells Omid, "Just go get [137:30] on a plane. Go to the airport Monday [137:32] morning and I'll call you and tell you [137:34] what market to where to buy a ticket to [137:36] and we'll just kind of go from there or [137:38] just pick a country and we'll, you know, [137:40] figure it out." They kind of build [137:41] basically these little startup teams in [137:43] a bunch of different geographies that [137:44] kind of act as their own company selling [137:46] Google ads on the new system. But it [137:48] basically works year one because OMID [137:51] spent the whole year on the plane. 18% [137:53] of revenue is now international. 2002 it [137:55] grew to 22%. 2003 it grew to 29%. Today [137:59] it's half of Google's business. And they [138:02] even had a business in China for a long [138:03] time until famously they got into a [138:05] fight with China that started in when [138:07] was that 2002 and 03 04 something like [138:11] that. Yeah. Eventually Google finally [138:14] just withdrew and rather than censoring [138:16] results in 2010. But basically [138:19] everywhere that is not China, starting [138:21] in 2001 with Schmidt being like, "We [138:24] need an international business," they [138:25] grew themselves a just fine [138:27] international business. Yep. Yep. Yep. [138:29] So 2002 is this year of transition to [138:32] AdWords for the company. So spoiler [138:35] alert, it worked. In 2002, the company [138:40] did $440 [138:42] million of revenue. So up like, you [138:45] know, whatever that is, five, sixx from [138:47] the 86 they did the year before while [138:50] transitioning all of that revenue to the [138:52] new model. And when I say all of that [138:54] revenue, I really do mean all of it [138:55] because a lot of that 86 million, [138:58] remember, was the the the portal [138:59] partnership deals. By midway through [139:02] 2002, [139:04] Google's realizing that paid search and [139:06] Adwords is working so well, we should [139:09] stop having portals pay us for organic [139:13] search. We should start paying them to [139:16] do paid search on their sites and share [139:20] the ad revenue. And that leads to the [139:23] landmark summer 2002 [139:27] deal with AOL. Oh yes. But before we [139:31] tell that story, now is a great time to [139:34] thank one of our favorite companies, [139:36] Verscell. Verscell is such an awesome [139:39] company. Over the past few years, they [139:41] become the infrastructure backbone that [139:43] powers modern web development and now [139:45] the AI wave, too. If you visited a fast, [139:48] responsive website lately or used a [139:50] slick AI native app with agents and [139:52] hyperpersonalized interfaces, there's a [139:54] good chance it was built and deployed on [139:56] Verscell. Yeah. And the reason for that [139:58] is that Verscell has completely [140:00] reimagined the developer experience for [140:02] the modern era. In the old world, [140:04] developers had two completely different [140:05] jobs. Write code, then wrestle with [140:08] deployment. Verscell eliminated that [140:10] second component with what they call [140:12] framework defined infrastructure that [140:14] transforms your code into live globally [140:16] distributed applications automatically. [140:19] But now Versell is enabling something [140:21] even more incredible. Shipping live code [140:23] as fast as AI generates it. So in the [140:26] past even the very best web companies [140:28] used to deploy new production code [140:30] roughly on a daily cadence. Now [140:32] companies like Door Dash and Notion are [140:34] just shipping constantly on Verscell. [140:37] And Verscell is increasingly the [140:38] platform to build full stack AI apps. [140:41] Verscell has been synonymous with [140:43] front-end development, but now they do [140:44] backend and agentic workloads as well. [140:47] Exactly. And Verscell is their own [140:49] guinea pig with this stuff, too. Their [140:50] own AI product, Vzero, has 2 and a half [140:53] million users who've generated more than [140:55] 100 million messages with six app [140:57] generations every second. And Vzero runs [141:00] entirely on Verscell's platform. We did [141:02] an ACQ2 episode with Verscell CEO Guemo [141:05] back in February where we talked about [141:06] all this. It's kind of wild. So if you [141:09] want to learn more about what Verscell [141:11] can do for AI and web development at [141:13] your company, join customers like [141:15] Runway, Supreme, PayPal, Ramp, [141:18] NerdWallet, Leonardo, Zapier, and Scale [141:21] AAI. And head on over to [141:23] versel.com/acquired. [141:26] That's verel.com/acquired. [141:30] and just tell him that Ben and David [141:31] sent you. Okay, so the AOL deal and one [141:36] context setting thing just to see how [141:39] fast Google's world changed in 2002 with [141:44] the new Adwords system it going [141:45] phenomenally well as recently as late [141:48] 2001 just months before there's a dinner [141:52] with Terry Seml who's the CEO of of [141:55] Yahoo kind of big media executive guy [141:58] comes in takes over is the big media CEO [142:01] and he sits down with the founder ers [142:02] and he says, "So guys, I think we're [142:05] your biggest customer, right? Like us [142:08] paying you for those organic results in [142:10] the portal deal?" And they say, "Yep." [142:13] And he says, "That's like less than $10 [142:15] million that we're paying you. So you [142:18] don't have a business, do you? The [142:21] business can't be that big if we're your [142:22] biggest customer at less than $10 [142:24] million." And everyone's all excited [142:26] about you. And they're like, "Uh, we're [142:28] excited about some of the things they [142:29] have in the works." and they're clearly [142:31] thinking about this new AdWords system [142:34] that they're going to launch on the [142:35] spot. He offers to buy Google for a [142:37] billion dollars. Wow. Even knowing you [142:41] aren't doing much revenue. And this is [142:43] at a time Google was so secretive preo. [142:46] I mean, no one knew other people in the [142:48] business. Terry Semble sits down. He [142:49] didn't have a real sense of their [142:51] revenue. He just knew that he was the [142:52] biggest customer. So, I think this kind [142:54] of illustrates just what an insane 12 [142:56] months it was to go from being in that [142:59] position to the numbers that you just [143:01] shared on what is it 5x revenue in a [143:05] year. Yep. They more than 5xed revenue [143:07] from 2001 to 2002 from that 86 million [143:10] to 440 million in revenue. All right. [143:14] So, they're feeling real good about this [143:17] and so they decide to go to AOL, the big [143:22] source of traffic. I mean, AOL has 34 [143:26] million users at this point. It's funny [143:28] imagining this recently, AOL was still a [143:31] big and important company, but if they [143:34] could figure out how to be the search [143:36] results provider and more importantly [143:38] the ad provider for AOL, that is a [143:41] potentially company-making event. Yes. I [143:43] mean, hell, 2002, [143:46] I was probably just transitioning from [143:49] using AOL as my way to access the [143:52] internet as a senior in high school. I [143:56] think we had maybe just gotten broadband [143:58] like that year, maybe the year before. [144:00] Funny how sometimes things feel like a [144:02] lifetime ago and sometimes they feel [144:03] like just yesterday. Right. Right. So, [144:06] yeah. Okay. Summer 2002, [144:09] the transition to the new AdWords model [144:11] is blowing the doors off. Google goes to [144:13] AOL, which by the way, Yahoo sees this [144:16] and they come back and say, "How about [144:17] three billion?" [144:20] I had heard about that three billion [144:21] number. Google comes back and says, "How [144:23] about five billion?" And of course, no [144:25] deal gets done. And this would be the [144:27] last time that they seriously try to [144:28] acquire the company. Yeah. But then [144:30] Yahoo would buy Overture. [144:33] But as I pointed out a minute ago, when [144:34] Yahoo bought Overture for $1.6 billion, [144:38] that was a huge portion of Yahoo's [144:40] market cap. If they had actually done [144:42] the $5 billion [144:44] price that the Google founders floated [144:46] at them, it would have been reverse [144:48] takeover. It would have been Google [144:49] taking over Yahoo. Oh, that's a good [144:51] point. In fact, Google throwing out 5 [144:54] million is almost a farce. It's like, [144:55] how about we buy you? It's like it's a [144:56] counter offering. Yeah, we'll we'll buy [144:58] you. That's so funny. So, this AOL deal. [145:02] The current state of things is that the [145:05] 34 million AOL users, their search [145:07] experience is powered by Inktomy on the [145:11] organic side since 1999, so the last 3 [145:14] years, and Overture for the last two [145:16] years. It's sort of a bake off of do we [145:19] want those two or do we want just one to [145:21] take over all of it since Google seems [145:23] to have kind of the whole package. Now, [145:26] Google wins the deal. Here's the shape [145:29] of the deal. And then we can kind of [145:30] talk about the philosophy behind it, but [145:32] it's worth [145:35] all [145:40] UI. So Google can sign them up or AOL [145:44] can say, "Hey advertiser, I know we have [145:46] a long-standing relationship. You can [145:48] buy ads on our properties other than [145:50] search, but for search, here's the URL [145:52] you go to to place your order with [145:54] Google." M okay. Go through the rest of [145:57] the deal points, but this is huge. That [146:00] is huge. Google will then share back 85 [146:04] cents on the dollar to AOL for all of [146:07] that revenue. AOL wants two things in [146:10] exchange for turning over their entire [146:12] business in the search advertising world [146:14] to Google. One is we want warrant [146:18] coverage. So what they end up getting [146:20] granted is the option to buy 7.4 4 [146:25] million shares of Google at $3 per [146:28] share. So a total of a $22 million [146:31] investment. They get that as part of the [146:34] deal. Two is a $100 million revenue [146:39] guarantee. Yep. We want to make sure [146:42] that hey, even if this whole thing falls [146:44] apart, you're going to pay us at least [146:46] $und00 million and hopefully more if [146:50] these ads perform well and we're getting [146:52] 85%. So, here's the crazy thing. Google [146:54] doesn't have $100 million. When they're [146:57] negotiating this deal in May of 2002, [147:00] it's like just starting to work. Yeah. [147:04] So, Sergey Brin has a quote. He says, [147:06] "We could have gone bankrupt." This is [147:09] quite literally Google betting the [147:10] company. And the way to kind of think [147:12] about it is financial leverage. They [147:15] took on a fixed dollar obligation with [147:18] that revenue guaranteed AOL. So if [147:21] there's upside to Google, it would have [147:22] been a huge huge win. But if there's [147:24] downside in their business of serving, [147:28] yeah, if they couldn't make it work on [147:29] AOL, right? They obviously have very [147:31] high confidence that it would. But if [147:33] something happens and they're like, oh [147:35] shoot, we actually can't sell these [147:36] things at the rates that we thought, it [147:38] would have gone from like, oh shucks, [147:39] bummer, to now that we've signed this [147:41] deal, we're bankrupt. Yeah, this was a [147:44] really, really contentious decision. And [147:46] I think it ultimately came down to Larry [147:48] and Sergey pushing for this. You are [147:50] absolutely right. There's a great quote [147:52] in Ken Alleta's book about this where [147:54] Omid says, "You're betting the company [147:56] if you do that." And Larry Page [147:59] responds, "We should be able to monetize [148:01] the pages. If not, we deserve to go out [148:03] of business." [148:07] That's great. So, yeah, those are the [148:09] deal terms, but that first