Alphabet Tumbles 12.5% as $200 Billion AI CapEx Strains Ad Model

Google Earnings, Jensen Targeted in Chinese Spy Plot, The Rise of Fake Wall St. Traders | Diet TBPN — TBPN

TBPN

Alphabet's record market drop reflects Wall Street's growing anxiety that an aggressive $200 billion infrastructure buildout cannot be sustained while generative AI search summaries cannibalize its high-margin advertising auction.

AI-assisted summary of Google Earnings, Jensen Targeted in Chinese Spy Plot, The Rise of Fake Wall St. Traders | Diet TBPN — original source. Check the original for full context.

Source published:

Key ideas

  • Alphabet shares tumbled 12.5% following Q3 earnings, wiping out nearly $500 billion in market capitalization amid an EPS miss of $2.12 versus expected $3.01, driven largely by an $8.7 billion write-down on its private SpaceX holding. (2:35)
  • Google Cloud's contract backlog expanded to $561 billion—with 55% slated for conversion into revenue over the next two years—as cloud operating margins hit 36.2%. (2:35)
  • U.S. intelligence reports uncovered that an executive at Singapore-based Megaspeed allegedly formed a close personal relationship with Nvidia CEO Jensen Huang while funneling intelligence to Beijing to facilitate export-controlled chip diversion. (14:52)
  • Retail algorithmic social media is increasingly saturated with rage-bait trading funnels and simulated prop firm SaaS platforms charging recurring $100 monthly evaluation fees that mimic actual institutional trading. (26:34)

Detailed notes

## Background This episode of Diet TBPN was published on YouTube by TBPN on October 29, 2026. Condensed from their daily live tech talk show hosted by John Coogan and Jordi Hays, the show notes highlight coverage of Google's earnings fallout, an alleged Chinese espionage operation targeting Nvidia CEO Jensen Huang, and the mechanics behind fake Wall Street day trading funnels.

## TL;DR The hosts break down Alphabet's historic 12.5% stock drop driven by an EPS miss and AI search monetization concerns, contrasting the decline with Google Cloud's massive $561 billion backlog and internal vertical integration. They explore emerging AI architectures trained on gaming simulations, including Google's Genie and Musk's Digital Optimist initiatives. The conversation examines a reported Chinese intelligence operation targeting Jensen Huang via Megaspeed executive Alice Wong, analyzing the mechanics of international GPU diversion and geopolitical arbitrage. Finally, they dissect the predatory online economy of viral 'losing our house' day-trading funnels and simulated prop trading firms that extract recurring subscription fees from retail hopefuls without executing real-market trades.

## Atmosphere & style The broadcast opens with bizarre, lighthearted physical comedy as a staffer wears a realistic silicone mask of co-host Jordi, creating an uncanny valley dynamic before shifting rapidly into sharp financial and geopolitical analysis. The banter between John and Jordi feels fluid, cynical, and quick-witted, blending Wall Street balance-sheet literacy with internet-native cultural fluency. The emotional tone shifts from genuine analytical curiosity over AI compute scaling and international smuggling schemes to exasperated amusement when exposing predatory social media grifts, culminating in a breezy wrap-up against the backdrop of half-a-trillion-dollar market sell-offs.

## Bulletin - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Alphabet shares tumbled 12.5% following Q3 earnings, wiping out nearly $500 billion in market capitalization amid an EPS miss of $2.12 versus expected $3.01, driven largely by an $8.7 billion write-down on its private SpaceX holding. - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Google Cloud's contract backlog expanded to $561 billion—with 55% slated for conversion into revenue over the next two years—as cloud operating margins hit 36.2%. - [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) U.S. intelligence reports uncovered that an executive at Singapore-based Megaspeed allegedly formed a close personal relationship with Nvidia CEO Jensen Huang while funneling intelligence to Beijing to facilitate export-controlled chip diversion. - [26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) Retail algorithmic social media is increasingly saturated with rage-bait trading funnels and simulated prop firm SaaS platforms charging recurring $100 monthly evaluation fees that mimic actual institutional trading.