one of [148:11] advertisers are going to use Google's [148:13] system. This is why it's worth betting [148:14] the company. This is when Google [148:17] discovered what we talked about on our [148:19] Meta episode when Bos Andrew Bosworth [148:22] the CTO had the insight that more ads [148:25] equals better ads and then argued to [148:27] Zuck and Cheryl like then we need to [148:29] show more ads in feeds and then they'll [148:30] get better. The more ad inventory you [148:33] have in your system, if you're serving [148:35] them dynamically based on a ranking and [148:37] targeting them, you want to have as much [148:39] inventory as possible to give you as [148:42] many candidates to choose the best ad to [148:45] serve. And so onboarding all of AOL's [148:48] search ad inventory into Google's system [148:51] was hugely strategically valuable. Yeah, [148:53] it's a market liquidity thing. Yes, [148:56] exactly. Basically, the more volume you [148:59] have in your market, the more deeply [149:01] traded the market, the more likely you [149:03] are to have an ad that has perfect [149:05] product market fit with the query. Yes. [149:08] If you have a thin pool of advertisers, [149:10] and several folks I talked to at Google [149:12] made this point to me that if Google had [149:15] come out of the gates with the AdWords [149:19] business model in all of its glory that [149:21] it ultimately became, it would have been [149:24] very hard to bootstrap from a cold start [149:28] because you would have had this [149:29] inventory problem. you wouldn't have [149:31] been able to deliver the magical high [149:34] quality ad experience because you would [149:36] have had a very thin inventory of [149:38] advertisers. You almost had to bootstrap [149:40] it up how they did and then onboard this [149:44] other supply into the marketplace to get [149:45] deep liquidity. It's funny. I just want [149:48] to pause for one second. When you say [149:49] magical, there's nothing magical about [149:52] search ads, but I think you're right [149:53] that they're like the least offensive. [149:55] They're the most likely to be what I'm [149:57] looking for without giving me any [149:59] delight whatsoever. Sorry, I meant [150:01] magical from an economic sense. It is [150:04] the most magical economic transaction I [150:08] think ever known to man. Right. I mean, [150:10] for an advertiser, you are reaching the [150:13] exact right person at the exact right [150:14] time when they have the most intent [150:16] possible to find your service. It is [150:19] actually a pretty magical economic [150:21] lever. Well, to your point at the [150:23] beginning of the episode that Google [150:25] with this business model makes more [150:28] profits than any other company. Yeah. [150:31] Airgo logically it's the most magical [150:34] business model ever discovered. Right. [150:37] So the ink to me comment on this. They [150:39] commented to the Wall Street Journal on [150:41] AOL's decision. They'll learn over time [150:44] that Google takes your users. It doesn't [150:46] help you build your property. [150:50] which wasn't wrong. I mean, how many [150:53] people use any portal today versus how [150:55] many people use Google directly today? [150:57] So, how did this actually go? They made [150:59] this huge bet. They put $100 million on [151:01] the line. You better be really, really [151:03] sure that you can come through when [151:04] you're betting your company. Uh, it [151:07] worked. It worked. AOL made $35 million [151:11] in 2002, the first half year of the deal [151:14] alone, and then in 2003 made $200 [151:17] million. Wow. Yeah. Yeah, blow through [151:21] the guarantee there. Absolutely. This [151:23] made Google a major player in the paid [151:25] listings market almost overnight and [151:27] they weren't at all before. Overture [151:29] dominated this market before. So this is [151:31] a absolute bet the company move that [151:34] couldn't have gone better and they were [151:36] taking all that inventory effectively [151:37] from Overture who was AOL's partner [151:39] before this. That's exactly right. This [151:41] is also where Cheryl Sandberg really [151:43] makes her mark. She joined David as you [151:45] mentioned sometime in the last year [151:47] right around when Eric Schmidt joined or [151:49] right after that and she was looking for [151:50] the right job to do. She's sort of [151:52] poking around the company. I think she's [151:53] a business unit manager or something [151:55] like that but Google didn't really have [151:56] business units. And so OMI sits down [152:00] with her and says you're looking for a [152:02] big job, right? And she says yes. And he [152:04] said we have this huge AOL deal that we [152:06] just signed. We have a ton of new [152:08] advertisers in a bunch of categories [152:10] that we have no idea how to service. We [152:12] need like an army of people to handle [152:14] these thousands of new advertisers and [152:16] they have to be smart and they have to [152:18] be adaptable because we have no systems [152:20] built for this yet. And then over time [152:22] they're going to have to figure out how [152:24] to scale themselves so that we're not [152:26] constantly hiring more people. They need [152:28] to feed ideas into our technology [152:30] organizations to make it so that we get [152:32] more leverage off of the people that we [152:34] hire. and she basically hired all these [152:36] great people, built out the entire [152:38] AdWords sales function to service this [152:40] monster AOL deal. That was what she did [152:43] at Google before going and becoming COO [152:46] of Facebook. We'll talk about that on [152:48] the next Google episode here. Yes. This [152:51] also I think as Google's digesting this [152:54] deal and realizing the huge strategic [152:57] value of it, this really I think gives [152:59] them license to then go play offense on [153:03] traffic acquisition everywhere. Like [153:05] basically the light bulb now goes off of [153:08] if we can have Google search paid and [153:12] organic be part of the user experience [153:17] anywhere on the internet. We should do [153:20] everything possible to do that because [153:22] it will build our liquidity pool and our [153:24] business and we will just monetize the [153:26] internet. Oh boy, will it ever. David, [153:29] you want to do distribution? Do you want [153:30] to go there right now? Oh yeah, let's go [153:32] there. Great. All right. I've been [153:33] chomping at the bit. So, we're going to [153:35] talk about all the crazy stuff they did [153:36] for distribution. [153:38] But before that, it's worth a discussion [153:40] of the business model of search. We've [153:42] been talking about it all episode, but [153:43] there's a very particular unique [153:45] characteristic that once you realize it, [153:48] it completely changes how you should [153:51] think about distribution. So, search is [153:53] a winner take all market. And not just [153:55] because it's large and consumerf facing [153:58] and it's horizontal across industries. [154:00] There's something more to it than that, [154:02] a second layer. So, they've got all the [154:04] traditional economies of scale that you [154:06] would expect it. If you make a thousand [154:08] of a widget, you get cheaper pricing [154:09] than if you make 10 of that widget. So [154:11] just like everyone else, they amortize [154:12] the fixed costs of their infrastructure [154:13] and their employees and they have a [154:15] better infrastructure model as we were [154:16] talking about earlier, etc., etc. [154:18] Exactly. But there's this crazy thing [154:20] that happens with Google where at scale, [154:23] not only do their costs decrease on a [154:26] unit basis, their revenue actually [154:29] increases per unit. So here's what I [154:32] mean by that. When you have more biders [154:34] on every keyword, you have better price [154:37] discovery in that little market and the [154:39] winning bid is a higher price than it [154:42] would be if they had less biders. Ah, [154:45] this is another reason why you want a [154:47] deeper market liquidity pool. Yes. The [154:50] second thing too is you have biders on [154:52] keywords that are less common. So let's [154:55] say you've just got the 100 biggest [154:56] advertisers in the world. You only get [154:58] to monetize some of your searches. But [155:01] if you've got a big long tale of [155:03] advertisers or just a lot of [155:05] advertisers, then you get to advertise [155:06] more of your searches. So it's more [155:09] likely that any given search results in [155:12] revenue. So having marketplace liquidity [155:16] means you always generate the most [155:17] revenue per search versus other smaller [155:20] search engines. So it's not just that [155:22] unit costs go down, it's that as they [155:25] scale, their revenue per search actually [155:27] goes up due to the auction system. Um, [155:30] yeah, there's a third statement that you [155:34] need to add to, you know, Bos's insight [155:37] from the Facebook days of more ads [155:39] equals better ads. It's more ads equals [155:42] better ads equals better business. [155:46] Absolutely. I mean, it is crazy that [155:48] there's this you make more money per [155:51] search the bigger you scale. Yeah. In [155:54] this auction-based marketplace system, [155:57] it's increasing returns to scale. [155:59] Exactly. So then keep following the [156:01] logic tree. So because each search is [156:04] worth more, well each user is worth more [156:06] over their lifetime, which means you can [156:08] pay more than other search engines can [156:11] to acquire a new user. And once you [156:14] realize this and you get a little bit [156:16] ahead, which this is where Google is [156:17] right now in history in that 2002 era, a [156:20] little bit ahead, you can start pressing [156:22] your advantage. And once you start doing [156:24] that, it's really hard for anyone to [156:26] catch up. So the cycle is get [156:29] distribution. And we haven't yet talked [156:30] about how, but somehow we've talked [156:32] about it a little bit in the portal [156:33] deals, which drives volume of searches. [156:36] More searches drives keyword bids. [156:38] Keyword bids drive up price in auctions. [156:41] The price creates more revenue for [156:42] Google. More revenue for Google means [156:44] they can pay more for distribution. The [156:46] virtuous cycle obviously goes on. So the [156:50] obvious lesson, do not just sit back and [156:54] let organic growth do its thing. even [156:56] though they've got great organic growth [156:57] and the best brand in the world and here [156:59] in 2001 2002 you want to be aggressive [157:02] and gobble up this market as fast as you [157:05] can because someone else is going to [157:06] have this insight too. Mhm. So then the [157:10] tactics what do you do? One pay massive [157:14] revenue share to your distribution [157:16] partners in some cases up to 100% of the [157:20] revenues generated. We'll talk about who [157:21] the distribution partners are in a [157:23] second, but even earlier we heard with [157:25] that uh AOL deal, they were willing to [157:27] give AOL 85%. That's a huge split. Yep. [157:31] Yes. So, with some partners, they were [157:33] incredibly aggressive. They was like, [157:34] "We're we're going to give you all the [157:35] revenue for a while just to get you on." [157:37] Yes. Exactly. If Google monetizes each [157:39] search the most, then their rev share to [157:42] distributors are going to be better than [157:43] anyone else. Let's say they give away [157:45] the same percentage as other people. Oh, [157:47] we're only giving away 70%. Well, that's [157:49] more than someone else's 70%. So, you [157:51] know, press that advantage, go 100%. I [157:54] even heard one example where they gave [157:56] more than 100%. Where they they realize [157:58] like the payback on this is just so this [158:01] property is so valuable to get this [158:03] distribution. Yes. Eventually, it