## Key takeaways - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Alphabet's sharp earnings decline was driven by bottom-line weakness, specifically an EPS of $2.12 against an expected $3.01, exacerbated by an $8.7 billion SpaceX equity write-down. - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Google Cloud has transformed into a massive profit center, generating $11.1 billion in operating profit with a 36.2% operating margin and a $561 billion contracted backlog. - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Unlike competitors' long-dated contracts, 55% of Google Cloud's backlog (roughly $275 billion) is slated to be recognized within the next two years. - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) Search ad revenue growth decelerated from 17% to 12% as AI overviews and Gemini modes consume screen space without offering the monetization density of traditional ten-blue-link sponsored auctions. - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) Alphabet is projecting roughly $200 billion in CapEx for 2026, already hitting a $51.4 billion quarterly run rate in Q3 to fund its vertically integrated TPU and model pipeline. - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) Gemini surpassed 1 billion monthly active users and processes 32 billion tokens per minute via API, enabling Alphabet to access frontier AI intelligence at an estimated 90% discount to external market rates. - [10:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=600s) Quarterly CapEx shortfalls across both Microsoft and Alphabet reflect physical infrastructure bottlenecks—such as powered shells and permitting delays—rather than reduced investment appetite. - [12:56](https://www.youtube.com/watch?v=1oxs8nUVurI&t=776s) Recent frontier AI reasoning advances increasingly rely on reinforcement learning inside gaming engines and physics simulations, revitalizing Demis Hassabis's original thesis that games serve as the shortest path to AGI. - [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) An alleged Chinese espionage plot targeted Nvidia CEO Jensen Huang via Alice Wong of Megaspeed, aiming to circumvent U.S. export controls on advanced AI accelerators through Southeast Asian data hubs. - [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) Third-party data centers in Malaysia and Singapore are emerging as key conduits for Chinese entities seeking remote inference access to restricted high-end Nvidia computing clusters. - [20:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1200s) Stringent export bans have created lucrative geopolitical arbitrage opportunities, where consumer electronics and computing hardware command massive markups when smuggled across sanctioned borders. - [21:33](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1293s) Viral social media trading accounts use extreme rage-bait hooks about financial ruin to feed algorithmic distribution, steering viewers into tiered info-products and mastermind subscriptions. - [26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) The rise of simulated prop trading firms has shifted retail speculative losses from asset market downswings into recurring $100-per-month SaaS subscriptions for paper trading challenges. - [26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) Many simulated prop trading platforms operate similarly to closed-loop Ponzi structures, paying out select winners using customer subscription pools rather than authentic market trading profits.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=0s) The Halloween Impersonation Reveal

The episode opens in confusion as John Coogan talks to what appears to be co-host Jordi Hays, who sounds flat, muffled, and off-key. When asked why he sounds so strange and whether he was out surfing all weekend in Malibu or tired from using his racing simulator, "Jordi" gives stiff, unconvincing replies. John even slips in an ad read for Ramp ("Time is money saved. Both is easy as corporate cards, bill pay, accounting, and a whole lot more") before looking closely and breaking character in shock. It is revealed that team member Tyler is sitting in the host chair wearing a hyper-realistic prosthetic silicone mask of Jordi's face, complete with dark sunglasses.

John notes how bizarrely accurate the mask looks from the back, particularly the hair, although Tyler admits he can barely function inside it: “I can't really move my mouth. I can't breathe through my nose, either.” John jokes that without the sunglasses, Tyler's eyes inside the eye cutouts would deliver an absolute jump scare to the audience, while viewers in the chat accuse him of having gotten extreme Botox. Although Tyler jokes about keeping it on for the rest of the stream, John insists on subbing the real Jordi back into the chair, laughing that wearing silicone masks feels like suffocating.