just [158:06] doesn't make economic sense for a [158:07] competitor to match your pricing. They [158:08] literally will run out of money to try [158:11] to spend the way that you can spend [158:13] because your monetization per user is so [158:16] high. So realizing this is kind of a [158:18] secret weapon. This is also where being [158:20] private was nice. Some of the other [158:22] search engines were public by this point [158:24] and so they were reporting very [158:26] consistent metrics that they wanted to [158:28] continue reporting. Google could [158:29] irrationally do things like eh we're [158:31] going to overpay for distribution in [158:33] this case and they could potentially [158:34] risk having a worse quarter. Ultimately, [158:37] Google discovered this property and they [158:40] had a belief that no one else had, which [158:42] is search is going to be really, really [158:45] big. Not like a billion dollars big or [158:48] 10 billion big. Search is currently half [158:51] a trillion dollar annual revenue market. [158:54] This is a market worth betting [158:56] everything on. and they had the stomach [158:58] to invest very very very heavily where [159:00] others kind of thought like geez is the [159:02] final payoff gonna actually be worth [159:05] investing into this market and Google [159:07] thought it's literally worth any amount [159:08] of money that we could invest in this [159:10] especially in being first and being [159:13] biggest. So we've been talking so far [159:16] about distribution deals in terms of [159:18] these you know search deals with [159:20] portals. tell us about some of the other [159:23] crazy stuff they end up doing because [159:25] once you realize this, the game just [159:27] becomes get users and advertisers at all [159:30] costs. That's exactly right. So, [159:34] currently people need to know how to [159:36] type in google.com. [159:38] That sucks. It would be really nice if [159:40] you could get users without having to [159:43] like hear that from a friend and load up [159:45] a web page to start searching and like [159:47] hopefully you bookmark it. You don't own [159:50] a browser. So, how do you get a Google [159:52] search box to actually appear in the [159:53] browser instead of on google.com? Which, [159:57] by the way, Microsoft owns the browser [160:00] right now. And if there's anybody you [160:02] need to be afraid about figuring out [160:03] this secret, it's Microsoft. And [160:06] Microsoft definitely did figure out this [160:08] secret. And just to be a little more [160:10] specific on that, Google was already [160:12] going and starting to pull away from the [160:14] rest of the market. So, it would have [160:15] taken a boatload of money to try and [160:19] compete with Google even a year or two [160:21] into this, which almost no one except [160:23] Microsoft has. And then, spoiler alert [160:25] for part two, who would a few years [160:28] later try to spend a boatload of money [160:30] to compete with Google? Microsoft. Yes. [160:33] Okay. So, Microsoft's got Internet [160:35] Explorer. Google doesn't have a browser. [160:37] What do you do? It's December of 2000. [160:40] We are not talking Chrome territory [160:42] here. Google Toolbar, baby. Google [160:46] Toolbar. [160:48] Man, when this came up in the research, [160:50] it was like the biggest blast from the [160:52] past of both. Man, I love that thing. [160:55] Man, I had not thought about that in [160:57] about 15 years. And holy crap, everybody [161:01] thought this was just this gift that [161:03] Google, the benevolent Google gods, [161:05] bestowed upon the internet ecosystem. No [161:08] way. It was a hugely strategic business [161:11] model piece for them. So, here's how it [161:14] worked. They shipped it super early in [161:16] December of 2000. This is like 2 and a [161:18] half years after the company was founded [161:20] before they've figured out AdWords v2. [161:22] They had just launched AdWords v1. So, [161:24] it's both the sort of offense we're [161:26] talking about here of go be aggressive, [161:29] get users, but also defense. Google's [161:30] paranoid about Microsoft entering and [161:32] using Internet Explorer as a weapon. If [161:34] Microsoft owns the browser, they can [161:36] direct the traffic wherever they want. [161:38] So once toolbar is installed by a user [161:42] and maybe we should for younger people [161:44] explain what tool bars are. What Google [161:46] toolbar is. Yeah. What toolbars are it [161:47] was a plugin the equivalent of a a [161:50] browser extension that would basically [161:52] create a bar underneath your bookmarks [161:55] bar or I don't know if bookmarks bars [161:56] were even a thing yet. Kind of where the [161:58] books mark bar is. Yeah. At the top of [162:00] the window. Yes. and it had a little [162:02] Google search box in it along with some [162:04] other functionality and you could just [162:06] search right from the toolbar without [162:07] having to go to the website. Right [162:08] nowadays, every modern browser you just [162:11] search from the bar at the top of the [162:13] browser. There used to be two different [162:15] things. There was a URL bar and first [162:17] that's all there was and then eventually [162:19] they put in a search field inspired by [162:21] the Google toolbar. So here's the [162:24] economics on how it all works. Once [162:27] Google Toolbar was installed, a user [162:29] averaged seven times the number of [162:32] searches, [162:34] obviously, which makes them seven times [162:37] more valuable, which means you could pay [162:40] a lot of money to get someone to install [162:43] it. Yes. So, how did they pay money to [162:46] get users to install the Google Toolbar? [162:48] Because they weren't paying users. The [162:50] average annual revenue generated by a [162:52] Google user was $2. But with toolbar it [162:56] was 10 plus dollars even if you're being [162:59] conservative. So that difference that [163:01] somewhere of you know eight bucks a user [163:03] call it is your budget to play with and [163:05] estimates are that Google ended up [163:07] spending on average way less than this [163:08] for a Google toolbar install. But you [163:11] can understand the amount of lift that [163:13] they get from a Google toolbar when you [163:14] understand wow it's worth $8 more per [163:17] user in this year. And by the way, [163:21] average revenue per user, Arpoo, is [163:23] skyrocketing. It's growing very quickly. [163:25] So this $8 is just this year's going to [163:29] become $20, $50, $100. Yeah. So Google [163:33] just paid everyone that they possibly [163:36] could to bundle Google Toolbar with [163:38] their installer of an application. This [163:41] includes Adobe. You downloading an Adobe [163:44] app. Hey, congratulations. You have [163:46] Google Toolbar. You don't know it, but [163:48] Google just paid Adobe a bunch of money. [163:50] Real networks, same thing. Windzip, same [163:54] thing. They were hyperaggressive. And [163:56] just to be really clear about what's [163:57] happening here, when users are [164:00] downloading programs, what apps used to [164:03] be called, to run on their computers, [164:06] Google is paying the maker of that [164:09] program to include a Trojan horse [164:12] payload of the Google toolbar, which [164:15] will then become a Trojan horse that [164:17] lives in your internet browser in [164:19] Internet Explorer, and Google will make [164:21] a lot more money from you. So the most [164:24] horrible way to describe this is that [164:25] it's adwear, it's spyw wear, it's a [164:28] Trojan, but users loved it. I love the [164:30] Google toolbar. Totally. But the [164:32] technique itself, are you going to get [164:34] into pop-up blocking? No. Lay it on me. [164:36] Oh, well, so as they were building the [164:38] toolbar and thinking about like, okay, a [164:40] users are going to love this because [164:42] users love Google and being able to [164:43] access search, you know, that's value [164:46] prop in and of itself for this thing, [164:48] but popup ads were a problem on the [164:51] internet at this point in time. And one [164:54] of the most popular plugins for web [164:57] browsers were pop-up blockers. And so [164:59] Google decided, well, hell, why don't we [165:02] make the Google toolbar also a pop-up [165:04] blocker? Just one more incentive to [165:06] install it and become a sticky Google [165:08] user. Yes, genius. They even did a deal [165:11] with Dell directly to make sure that [165:13] when new PCs shipped with Windows, they [165:15] shipped with Google Toolbar [165:17] pre-installed on Internet Explorer. [165:20] Amazing. They famously did a deal to [165:23] become the default search engine in [165:25] Firefox just as it was becoming popular, [165:27] which served as Mozilla's main revenue [165:30] source for decades. And this is a great [165:33] one that'll be real close to home. [165:34] David, you remember the uh Google Earth [165:37] acquisition? Oh, yes, I do. Classic [165:39] acquired episode. So, Google's an [165:42] ad-based business. They buy Earth. [165:43] There's conversation in Google. How do [165:45] we put ads inside Google Earth? Well, [165:48] instead of doing that, they realized [165:50] that Google Earth is going viral. People [165:52] are downloading this thing like crazy [165:53] because it's really cool to just play [165:55] with a globe on your And the original [165:56] Google Earth was a program that ran on [165:58] your computer. It wasn't like baked into [166:00] Google Maps, the web app. Yep. And [166:02] Google Maps was like very lame compared [166:04] to Google Earth. They did very different [166:05] functions. Maps was clearly for driving [166:08] directions. Earth was, "Oh my god, I can [166:10] zoom in on my house and I can it's all [166:12] 3D. It's super cool." They just bundled [166:15] Google Toolbar with the installs of [166:17] Google Earth. And then it more than paid [166:19] for itself. We don't need to do ads. Way [166:21] more than paid for itself. Yeah, it's [166:23] crazy. No need to put ads in Google [166:24] Earth. It's so crazy. In the mid200s, [166:27] Arpoo would eventually grow $10, $20, [166:31] $30, and always a Google toolbar user [166:34] was stickier than a non-toolar user. And [166:36] so, they just had more and more and more [166:38] budget to play with in acquiring users. [166:40] And not to spoil too much, but obviously [166:42] this still plays out all the way to the [166:46] much debated Apple Safari deal today and [166:48] the tens of billions of dollars that [166:50] Google still pays to Apple in traffic [166:52] acquisition costs. Yep. [166:55] So the takeaway here is yes, they had [166:58] the best product and yes, it was fast [166:59] and yes, it was the best technically [167:01] competent and yes, they had this amazing [167:02] culture, but everyone kind of forgets [167:05] about the fact that they were so [167:07] aggressive in distribution deals. They [167:10] didn't just let people magically find [167:12] their way to Google. Yeah. And it was [167:14] all so strategic. As I've been talking [167:16] to people over the last month about [167:19] making this episode, talking to friends, [167:21] think about how to position it. I've [167:22] been like, "This first Google episode [167:25] feels like when we made the Costco [167:28] episode. It's the same beautiful ballet [167:32] where every piece of the Google business [167:36] model works together and reinforces the [167:38] other pieces and in concert it creates [167:41] the best business model of all time." [167:43] Yep. And it's funny, Toolbar happened to [167:45] be the one that worked, but it wasn't [167:47] the only thing they tried. They tried so [167:49] many desktop applications. They even had [167:51] one called Google Desktop that would [167:53] search your desktop and then incorporate [167:55] those results privately into your web [167:58] searches. The original Google Enterprise [168:00] Search application reincarnated as [168:02] Google Desktop. But that's basically the [168:03] strategy for all these applications and [168:05] clients is how do we make you a more [168:07] sticky Google user? So, do you know the [168:10] final chapter of this story of in 2004, [168:14] a new PM is hired by Google. They come [168:18] in and they take over this applications [168:21] client team that includes Google [168:23] Toolbar. That PM is Sundar Pachchai. [168:27] That's right. And that is all we will [168:29] talk about for Sundar on this episode, [168:30] but obviously he will come into play [168:32] much more in the future. Well, it really [168:34] highlights how important Google Toolbar [168:37] was and how secretive the company was [168:39] about this really being like the [168:40] strategic lynch pin of what they were [168:42] doing or a strategic lynch pin. So [168:44] there's one more business building story [168:47] to tell here before we get to the IPO [168:49] and the end of this episode and it's [168:52] another version of this sort of [168:53] extension of the Google business model [168:56] and that is AdSense which is Google's [168:59] kind of second big business line after [169:01] Adwords on search pages. Yep. And it's [169:04] another brilliant insight of how to [169:07] extend the strategic Google business [169:10] model. Before AdSense, [169:13] Google was limited to making money when [169:16] a search happened. That was the atomic [169:20] unit of the Google business model was a [169:22] search query, which actually doesn't [169:24] happen that often. It's a really [169:27] valuable thing when it happens because [169:28] it's high intent. But most of the time [169:30] someone loads up a page, it's a website [169:32] that is not google.com, right? You're [169:35] consuming content on the internet much [169:37] more often or for a higher share of time [169:40] than you are running queries and [169:43] searches. And queries and searches have [169:44] high intent. So they're really really [169:45] valuable. But there's all this other [169:48] time and content on the internet that [169:50] Google can't monetize. [169:52] But because of everything that they [169:54] built for page rank and organic search [169:57] and understanding what's on a page and [170:00] then serving the page as search results [170:02] and then also for the ad system for ad [170:05] targeting and ad quality and predicting [170:07] click-through rate, they realize that [170:10] well actually we don't really need a [170:13] query to happen to serve effective ads [170:15] against what a user is consuming. What [170:19] if we essentially run a version of the [170:22] same algorithms on static pages on [170:24] publishers web pages and then reverse [170:28] serve the keywords that we would have [170:30] served for a search query that would [170:33] have landed on that page. [170:35] It's absolutely brilliant. It's a little [170:37] bit different because they're matching [170:39] the ads to content instead of to intent. [170:44] It's sort of try to fit in with the [170:45] content around it. But, you know, if [170:48] you're consuming content, you likely [170:50] have some future intent around that [170:53] content or maybe even loose intent right [170:55] now. Yep. And we've got this existing [170:58] pool of advertisers in our system and [171:02] their ads. And their ads in the system. [171:05] We could literally just run the same [171:06] ads. We could run the same ads on web [171:10] pages. So in February of 2003, they have [171:13] this idea and Google is still massively [171:17] inventory constrained for serving ads. [171:19] There's way more demand from advertisers [171:21] to be serving their ads against queries [171:24] than there are query supply in the [171:26] system, so to speak. Which is why the [171:27] auction works. If you had way more [171:29] searches, then everybody would just be [171:31] bidding, you know, one and two cents on [171:32] everything all the time and winning. [171:34] Right. Right. Right. You always want to [171:36] be supply constrained as a business. [171:37] Yes. [171:39] So, legendary Google engineer Jeff Deian [171:43] builds AdSense in six weeks. The whole [171:46] system, of course, he does. They're [171:48] great stories. You know, they launch it [171:49] first on Google groups and then they [171:51] want to test it on true third-party [171:54] websites to see how this works. And as [171:57] they're testing it, what they decide to [171:58] do is, oh, we'll just buy display ad [172:01] space on these other publishers and [172:04] rather than running display ads, we'll [172:06] serve it however they need to serve it. [172:08] but will effectively serve a display ad [172:10] that is just a window into the text ads [172:12] of AdWords ads that we're serving onto [172:14] that page. I remember seeing these for [172:16] the longest time when you publisher [172:17] enabled Google AdSense. You'd get like [172:20] what looked like search results just [172:23] like three across in a banner. Exactly. [172:26] That's what AdSense was in the [172:28] beginning. And their favorite website [172:30] for testing this was the website how [172:32] stuffuffs.com [172:34] because all of the pages on how stuff [172:37] works it turned out were like high [172:40] intent highly commercializable Adwords [172:43] pages for AdWords queries like if you [172:45] reverse engineered the ad search queries [172:47] that would have run AdWords against them [172:49] they were great pages. [172:52] So Susan Wiski came in as the product [172:54] manager for this into the process and it [172:57] becomes another big business for Google [173:00] much lower margin than their own [173:01] first-party AdWords business but adds [173:03] hundreds of millions of dollars of [173:04] revenue to Google off the bat. Yeah, [173:07] I've seen different estimates from [173:09] different points in time. Sometimes that [173:10] they share 67% of revenue, sometimes [173:14] that they share 80% of revenue. But the [173:16] right way to think about it is most of [173:17] the revenue on a click in Google AdSense [173:21] actually goes to the website publisher [173:23] and Google takes the smaller part as [173:24] their spiff. Yes, that's right. Whereas [173:27] if you actually own the search results [173:29] page and you are running firstparty ads, [173:31] you get 100%. Yeah, but it's very [173:33] similar to the portal ad deals that they [173:36] were doing with AOL and others of [173:37] sharing revenue with the publisher. It's [173:38] the same model, right? Yeah, it's a [173:40] great point. It's someone else's [173:42] traffic. The net result of this, by the [173:44] way, of the AdSense launch, Google has [173:47] had a troubled dance over the years back [173:51] and forth with publishers. Are they good [173:52] for publishers? Are they bad for [173:54] publishers? What does it mean for the [173:55] news industry? Blah blah blah. Right in [173:58] this moment when they launch AdSense, [174:01] publishers, and especially small ones, [174:03] love them. I'm making content on the [174:06] internet and all I have to do is drop in [174:07] some HTML and Google just starts [174:09] depositing money in my bank account. [174:12] This is amazing. Think about this, too. [174:15] This is the precursor to the YouTube [174:17] business model for creators. Oh, wait. [174:20] You mean I get to just create content [174:22] and put it on YouTube and Google [174:24] deposits money in my account? This is [174:26] the first version of that. An ad network [174:28] running on a thing that you make is a [174:30] beautiful thing for a small business [174:32] owner. Totally. And with all the [174:33] liquidity of advertisers of Google. So [174:36] they launched this thing, you know, Jeff [174:38] codes it up at the beginning of 2003, [174:39] kind of call it end of Q1 in 2003, they [174:42] launch AdSense. By the end of the year, [174:44] so just a few months later, it's doing [174:46] over a million dollars a day in revenue. [174:49] That's crazy. It's just crazy how fast [174:52] this grew. And we're zoomed in on [174:54] AdSense right now, but you know, we're [174:55] still preipo here. The rest of the [174:58] business is still very much developing. [175:01] It's worth sort of bouncing around to a [175:03] few different parts of Google to share [175:04] some updates that kind of didn't fit [175:07] into the story arc along the way. One is [175:10] 20% time is happening. People are [175:13] launching all sorts of fun side projects [175:16] and there's Google Labs, which is this [175:18] really great way that they're starting [175:19] to surface this stuff to users. Google [175:22] News comes out of this. is legitimately [175:24] someone's actual 20% time, their own [175:27] personal motivations. Actually, [175:29] particularly after September 11th, [175:31] there's this super strong hunger for [175:33] people to have a way to get rapidly [175:36] updating news on a topic. Yeah, you [175:38] couldn't get that before. You had to go [175:39] to CNN.com, right? So, Google News is [175:41] starting to really get some traction. [175:43] And speaking of disputes with [175:45] publishers, that is starting to heat up [175:47] as well. The second thing is their [175:50] organic search ranking is really [175:52] developing. what started as just page [175:54] rank plus then using the anchor text, [175:57] they're starting to use all sorts of [175:59] things later by 2007 they were using 200 [176:04] different pieces of information to [176:06] determine the ranking on a query and [176:09] early on I know some of these early [176:11] signals included data that they were [176:14] actually feeding back from observing [176:17] traffic. there's this data network [176:19] effect that's starting to happen where [176:20] more people use Google and that makes [176:23] Google better a on the whole so they can [176:26] understand oh if someone keeps bouncing [176:28] off that page every time this query is [176:30] searched then clearly that thing doesn't [176:33] belong we're doing a bad job on this [176:35] query yeah yeah or B personalization [176:38] what can we learn about you from a whole [176:41] bunch of things that you've done in the [176:42] past now without spoiling too much of [176:44] the future there's not a strong reason [176:46] to be logged into Google yet [176:47] personalization only works so well, but [176:50] once Google accounts become a thing, [176:51] then that'll really take off. So, it's [176:54] worth knowing, and I really didn't know [176:56] this, page rank really is the thing that [177:00] got Google going, but that part of the [177:04] algorithm isn't really the thing that's [177:06] the main differentiable asset today. It [177:09] was just the start. And I think this is [177:11] also a major difference in mindset of [177:12] Larry and Sergey and Google versus the [177:15] other search providers. Everybody else [177:17] just said, "Oh, we've got our insight." [177:18] Like, "Okay, great. We're done." You [177:20] know, yes, we solved search. Google has [177:22] never said, "We solved search." And they [177:24] just keep investing and investing and [177:26] investing. I think a they believe search [177:28] is going to be bigger than anyone could [177:29] have realized, but b they sort of had [177:31] the insight of the exponential curve of [177:34] content on the internet is growing [177:36] faster than Google will ever be able to [177:39] sort of index it all or come up with [177:40] clever strategies to sort it all. And [177:42] it's so dynamically changing that we'll [177:44] actually never catch up. And so we need [177:47] to constantly be investing to [177:49] approximate good results because we'll [177:50] actually never have optimal results. [177:52] Yep. Totally. And then the third big [177:55] sort of thing that kept developing is [177:56] their infrastructure. They put so much [177:59] over these years into building out all [178:02] the hardware that we were talking about. [178:04] A ton of software stuff. Yeah. And