[02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Google Stock Plunges 12.5% and the $561B Cloud Backlog

The hosts shift to the market carnage following Google's earnings report, which sent the stock tumbling 12.5% in a rare and sharp post-earnings sell-off. While top-line revenue consensus was pegged at $128 billion and Google delivered $127.something billion (a modest third-of-a-percent miss), the bottom line took an enormous hit. Earnings per share (EPS) arrived at $2.12 against an expected $3.01—a massive 30% miss. Some of that drag came from a non-operating hit: Alphabet absorbed an $8.7 billion loss on equity investments, primarily due to writing down its stake in SpaceX as private shares traded down, shaving 55 cents directly off EPS. Even excluding that SpaceX write-down, however, normalized EPS would have landed around $2.67, falling well short of Wall Street's expectations.

Despite the stock cratering, the hosts highlight Google Cloud as an absolute juggernaut. They recall 2018 reporting revealing that back in 2017, Larry Page had actually contemplated shuttering Google Cloud entirely amid infighting between executives and the daunting challenge of competing against Amazon Web Services (AWS). Today, the business tells the opposite story: Cloud generated $11.1 billion in operating profit with a 36.2% operating margin, prompting John to remark: “If it were independent, it would be worth it. an incredible amount of money, hundreds of billions, maybe a trillion dollars right there, just in that value.”

```mermaid flowchart LR accTitle: Google Cloud Backlog Realization accDescr: Diagram illustrating the recognition timeline for Google Cloud's 561 billion dollar backlog. subgraph Backlog["Google Cloud Backlog ($561B)"] direction TB A["Near-Term (Next 2 Years)<br/>55% (~$275B Revenue)"] B["Long-Term Tail (Up to 2033–2036)<br/>45% (~$286B Revenue)"] end Backlog --> C["Recognized into Business<br/>(Current Run-Rate $30.7B/Quarter)"] ```

- Total Google Cloud backlog stands at an enormous $561 billion, driven by frontier labs (including OpenAI) purchasing cloud capacity and Google TPUs. - Unlike long-dated enterprise contracts seen in Anthropic’s S-1 extending out to 2033 or 2036, 55% of Google's cloud backlog (roughly $275 billion) is contracted to convert into recognized revenue within the next two years.

[05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) Google Search's AI Monetization Problem and the $200B CapEx Ramp

The primary anxiety troubling Wall Street lies in search advertising. Search ad revenue growth visibly decelerated, sliding from 17% in the previous quarter to 12% this quarter. While AI overviews and Gemini modes are driving rapid token generation and query volume, they present a dangerous monetization mismatch compared to traditional ten-blue-link search. Over the past decade, Google meticulously optimized desktop and mobile 9:16 screen space until the entire first page was saturated with high-intent, auction-bidded sponsored ads. AI mode answers, by contrast, summarize user intent in plain text without yet monetizing every single pixel or sentence with commercial links.

Meanwhile, Alphabet is aggressively scaling up its capital expenditure. The company outlined roughly $200 billion in AI infrastructure CapEx for 2026, signaling that 2027 will see an even steeper climb. In Q3 alone, capital expenditures reached $51.4 billion, already putting Alphabet firmly on a $200 billion annual run rate.

```mermaid flowchart TD accTitle: Alphabet Frontier Vertical Integration accDescr: Flowchart showing how Alphabet's proprietary frontier models feed into internal products at low cost. A["Proprietary Frontier Models<br/>(Gemini for Carbon / Gemini Core)"] --> B["Internal Workloads & Infrastructure"] B --> C["Google Search & AI Mode"] B --> D["YouTube Context & Transcripts"] B --> E["Third-Party APIs & Cloud TPUs"] B --> F["90% Lower Cost vs. Market Frontier Pricing"] ```

On the consumer and developer adoption front, Gemini surpassed 1 billion monthly active users (MAUs), with its API processing 32 billion tokens per minute—a 40% jump quarter-over-quarter. Alphabet also timed the reveal of its newest frontier model, Gemini for Carbon, immediately prior to earnings to reassure investors of its competitive positioning against OpenAI and Anthropic. The hosts stress that Alphabet’s decisive advantage is full vertical integration: owning the TPUs, the models, and the consumer distribution surfaces allows them to bypass external API markups entirely. Jordi points out the bottom-line significance of this architectural independence: “Yeah, it seems like it's going to be incredibly strategic to be able to get frontier intelligence at 90% lower than market prices.”

- Alphabet's heavy infrastructure enables custom model tuning tailored directly to end products—optimizing for raw latency on AI Search queries while empowering YouTube models to parse full video transcripts without web-scraping hurdles or CAPTCHAs. - The sudden market decline also delivered an immediate paper loss to Warren Buffett's Berkshire Hathaway, which recently established a $10 billion position in Alphabet at a blended cost basis around $300 per share.

[10:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=600s) Google Search Ad Degradation and CapEx Delays

The hosts review Alphabet's share performance, pointing out that from a peak around $350 per share (across Class A and Class C stock), it dropped to $280 in roughly six months, marking a dramatic pullback as the "OG Google Bear case" reemerges. They point back to a viral July 23rd post by Max Anderson, who stated: “as someone who has personally spent $500,000 a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term.” Although analysts like Eric Seufert (the "Soufinator") and others have pushed back on Anderson's argument, the hosts observe that search engine results pages have clearly become over-monetized to the detriment of user experience.

```mermaid flowchart TD accTitle: The anatomy of a degraded Google search result accDescr: Visual breakdown of how modern commercial Google Search results are dominated entirely by paid units, widgets, and affiliate-owned assets before any organic links appear.

subgraph Paid["Top Paid Placements"] A["Google Shopping Widget"] B["3–4 Sponsored Direct-Brand Ads"] end

subgraph Mediated["Commercial Affiliates"] C["Affiliate Review Sites (Purchased by Mattress Brands)"] end

subgraph Organic["Organic Real Estate"] D["Zero Neutral Results (No Wikipedia, Minimal Organic)"] end

A --> B --> C --> D ```

A prime example is searching for a mattress: users are greeted first by paid Google Shopping callout widgets, followed by three to four direct-to-brand keyword advertisements, and then mattress review blogs that are actually owned by the mattress manufacturers themselves. There are no independent resources or neutral Wikipedia links left on the page. Despite the poor user experience, Google's ad revenue shows only slight deceleration rather than a cliff fall because consumer habits remain deeply entrenched and sticky. Turning to earnings, Microsoft reported CapEx of $38.37 billion against an expected $45.9 billion, while Google reported $51.36 billion against an expected $55.7 billion. These shortfalls are not caused by a desire to curb spending; rather, they are driven by physical infrastructure delays—such as powered shell constraints, slow permitting, and GPUs sitting idle waiting to be racked. For companies currently selling tokens, these supply bottlenecks inadvertently protect their margins.

- Alphabet stock dropped to roughly $280 per share within about six months of its ~$350 peak. - Max Anderson spent $500,000 monthly on Google Ads before sounding the alarm on search growth quality. - Microsoft quarterly CapEx hit $38.37B (vs. $45.9B estimate). - Google quarterly CapEx hit $51.36B (vs. $55.7B estimate).

---

[12:56](https://www.youtube.com/watch?v=1oxs8nUVurI&t=776s) Can Video Games and Simulation Train AGI?

The conversation shifts to emerging AI architectures and reinforcement learning (RL) breakthroughs. Industry commentary from accounts like *I'm Just New at AI* highlights rumors that massive leaps in model reasoning stem from aggressive RL inside gaming environments and simulated worlds. Research shows strategies mastered within simulated physics engines transfer remarkably well to research mathematics, Navier-Stokes fluid equations, and ETH benchmarks. Anthropic has moved in a similar direction with internal Claude checkpoints, while Elon Musk's Digital Optimist program explicitly targets measuring transfer from games into general computer use.