that [178:07] hardware stuff shifted from we're being [178:10] really uh you know entrepreneurial shall [178:12] we say and how we use commodity hardware [178:15] cheaply to oh we're designing our own [178:18] data centers. Yep. Absolutely. And we're [178:20] designing our own file systems with GFS [178:24] or you know I think we'll talk about a [178:25] lot of it in the next episode but very [178:28] real system level software that is [178:32] pioneering. Yep. Big table map produce [178:34] etc etc. Yep. That's on top of the [178:37] clever search software that they're [178:39] writing for things like synonyms. People [178:42] don't pay that much attention to this, [178:43] but synonym matching is actually a [178:45] crucial part of getting search right. If [178:48] you're searching for cat and there's a [178:50] whole incredibly relevant page about [178:52] kittens and you don't have a good way to [178:55] understand synonyms, then you're never [178:57] going to surface it even though it might [178:58] be incredibly relevant. And Google had [179:01] all these little tricks like realizing, [179:03] oh wait, when users are searching for [179:05] cool cat pictures and then they change [179:08] cat to kitten and that happens a lot, we [179:11] can infer that that is actually a [179:14] synonym. And then we can develop our own [179:17] constantly updating synonym dictionary [179:20] in real time, which will make our search [179:22] results better. And they have a thousand [179:24] of these things. And so they're just [179:26] pushing and pushing and pushing. So [179:27] we're finishing out 2003 here. They're [179:30] effectively never capital constrained [179:32] from this point on. They can always fund [179:35] every idea that they have the business [179:37] has flipped from one that kind of [179:40] prejune of 2002ish they had to make [179:43] trade-offs. And after call it the end of [179:46] 2002, there's no more trade-offs ever. [179:49] They always have the cash for every [179:51] single thing they want to do. Yes, we [179:53] talked about 2002 and the 440 million of [179:56] revenue and the 185 million of profits. [179:59] 2003 that explodes to one and a half [180:02] billion in revenue and almost 350 [180:04] million in operating income. What was [180:08] operating income the year before? 185 [180:10] million. So went from 185 to 350 in one [180:14] year. Yep. In one year. Wow. So some of [180:17] you might be listening and be like, [180:18] "Well, wait a minute. Sounds like their [180:20] margins got a lot worse." And yeah, [180:22] depending on your accounting, it did. [180:24] You get to keep all the money from [180:26] AdWords and you only get to keep 20% of [180:29] the money from AdSense. Yep. AdSense is [180:32] the answer there. AdSense added like [180:34] half a billion dollars in 2003 at much [180:37] lower margin, but was awesome. Yeah. And [180:41] as best we can understand it, Google [180:42] AdWords stayed an 85% gross margin [180:46] business. Yep. Incredible. Okay, so that [180:49] takes us to the last chapter of our [180:51] story today in 2004, which is the I [180:56] think today thought of as famous Google [180:58] IPO and today thought of as successful [181:01] and today thought of as yes successful [181:03] IPO at the time infamous and horribly [181:07] unsuccessful [181:09] Google IPO of 2004. So, I don't think [181:12] other than Microsoft there had ever been [181:16] another company like this where there [181:20] was no good reason for Google to go [181:24] public. It was wildly profitable, [181:28] generating plenty of cash, did not need [181:31] the investment money. I assume they've [181:34] never spent their IPO proceeds. No, of [181:36] course not. They've never not been [181:37] wildly, wildly, wildly profitable. And [181:40] actually, even more so than Microsoft, [181:42] Google had a really, really good reason [181:44] not to go public, which was Microsoft. [181:46] As we alluded to, there was desperate [181:49] paranoia in the company of we can't let [181:52] Microsoft, they are the actual front [181:56] door to the internet for all of our [181:58] users through Internet Explorer, who [182:00] actually has the capital to fight us, [182:02] and they have no idea what a good [182:03] business this is. Yes, we can't let them [182:06] know how good this is. Fortunately, I [182:09] guess for the investing public and [182:11] unfortunately for Google, the jobs act [182:14] had not been passed yet. And so the 500 [182:17] shareholder rule was still in effect for [182:21] companies in the US, which was that if [182:23] you crossed the threshold of having 500 [182:26] distinct shareholders for your company, [182:29] you had to report your financials [182:30] publicly as if you were a public [182:32] company. You know, David, I read all [182:34] this, too. They had venture capital [182:36] backers. They had to go public. Sequoa [182:38] and Kleiner Perkins are not going to be [182:40] just sitting there on their hands like [182:41] we love being private shareholders [182:43] forever. We like dividends. Yeah. No. In [182:46] 2003 in these funds that just went [182:49] through the dot crash. So most of their [182:51] other companies got wiped out. And it's [182:54] not like they're the venture capital [182:55] funds of today that come up with all [182:57] these clever strategies to look like [182:58] more permanent vehicles and offer [183:00] liquidity. These were closedend freaking [183:03] funds that need to get their money out. [183:05] Well, I think the question is did Larry [183:07] and Sergey care about that? Whether they [183:10] did or didn't, the 500 shareholder rule [183:12] was a forcing function, but obviously [183:13] the VCs, but to your point, I guess the [183:15] VCs didn't control the board. Larry and [183:16] Sergey controlled the company. Yeah, [183:18] exactly. Just like Bill and Paul and [183:21] Microsoft, it was extremely rare that [183:24] because Google never needed to raise VC [183:26] money after the series A, Larry and [183:28] Sergey together with the employees in [183:31] the option pool, you controlled a [183:32] majority of the votes in the company. [183:33] Hm. Meta, Google, Microsoft. There's [183:37] something correlated between founder [183:39] control and incredibly good capital [183:42] efficiency in a business. Interesting. [183:45] Regardless, this is all academic because [183:48] truly the 500 shareholder rule would [183:50] have required them to disclose their [183:51] financials anyway. So, like might as [183:53] well go public and make the VCs happy, I [183:56] guess. And employees, too. There was [183:57] clearly pent up employee demand and [183:59] there wasn't the same kind of liquidity [184:01] markets that there is today. David, put [184:04] four underlines under that. This IPO [184:07] made half of the 2,000 people who worked [184:10] at Google millionaires. Yes. Lot of [184:15] interested for this thing to go public. [184:17] Yes. So, as we get to end of 2003, [184:19] beginning of 2004, they know they're [184:21] going to cross the threshold during [184:23] 2004. They're going to have to go [184:25] public. They start interviewing [184:28] investment banks. and Larry and Sergey, [184:31] you know, they really don't want to do [184:32] this. They don't like anything about the [184:34] process. They don't like the IPO pops. [184:36] They don't like how much money the banks [184:38] make, etc., etc., etc. So, they're [184:40] interviewing banks. And we should say [184:41] this IPO pop thing, it sounds good, [184:43] right? It's like a phrase that the [184:46] investment banking community invented to [184:48] make it sound like a good thing. A pop [184:50] is a bad thing for existing [184:53] shareholders. It means that in the IPO [184:55] you incorrectly priced at too low and [184:58] then within one day upside that should [185:01] have or really like is yours because it [185:04] happened in all the intrinsic value was [185:06] built over these years goes to the [185:08] people who had access to buy your IPO [185:10] shares the investment bank's clients and [185:12] then they get this nice little pop on [185:15] day one and you were mispriced. That is [185:18] the problem that a lot of people try to [185:20] solve for in different ways. Yes. So [185:23] during the IPO process, they learn from [185:26] Bill Hamrich of WR Hamchrich, a boutique [185:29] investment bank in San Francisco, who [185:31] really doesn't have the same incentives [185:32] that the big banks in New York have with [185:35] their clients, that actually there is an [185:37] alternative way to price your IPO, [185:40] something called a Dutch auction IPO [185:43] process, which is this arcane thing that [185:45] been done before. How do you do good [185:47] price discovery? Well, the optimal way [185:49] to do this is a reverse auction where [185:52] you start the bidding high and you come [185:55] down in price incrementally [185:58] until you reach a clearing price where [186:01] the entire offering size is spoken for [186:04] with bids at that price. And you can [186:06] imagine just how much this appeals to [186:09] Larry and Sergey and Google. Oh my god, [186:12] it sounds perfect. They're like, "This [186:14] is like the whole business. This is what [186:15] we do anyway. This is delightful." [186:17] There's this legend that Eric Smith [186:19] talks about of they also they got a [186:20] letter from a little old lady that uh [186:23] hearing that Google was about to go [186:24] public and really hoping she could get [186:26] it and that the small retail investor [186:28] would have access to and that pulled at [186:30] their heartstrings and I'm sure that's [186:32] true too. But you can see why this [186:34] appeals in a vacuum. This sounds perfect [186:36] and like everyone should do it. Yes. And [186:39] theoretically this is also like kind of [186:40] what the investment banker what [186:42] algorithm are they actually running? It [186:43] should be something like this, right? [186:45] They're meeting with clients. They're [186:46] picking up the phone. Are you in at this [186:47] price? Are you in how much would you [186:48] want at that price? Like you kind of [186:50] should be running this algorithm in a [186:52] loose human way anyway. Yes. Now, so [186:55] they're worried about pricing and [186:56] they're worried about the IPO mechanism. [186:58] They're also really worried about losing [187:00] control because once they go public, [187:03] even though they have control of the [187:04] company now as a private company, the [187:06] employee shares are sort of captive. [187:07] People are going to start selling. Now, [187:09] all of a sudden, the public markets are [187:10] going to control a lot more of the [187:12] company. there's risk that Larry and [187:14] Sergey might collectively lose control [187:16] of the company here. So, they do a thing [187:19] that no one else in the technology [187:21] industry does, and Google has been [187:23] swearing up and down that they're not a [187:25] media company. And then they look to the [187:27] media companies and they go, "Wait a [187:29] minute." when the media companies need [187:30] to separate editorial control and have [187:34] sort of family stewardship of editorial [187:38] control, but they want to let the [187:39] shareholders come in and they don't want [187:41] the business to be able to affect [187:43] editorial too much. They've got this [187:45] great dual class structure or a tech [187:49] company. Yes. where the families of the [187:52] New York Times company or Dow Jones back [187:55] then or basically all the major [187:57] newspaper companies, the original family [188:00] owners had super voting shares that [188:05] ensured that collectively the family [188:07] would retain majority voting control [188:09] over the company even if they lost [188:10] economic control. So Larry and Sergey [188:13] decide, oh great, we're going to do a [188:15] dual class share structure for Google, [188:17] too. Which today is super common. Yes. [188:19] But Google started it. Google was the [188:22] first tech company to do this. I mean, [188:23] today it's it's everybody. It's, you [188:25] know, Facebook Meta, Alibaba, Shopify, [188:27] Spotify, Coinbase, Airbnb, Zoom, Data [188:30] Dog. Every major IPO since Google, every [188:34] major tech IPO has had this. Famously, [188:37] Snapchat even pushed the envelope so [188:39] far. When they IPO, the public shares [188:42] have no votes. So, it's not even just [188:44] super voting. It's like, oh, you public, [188:46] you you get no votes whatsoever. That's [188:48] like being a Green Bay Packers [188:49] shareholder. Yeah, exactly. Google [188:52] pioneered all of this, which I think is [188:54] why in retrospect this IPO is viewed as [188:57] a famous success. Okay, so they do the [189:00] Dutch auction. In practice, it does not [189:02] go well. [189:04] But why? This is the so the numbers on [189:07] this are pretty crazy. They are [189:09] initially floating in the Dutch auction [189:12] using the software. And by the way, [189:13] Google software engineers wrote the [189:15] software. I know. Amazing. Isn't this [189:17] crazy? And I don't think it's like [189:18] Google owned. I think they were [189:19] collaborating with the investment bank. [189:21] So the it's like this weird joint [189:23] partnership that they're doing where [189:24] it's Google engineers, but it's this [189:26] investment bank running the process. And [189:28] they're trying to figure out where in [189:30] the range between $108 a share and $135 [189:33] a share should we price? Will it be [189:35] fully subscribed? Well, the actual price [189:38] where they end up filling the order is [189:40] at $85 a share. Yep. which gives Google [189:43] a $23 billion market cap at IPO and they [189:47] raise $1.7 billion. You know, this is [189:50] like great, right? It's a $1.7 billion [189:53] raise, $23 billion market cap. That [189:56] looks like an astronomically high [189:58] multiple that the company's been given. [190:00] So, you should walk away and say they [190:02] really maximized value there. And they [190:04] probably were just wrong in that initial [190:06] range that they were looking for. $108 a [190:08] share at 135 and it only priced at 85. [190:11] Well, trust the mechanism. I guess it's [190:12] only actually worth $85 a share. Nope. [190:15] Pops to 100 bucks on closing of trading [190:18] the very same day. 18% bop day one. Then [190:22] by the end of the very next year, 16 [190:25] months later, it is almost a 5x. Yeah, [190:29] this thing was not at all priced [190:31] correctly. So there's a reason why when [190:34] you look at the legacy of the Google [190:35] IPO, dual class share structure, great [190:39] idea. Everybody does it. Dutch auction [190:41] IPO, not a great idea. Nobody has done [190:45] it since. It doubled within the first [190:47] few months. Yeah. I mean, it's great PR, [190:49] right? The stocks doing well. People [190:50] think high of your company. That's good [190:52] for all sorts of reasons. But this did [190:54] absolutely zero for making sure that the [190:57] company doesn't leave money on the [190:58] table. Yep. So funny. It's all kind of a [191:02] footnote of history anyway because know [191:04] what's a few percentage points between [191:06] friends when the company would go to [191:08] over$2 trillion dollars today as we are [191:11] recording this or roughly a h 100redx [191:14] the market cap when it IPOed and David [191:16] you're not counting dividends not [191:18] counting dividends right of course if [191:19] you reinvested dividends you'd make [191:21] significantly more than uh 100x since [191:23] IPO well after our Steve Bomber [191:25] interview never going to not count [191:26] dividends again but to preview a little [191:29] bit the rest of the series. Google's a [191:34] 2.1 call it trillion dollar market cap [191:38] company today. Roughly 100x since the [191:40] IPO. Amazingly, I'm going to ask, do you [191:42] know? I know you know what Google [191:45] Alphabet's price to earnings ratio is [191:48] right now. Oo, baby, I do know because I [191:50] was just looking this up. It is kind of [191:53] an all-time low. 20 [191:57] 20 earnings 6x revenue. So compare that [192:01] to its peer companies. Amazon's PE is [192:04] 35. Microsoft is 37. Nvidia is 46. Apple [192:10] is 30. And Meta is 27. And Google [192:13] Alphabet is down at 20. And it's not [192:16] like Alphabet's not growing revenue. [192:18] They're growing revenue just as fast, if [192:20] not faster, than all of those companies [192:22] except Nvidia. Something is going on [192:24] here. This price sure seems to reflect [192:27] that even though revenue is growing [192:29] nicely and margins are quite high, [192:32] somebody and that somebody is Mr. Market [192:34] thinks the future is a lot bleeer than [192:36] they do for those other companies. Never [192:39] mind that Google invented AI and [192:42] published the transformer paper. We will [192:43] get to all and all that. We will get to [192:45] all that. No spoilers. Okay, no [192:47] spoilers. But okay, that's where we're [192:50] going to leave Google for part one. But [192:52] one more little little start of a story [192:55] to tease you with for part two next [192:57] time. So the same month in April of 2004 [193:03] when Google files its S1 for its IPO, [193:06] Google does a unexpected product launch [193:09] on April Fool's Day, which really was [193:12] not a good idea because Google had a [193:13] history of fake April Fool's joke [193:16] announcements. But if you have one that [193:17] kind of sounds ridiculous, don't launch [193:18] it on April Fool's Day because people [193:20] will think it's a joke. Because the [193:21] product they launch actually sounds way [193:23] too good to be true. [193:25] Webbased [193:27] email [193:28] from Google with one gigabyte of free [193:33] storage for every single user. Now, to [193:36] put that in context, Yahoo and Hotmail, [193:40] Yahoo Mail and Hotmail at the time had [193:42] like 2 megabytes of free storage per [193:44] user. I think it was 20x the next best [193:47] is the stat that I read on Gmail. Yeah. [193:50] And it comes with Google search baked in [193:52] across all of your emails and it's [193:54] entirely web- based. Runs in your [193:56] browser anytime, anywhere. It's like the [194:00] greatest [194:01] April Fool's gift to, you know, internet [194:04] users everywhere that Google could [194:06] provide here. So, the question though is [194:08] why did they do this knowing what we now [194:10] know about Google? David, wouldn't it be [194:13] great if there was a reason, like a [194:16] really compelling reason, for someone to [194:18] be logged into Google? [194:22] And wouldn't it be great if we could [194:23] just attach more things to a user's life [194:28] that could be entry points to Google [194:31] search and the greatest business of all [194:32] time, search ads? What if, then? What [194:36] if? Okay, David, we're gonna tell the [194:37] whole Gmail story as a part of chapter [194:40] two, but I do have to give you one thing [194:43] that is specific to this episode. Go for [194:46] it. So, the engineer who started Gmail, [194:50] Paul Bukite, now of course a partner at [194:52] Y Combinator and actually with Brett [194:54] Taylor, started Friend. That's right. [194:57] Paul Bukite is awesome. and recently [195:00] launched a new venture fund and actually [195:02] the original coiner of the term don't be [195:04] evil at Google. That's right. So, he's [195:09] working on Gmail. It's very early. It's [195:11] like 2001. He's been working on this [195:13] thing for two and a half, three years [195:14] before it launches. So, we're at the [195:15] very beginning of it in his 20% time, [195:18] right? It's a 20% project. And it starts [195:22] as I'm going to look in your Unix [195:24] directory at your mail and I'm just [195:26] going to treat that like the web just [195:29] the same way that we treat web pages. [195:31] And so I'm just going to take a search [195:32] box and I'm going to point it at your [195:34] mail folder and I'm going to let you [195:37] search. That's it. That's like the only [195:40] functionality of what would become [195:42] Gmail. So the search bar is actually the [195:44] first feature of Gmail and everything [195:46] else came later. And as he's playing [195:49] around with this, he has this idea. [195:52] Well, if our core business is indexing [195:55] organic results and showing some ads, [195:57] maybe in addition to indexing and [196:00] searching this organic results out of [196:03] your mail folder, I should just go grab [196:06] ads from our ad database and just kind [196:08] of display them around and see how well [196:11] the content matches. He's showing this [196:14] off internally. [196:16] Larry and Sergey see it and they go, [196:18] "Wait, does this work on websites, too?" [196:22] And so, the thing that led to AdSense, [196:26] ah, this was the beginning of the idea [196:28] for AdSense was actually part of the [196:31] prototyping process of Gmail. Ah, [196:34] amazing. I love it. I love it. And I [196:36] love how you saved this to the end [196:37] because you knew we were going to do the [196:38] little uh teaser on Gmail. Well, you [196:42] texted me. You said, "I think we should [196:43] do a little Gmail foreshadow." in the uh [196:46] great yes great great great great so [196:49] thank you to uh Paul Bukite for sharing [196:51] the story with us amazing all right [196:53] bringing it home that is the building of [196:57] Google search business let's bring this [197:00] one home yeah so David this chapter this [197:02] episode definitely feels like the [197:04] building of the castle yeah and maybe [197:07] next episode is going to be the uh the [197:09] building of the city around it the state [197:11] around it the nation state around it [197:13] yeah and depending on your metaphor Or [197:15] is it a, you know, an entire property, a [197:18] platform that they're building around [197:20] it? A city? Is it a moat? Is it a Yeah, [197:23] but it's definitely uh This one is [197:25] building the castle. We'll have to see. [197:27] All right, let's go into playbook for [197:29] part one. Ben, what do you got? The way [197:31] that I framed playbook for this one is I [197:33] tried to just itemize the bullet points [197:35] of why did Google work? And as I think [197:38] through them, if I had to sort of lay [197:40] them out to someone, starts with the [197:42] best original algorithm insight. They [197:45] had the best organic relevance out [197:47] there, which created the best results in [197:49] order, fast, delightful, clean, simple [197:51] UX. And they were truly dedicated to [197:54] organic search. I mean, the aversion to [197:56] paid inclusion for as long as they were [197:58] served them very well. So this amazing [198:02] original algorithm for organic search is [198:04] one. two, best execution of the search [198:07] advertising model. I mean, once you get [198:09] all those puzzle pieces in place, the [198:13] auction, the switch to cost per click, [198:17] factoring in relevance, you know, with [198:19] click-through rate, it really is this [198:21] truly beautiful system. Advertisers are [198:23] incentivized to make their ads more [198:25] relevant and only bid on the most [198:27] relevant keywords because it means they [198:28] don't have to pay as much. It's the best [198:30] ads to the right users at the right [198:32] time. And to your point, David, it [198:34] maximizes, it is literally the algorithm [198:36] to maximize Google's expected value. It [198:38] is like a harmonious system that they [198:41] developed. Three, clever infrastructure [198:44] advantages. They just invented stuff and [198:47] they thought about problems differently [198:48] and they reasoned from first principles. [198:51] Four, they hired the best people truly [198:54] only world-class people for a very long [198:56] time. And because when the.com