This brings up Ashlee Vance's commentary at *Core Memory*, questioning whether the industry is circling back to Demis Hassabis's original vision: using games as the fundamental path to AGI after a temporary detour into brute-force LLM scaling. Google's persistence with its Genie project stands out. While Anthropic stayed strictly focused on coding (initially avoiding image generation entirely), OpenAI took various compute-heavy "side quests" like Sora. OpenAI also pursued early virtual computer interaction with projects like Operator—which originally seemed clunky and janky when attempting to navigate websites, but now appears prescient as computer use becomes solved. Google's Genie could either prove to be a vanity expense or a foundational sandbox for the next generation of model training.

- Digital Optimist hiring under Elon Musk seeks to benchmark cross-domain transfer from gameplay to general operating system navigation. - OpenAI's Operator served as an early, clumsy forerunner to the modern computer-use agents that labs are perfecting today.

---

[14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) Jensen Huang Targeted in Alleged Chinese Spy Scheme

The hosts pivot to an explosive *New York Times* report revealing that NVIDIA CEO Jensen Huang was the target of a Chinese espionage operation. American intelligence agencies discovered that Alice Wong, a Chinese citizen and top executive at the Singapore-based data center operator Megaspeed, had formed a close personal relationship with Huang while simultaneously funneling sensitive intelligence directly back to Beijing. U.S. officials had Megaspeed under scrutiny for orchestrating the diversion of export-controlled advanced NVIDIA AI accelerators into mainland China. The revelation that an intelligence asset with direct access to Huang—one of the most critical figures in the global AI race and an ally of President Trump—was providing updates to Beijing triggered immediate alarm in the White House. Intelligence agencies briefed President Donald Trump on the findings roughly a year prior, though he opted to take no action at the time.

The hosts note that they had actually examined Megaspeed and Alice Wong on their October 10th episode. Reviewing an AI summary of their prior discussion, they confirm they had previously noted Wong's background in Chinese state media and questioned Megaspeed's operations:

> "Could the discussion be interpreted as raising suspicions? Yes. It was too early to go out and say we didn't have any real details around the situation, but the whole situation was so suspicious. I'm surprised it's taken this long for it to kind of bubble up."

The hosts banter about George W. Bush’s famous gaffe: “There's an old saying in Tennessee. I know it's in Texas, probably in Tennessee. That says, fool me once. Shame on... Shame on you. If you fool me, we can't get fooled again.”

They note that Huang had not addressed the reporting publicly and had not posted on social media since late October. The core strategic concern is that third-party jurisdictions like Malaysia and Singapore are serving as conduits: the more advanced NVIDIA hardware that lands in Malaysian data centers, the more inference compute is accessible to Chinese entities over the wire. The hosts compare monitoring national compute import charts to tracking a sick patient:

> "If you're measuring like an athlete and you're noticing like insane blood glucose levels or something like you might notice that there's a problem before you can diagnose it. Malaysia seems to be in that camp of like... it's, I don't know, if they're scaling dramatically given the local economy. It's suspicious."

- Target: NVIDIA CEO Jensen Huang. - Asset identified: Alice Wong, executive at Singapore-based Megaspeed and former Chinese state media reporter. - Political briefing: President Donald Trump was briefed on the intelligence findings roughly one year prior and took no executive action. - Mechanism: Routing high-end GPUs through hyperscale build-outs in Southeast Asia (notably Malaysia and Singapore) to run remote inference workloads for Chinese firms bypassing U.S. export sanctions.

[20:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1200s) Geopolitical Arbitrage and the Chip Smuggling Supply Chain

The discussion turns to the aggressive, lucrative mechanics of bypassing export restrictions to move cutting-edge technology across borders. The hosts recall reports of people using hair dryers to strip labels off servers—identifying Supermicro hardware as a prominent example—in order to smuggle them past customs checkpoints. Whenever strict geopolitical barriers emerge, immense financial incentives for arbitrage quickly follow.