crash [198:59] happened, they could basically get [199:01] anybody that they wanted there in the [199:03] second half of this episode. Paul Bukite [199:05] had a great quote when I was talking [199:06] with him. I don't even think I realized [199:08] it at the time that it was truly just [199:11] the best people in the industry working [199:12] around him. So that's four. Five, [199:17] culture. A culture of thinking insanely [199:20] big, mostly by inexperienced, untainted [199:24] people that helps you with creativity, [199:26] that helps you come up with new ideas. [199:27] It was like the naivee of kids on a [199:29] college campus who are dreaming matched [199:31] with the brain power of the very best [199:33] PhDs. And this hardcore belief that [199:36] whatever our big ideas are, we always [199:38] have to think with scale. Every little [199:41] implementation detail has to be as this [199:43] scales, will this work or do we need to [199:46] rearchitect the system is very [199:48] impressive. So culture, which includes [199:50] power law dynamics, by the way, being [199:52] willing to make big bold bets because [199:54] they could be these multi-billion dollar [199:56] payoffs. That's five. Six, the [199:59] self-reinforcing data network effects [200:01] once it takes off. I think that is [200:03] underappreciated about Google. A lot of [200:06] people say, "Oh, the algorithm." But [200:07] like the algorithm is so dependent on [200:09] all the data that is generated. And then [200:12] lastly, a mission that has stood the [200:14] test of time. Organize the world's [200:16] information. It's not too broad. It's [200:18] not too narrow. It feels altruistic, but [200:21] of course, the business behind it is [200:24] actually the best business of all time. [200:26] Yep. I would add on to the second to [200:28] last one you had there, the data network [200:30] effects. It also is the flywheel effect [200:33] of liquidity in the marketplace of users [200:36] and queries and advertisers. Yeah. [200:39] Everything that we just talked about [200:40] like once Google had that realization of [200:44] oh our business gets better the more [200:48] users and advertisers we have and thus [200:52] we should be willing to spend basically [200:54] anything to increase those two pools. [200:56] This is my quintessence. Oh, okay. I'm [200:59] stealing your quintessence. I love it. I [201:01] feel like this is the most unique [201:02] insight of this episode is wo, these are [201:05] economies of scale that don't just [201:07] reduce your cost as you get bigger, but [201:08] it increases your revenue as you get [201:10] bigger. Yeah. Okay, great. Well, I [201:12] didn't mean to steal your thunder with [201:13] quintessence. Sorry about that. We did [201:14] our quintessence earlier this episode. [201:16] Okay, great. Great. Great. All right, [201:18] give me your playbook. Great. I've got [201:20] two other sort of meta points that [201:23] jumped out to me from this episode in [201:24] addition to what you just said about the [201:26] incredible encapsulation of why Google [201:28] worked. Great. When you and I were [201:32] talking about doing this and starting [201:35] the Google series, the reason we decided [201:38] now was the right time was because of [201:40] everything going on in AI and it feels [201:42] like understanding Google has never been [201:44] more relevant. And like if we're going [201:45] to do Google unacquired, we got to start [201:47] at the beginning and understand how [201:49] Google was built because that's what we [201:51] do. And I thought, oh, this episode will [201:54] set the stage to then get to today. [201:58] telling the story though and doing the [201:59] research I was like today is exactly the [202:01] same. The parallels I had the exact same [202:04] thought between what happened between [202:07] 1996 and 2002 feels like everything that [202:12] we are living through right now 2021 to [202:16] today right or even let's start with you [202:18] know the chat GPT moment put sharper I [202:21] thought this was going to be well we're [202:22] going to have to eat a lot of vegetables [202:24] to understand Google so that we can [202:26] understand where the transformer came [202:27] from to get to the real great meat and [202:31] what we can learn about AI by studying [202:33] the present. But I think by studying the [202:35] way that search played out, how did [202:37] monetization work? How did the value [202:40] chains work? How did distribution work? [202:42] How did monetization work that uniquely [202:44] enabled distribution? Where did all the [202:46] competitive dynamics come from? This is [202:48] a history doesn't repeat, but it rhymes. [202:51] And God, does this rhyme? Totally [202:54] transferable [202:55] lessons and dynamics. Yep. when you were [202:59] telling the story of GoTo and Overture [203:02] and the launch at TED and you know how [203:04] upset people were but how brilliant I [203:06] was thinking like well what would the [203:08] analogy be today like what if somebody [203:10] made a chatbot you know an LLM a model [203:14] and what it told you was just what [203:16] people paid it to tell you like people [203:18] would go crazy you know if that happened [203:20] but like yes is that worth trying should [203:23] somebody try that like well let's see [203:25] you know product design here on acquired [203:27] by David Rosen Yeah. Right. Right. [203:28] Right. Probably a bad idea, but it it [203:30] feels like such a similar Yeah. moment [203:33] that we're in. That's just what struck [203:35] me over the head doing all of this. So [203:37] like, wow, history doesn't repeat [203:39] itself, but it does rhyme. Yep. And then [203:41] the other big playbook theme I had was, [203:44] God, did Google really come of age at [203:46] exactly the right time? We talked about [203:48] this earlier, but if Larry and Sergey [203:51] had met and started working on this a [203:54] few years earlier, it would have been [203:55] Yahoo because the web was just so much [203:58] smaller. Like you didn't need a [203:59] technology-based search engine to [204:03] understand it, right? And then if they [204:04] had started a few years later, it would [204:06] have been too late. It would already [204:07] been too big. You would have needed too [204:09] much technology and power to make it [204:11] work. It was the perfect window. There [204:13] was a very narrow window to start the [204:16] Google of that era. And again, maybe [204:19] this is a subpoint of my first playbook [204:21] theme of just the parallels to today. [204:23] Yep. All right. Powers. [204:27] What of the seven powers does Google [204:30] have? And for new listeners to the show, [204:32] this is based on a book called Seven [204:34] Powers by Hamilton Helmer. And it is the [204:36] seven factors that enable a business to [204:39] achieve persistent differential returns [204:42] or basically how to be way more [204:43] profitable than your closest competitor [204:45] sustainably. And the seven are [204:47] counterpositioning, scale economies, [204:49] switching costs, network economies, [204:52] process power, branding, and cornered [204:54] resource. Let's see. To start, I guess [204:58] let's just go down the list. I actually [205:00] don't think there was tremendous [205:01] counterpositioning here. You could argue [205:03] versus Yahoo. Well, it's interesting [205:06] because it was a new industry. They were [205:08] counterpositioned against the other [205:10] search engines that existed at the time [205:11] in that they as we talked about with [205:14] Excite when they were trying to sell [205:15] back rub to Excite. The other search [205:17] engines wanted you to stay on the page [205:19] and Google didn't. But I don't think [205:21] that really counts because it was a new [205:23] industry and those were bit players. [205:25] There was no incumbent already there, [205:27] right? And Google was just better. Being [205:29] better is not counterpositioning. The [205:31] best argument for counterpositioning is [205:33] versus the portals. Yeah. Versus Yahoo. [205:36] It may have become clear at some point [205:39] that search actually is important, but [205:42] the portals couldn't really pivot to it [205:44] because they couldn't give up all of [205:45] their portal ad revenues. Yeah, I think [205:47] that's also right. And also take Yahoo. [205:49] Yahoo had constructed itself as a media [205:52] company even though it was started by [205:53] two electrical engineering PhDs from [205:55] Stanford, right? And just couldn't [205:59] pivot. I mean famously it tried to right [206:01] it bought Overture then it bought [206:02] Inktomy and then they spent years trying [206:04] to put Overture and Inctomy together to [206:06] create a packaged competitor to Google. [206:08] This was the ill- fated project Panama [206:11] at Yahoo which by the way acquired [206:14] Easter egg that is where Yan Kum and [206:17] Brian Acton met at Yahoo and then they [206:20] would get so frustrated and leave and [206:21] start WhatsApp. But yeah, I think [206:23] there's counter positioning against [206:24] Yahoo there. Okay, scale economies for [206:28] sure. There's more powers here too, but [206:30] the whole thing is scale economies. And [206:31] it's more than just the traditional one. [206:34] I mean, think about Hamilton's [206:36] traditional definition here is that [206:38] Netflix has scale economies because it [206:40] can amortize the cost of buying a given [206:44] piece of content across more users. This [206:47] is more than that. I mean this is for a [206:50] given piece of infrastructure or [206:52] software or hardware investment that [206:54] Google wants to make or a user [206:56] acquisition cost right they can amortize [206:59] that across more users but also as they [207:05] scale they make more revenue per user [207:08] right I don't know what that is is that [207:11] a new power supercale economies yeah [207:14] where does this come from auctionbased [207:17] businesses whenever you have auctions to [207:19] determine pricing. [207:21] The more liquidity you have, the higher [207:24] price is. And so, are there other [207:27] businesses that we can kind of look at [207:29] that are similar? Do scale economies [207:31] ever explain why scale gets you more [207:34] revenue? What is the thing where with an [207:37] increase in scale, their prices go up? [207:41] They maximize their available take on [207:45] any given micro auction. This is weird. [207:49] I'm trying to think in another [207:50] auction-based world like Christy's would [207:53] kind of have this or Sabes as you get [207:55] more and more people into the auction [207:58] house audience any given sale is likely [208:01] to go at a higher value. A real estate [208:04] brokerage if people were actually like [208:06] loyal clients of a real estate [208:08] brokerage. Oh, is this just network [208:10] economies that the more queries you [208:13] have, the more advertisers you'll have, [208:14] the more advertisers you'll have, the [208:15] more But typically network economies are [208:18] when people join the network, it creates [208:20] value for other people in the network. [208:22] That is true from a advertiser to a [208:24] searcher and a searcher to an [208:26] advertiser. So, we should say this [208:27] definitely has network economies. It's [208:29] almost like there's negative network [208:31] economies from advertiser to advertiser. [208:33] You don't want your competitors to be on [208:35] the platform, but Google does. Yeah, [208:37] maybe you're right. Maybe there's [208:38] something unique to auction models here [208:40] because the price is dynamic. Hamilton, [208:43] if you're listening, we need to talk. [208:46] Yeah, [208:48] great. Okay, let's keep going. Yep. [208:50] Switching costs? Not yet. Not yet. Talk [208:52] about that in the next episode. When [208:54] you're not logged in and there's no [208:55] personalization, no real switching costs [208:57] yet. Yep. And everything else you'd [209:00] switch to is worse, honestly. Branding. [209:03] Yeah, to a certain extent. Absolutely. [209:06] They built a brand of trust and speed [209:08] and and fun. I had a Google shirt. I [209:11] used to read Google blogs. I mean, I'm [209:13] trying to think, let me date the year. [209:15] 2002 [209:17] to six. I was as big a Google fan of [209:21] things that were googly as I was an [209:23] Apple fan for that period of my time. [209:25] And I think a lot of people were. I [209:27] think for those of us who weren't in [209:29] Silicon Valley, that's what it meant to [209:31] be successful at Silicon Valley was to [209:34] become Google. Yeah, I would agree with [209:36] that too. I think it's weak branding [209:38] power though because that's not a [209:40] branding power like Hermes has branding [209:41] power. It's gone away over time. Now [209:43] it's just a and for the literal [209:45] definition, are people willing to pay [209:46] more for the brand? Like are advertisers [209:48] willing to spend more on Google than [209:51] elsewhere? No, they're rational actors. [209:53] They had an employment brand though. To [209:55] your point, they absolutely had an [209:56] employment brand. Smart people would be [209:58] willing to do anything to work at [209:59] Google. Yep. And lastly, cornered [210:01] resource. Not really at this point in [210:04] time. Not really. And process power. I [210:06] also don't think there's much there. [210:08] Yeah, I don't think so. Okay. [210:10] Quintessence, we already talked about [210:12] yours. Do you want to say another word [210:13] on it? No, the increasing returns to [210:15] scale, the revenue side is unbelievable. [210:18] And the fact that they can have that [210:20] insight and then realize they need to go [210:23] be super aggressive on spending. It [210:25] makes total sense if you have a long [210:27] view and think our RPOS are only going [210:30] to grow up. People are going to be [210:31] sticky forever. It is worth investing [210:33] heavily to win this race. It's amazing. [210:36] Yes, I love your quintessence. It's [210:39] totally right. I will second and [210:41] underline it. One not quite as good but [210:44] alternative quintessence I want to put [210:46] out there about Google is a quote from a [210:50] page in Steven Levy's book in the Plex. [210:53] On June 8th, 2007, Justin Rosenstein, [210:57] who until recently had been a Google [210:59] product manager, sent an email to his [211:01] colleagues. I am writing to spread good [211:04] news. The missive said, Facebook really [211:08] is that company. Which company? That [211:12] one. The company that shows up once in a [211:14] very long while. The Google of [211:16] yesterday. The Microsoft of long ago. [211:19] That company that's on the cusp of [211:21] changing the world. that's still small [211:23] enough where each employee has a huge [211:25] impact on the organization where you [211:27] know you'll kick yourself in three years [211:29] if you don't jump on the bandwagon now [211:32] even after someone had told you it was [211:34] rolling toward the promised land that [211:37] was Google was that company Microsoft [211:40] was the first that company Google was [211:43] that company and then Facebook was that [211:45] company that's exactly right and they [211:49] are exceedingly exceedingly rare yes [211:52] they They are. So, that about captures [211:54] it. That's my quintessence. Google was [211:57] that company. All right. Carveouts. [211:59] Carveouts. I have a three-way tie of [212:03] three excellent TV shows that I watched [212:05] in the last I guess two months because [212:07] we didn't do uh Carvouts with Steve [212:09] Bulmer. Great. The first and I think the [212:13] most landmark of mine is the rehearsal [212:16] with Nathan Fielder. Season two. Oh my [212:19] god. I don't want to spoil anything for [212:22] anyone. So, if you're a person who [212:26] believes that anything is spoilers, [212:27] stop. I'll tell you things that you'll [212:29] learn in the first 10 minutes of the [212:31] first episode. Nathan, just to give you [212:33] some quick background, a decade ago did [212:35] a show called Nathan for You went and [212:38] helped small business owners figure out [212:40] how to make their businesses better, [212:42] improve on a key area. But it's all kind [212:46] of satirical. the way that he helps them [212:48] accomplish their goals is very bad for [212:52] their business in in most other [212:53] respects. He's an incredible comedian [212:55] and really dry sense of humor. And the [212:58] thing that he did in Nathan for you was [213:01] like a pretty good commitment to the [213:03] bit. The lengths that he would go to, [213:05] for example, to get a coffee shop owner [213:07] to get more traffic in their store, they [213:10] rebranded the store to dumb Starbucks. [213:12] and he spent hundreds of thousands of [213:14] dollars of the studios money or maybe [213:15] millions of dollars to commit to this [213:17] rebrand. Obviously a bad idea, but great [213:20] TV. In the rehearsal, he commits to the [213:23] bit so unbelievably hard it takes years [213:26] of his life. And he has a a goal to [213:32] reduce the number of plane crashes by [213:34] doing a deep study of the thing that [213:36] causes plane crashes, which he believes [213:38] to be pilot communication. [213:41] And in this season of the rehearsal, he [213:45] builds elaborate elaborate sets and [213:49] hires a bunch of actors to simulate [213:52] different experiences to help these [213:54] pilots feel more comfortable [213:56] communicating with each other so that [213:58] fewer planes will crash. And I am [214:00] telling you, David, this is the tip of [214:01] the iceberg. It gets crazy. [214:05] Sounds amazing. Commitment to the bit at [214:07] an all-time great. Are we committed to [214:09] the bit? We are. We're nowhere near as [214:11] committed to the bit as Nathan is. It [214:13] doesn't It sounds like we're not. Yeah, [214:15] it's inspiring. Okay, so that's one. Two [214:18] is a much more casual, very enjoyable [214:21] show called Your Friends and Neighbors [214:23] on Apple TV. It's John Ham. It's [214:25] beautifully shot. It's a little bit like [214:28] the, you know, following rich people [214:30] around like Succession is these sort of [214:32] fictional characters, but with an [214:34] unexpected twist. Love your friends and [214:36] neighbors on Apple TV. And then Andor [214:39] season 2. on Disney was excellent. [214:42] Starts a little slow. First three, four [214:44] episodes are not as good as season one [214:46] in my opinion, but the last eight [214:48] episodes are like some of the best Star [214:50] Wars canon that exists. I love it. Ben, [214:53] you are my in addition to being my best [214:55] friend, you are also my smart friend who [214:56] has like great TV recommendations. So, I [214:59] love that you get to be that smart, you [215:00] know, TV recommener friend for the [215:02] internet, too. I'm here for you and I [215:05] will continue dumping this on you even [215:06] though I know you don't watch TV. You'll [215:09] never get like to this is for listeners. [215:12] This isn't for you. Two little kids. [215:14] Tough. Tough. [215:16] My carve outs. I've got two. Well, I've [215:19] got one standard carve out. Gamecraft [215:21] season 3. Gamecraft podcast. We did a [215:24] crossover with Mitch and Blake years [215:26] ago. They have really committed to the [215:28] bit. Season 3 is excellent. I'm so glad [215:30] they've kept up the podcast. It's really [215:31] great. If you like gaming, the gaming [215:34] industry, Mitch and Blake are two of the [215:36] best in the business. My next carveout [215:38] related to that is actually sort of a [215:41] real- time dilemma carveout. And this is [215:43] good because it could be a multi-part [215:44] series here on Google. I'll report back [215:47] in the next episode which direction I [215:50] went with this carveout. So, as I said [215:53] before, I think I did a preemptive [215:55] carvout. I was so excited for Switch 2. [215:57] Switch 2 finally launched. I haven't [216:00] gotten one yet. I have a reservation for [216:02] a shopping appointment at the Nintendo [216:04] store in San Francisco. the new Nintendo [216:06] store here in San Francisco. Super [216:08] excited. I can't wait for it. I can't [216:10] wait to play it with my daughters [216:11] someday as they're approaching that age. [216:13] All right. What's the decision? As I was [216:15] watching reviews on YouTube, I started [216:18] getting into Steam Deck content. And now [216:22] I'm desperately conflicted. [216:25] Do I want to get the Switch 2 like I'd [216:27] been planning or do I want to go in a [216:30] totally different direction and get a [216:31] Steam Deck? M [216:34] well acquiring listeners, tune in. If [216:37] you weren't interested in Google part [216:39] two on its own merits, now you're going [216:41] to be on pins and needles from I would [216:44] ask our listener base for help with this [216:47] decision and I would love all of your [216:49] thoughts, but because of our editing [216:50] process, the reality is I will have made [216:52] my decision already by the time this [216:54] episode goes live. So, tweet a picture [216:57] listeners. I'll tweet a picture. Yes. [216:58] Yes. Yes. All right, that's what I got. [217:01] Awesome. Well, listeners, we would love [217:03] to see you in New York City. [217:05] Acquired.fm/nyc [217:07] if you want to come and be a part of the [217:11] ridiculous acquired experience we are [217:12] planning at Radio City Music Hall with [217:14] our good friends at JP Morgan Payments. [217:17] Speaking of JP Morgan Payments, thank [217:19] you to our partners this season. JP [217:21] Morgan Payments offers trusted, reliable [217:23] payments infrastructure for your [217:24] business, no matter the scale. to [217:27] Anthropic, the makers of Claude, an [217:30] excellent AI assistant that I used a ton [217:34] in prepping for this episode. Me, too. I [217:36] mean, it is transforming the way [217:38] acquired works, which is just awesome. [217:40] To statig, the best way to do [217:43] experimentation and more as a product [217:45] development team and Verscell, your [217:48] complete platform for web development. [217:51] We've got some thank yous. We talked to [217:53] a zillion people as seems to be the new [217:56] president when we do these large tech [217:57] companies. One off the top, Arvin [218:00] Navaratnam for Worldly Partners did an [218:02] awesome write up on uh the company as [218:05] usual, which he will make available by [218:07] clicking the link in the show notes. And [218:09] then David, you've been maintaining the [218:11] list. Yes, I've been maintaining the [218:14] list of some of the folks we should [218:15] thank for helping us with this episode. [218:18] In addition to the many other folks we [218:19] talked to who we can't mention, but [218:22] thank you. You know who you are. But [218:24] specifically, thank you to Craig [218:26] Silverstein, Google's first employee, to [218:29] Anna Patterson, to Omid Cordistani, Alan [218:32] Eustace, Clay Bavor, Brett Taylor, Jeff [218:34] Dean, Jen Fitzpatrick, Danny Sullivan, [218:38] Nick Fox, and to Paul Bukite, to Bill [218:41] Gross, to Wesley Chan, and to Esar [218:44] Lipkovitz. Thank you so much for the [218:46] conversations. David, I feel like we got [218:49] a dream team of people if we want to go [218:51] start a tech company. Yeah. Allstar [218:53] lineup. I started joking by the end of [218:55] the research process. I was like, I [218:57] think we are sapping billions of dollars [218:59] of market cap out of the economy by [219:02] taking people's time to have these [219:03] conversations. So, we greatly appreciate [219:05] it. Yep. All right, listeners. We'll see [219:08] you next time. We'll see you next time. [219:10] Who got the truth? [219:13] Is it you? Is it you? Is it you? Who got [219:16] the truth now? Huh? [219:20] [Music]