```mermaid flowchart LR accTitle: Hardware Arbitrage Flow accDescr: Flow diagram showing how consumer goods and restricted chips are smuggled into conflict and sanctioned zones for massive arbitrage profits. subgraph S1["Western Sourcing"] A["Retail Purchase<br/>(e.g., DJI Drones, GPUs)"] end subgraph S2["Smuggling Route"] B["Luggage / Border Crossing<br/>(Labels stripped, relabeled)"] end subgraph S3["Restricted Markets"] C["Conflict / Sanctioned Zones<br/>(10x markup / $90k profit)"] end A --> B B --> C ```

Drawing a direct parallel to the Russia–Ukraine war, a host explains how individuals exploited consumer gear: someone could take a first-class flight to a Western democracy, walk into a retail store like Best Buy, purchase ten DJI drones for roughly $1,000 each, pack them into luggage, fly to a neighboring country, and cross into the conflict zone via bus or train to resell them for $10,000 apiece. That single trip generates $90,000 in pure profit. The exact same arbitrage dynamics now govern high-end Nvidia GPUs. Because the stakes are so high, bad actors face enormous incentives not merely to divert cargo or steal shipments, but to infiltrate the organizations themselves—attempting to get close to the CEO to decipher internal compliance procedures and figure out how to circumvent anti-diversion mechanisms. Reflecting on scenarios discussed around the "AI 2027" timeline where hostile foreign powers wake up to the strategic necessity of espionage, the hosts note that nation-states have already been executing these covert campaigns aggressively for years.

- A $1,000 retail DJI drone purchased in a Western market could command around $10,000 in a conflict zone. - Smuggling ten drones in checked luggage yields up to $90,000 in arbitrage profit per trip. - Techniques for smuggling computing hardware include using hair dryers to remove manufacturer labels and server branding to disguise restricted systems.

[21:33](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1293s) The "Losing Our House" Rage-Bait Funnel

The conversation transitions into the bizarre world of algorithmic social media after both hosts realize they were served the exact same viral Instagram Reel. After joking about their identical feeds and teasing Tyler for having to take off his on-set Jordi mask, they examine a viral video headlined with the grammatical oddity: *"Losing Our House Because of Day Trader."* Accompanied by swelling, melancholy saxophone music, the video features a distraught-looking family appearing to face the catastrophic loss of their home.

```mermaid flowchart TD accTitle: The Day Trading Rage-Bait Funnel accDescr: Step-by-step breakdown of how a viral tragedy reel converts into paid trading courses. subgraph ViralHook["Algorithmic Hook"] H["Viral Tragedy Reel<br/>'Losing Our House Because of Day Trader'"] end subgraph NarrativePivot["Redemption Story"] P["Follow-up Content<br/>'How I'm planning to win it all back'"] end subgraph Monetization["Info Product Upsell"] M1["Free Starter Course"] M2["$49 Tier"] M3["$100 'Mastermind' Tier"] end H --> P P --> M1 M1 --> M2 M2 --> M3 ```

Upon inspecting the creator’s profile, however, the apparent tragedy reveals itself as the top of an aggressive sales funnel. The very next post frames the narrative as a redemption arc—*"I lost my house day trading and here's how I'm planning to win it all back"*—directing viewers to a link in the bio that leads straight to a tiered trading info-product. The hosts describe the tiered offerings, moving from a free teaser course up to a $49 course and a $100 mastermind community. While counterintuitive to market trading expertise by admitting you lost your own home, it ruthlessly exploits algorithmic negativity bias: viewers are gripped by the initial shock, click through, and end up viewing the creator as an inspiring underdog refusing to quit before hitting it big. As one host summarizes the exploitative packaging: “And so all of this is actually a funnel to an info product and this is designed to go viral.”

- The viral video employed the specific title *"Losing Our House Because of Day Trader"* alongside dramatic saxophone music to maximize watch time. - The creator's bio funnel offered a progression of products: a free course, a $49 paid course, and a $100 mastermind tier. - The hosts observe that it remains entirely ambiguous whether the family actually lost their house or simply staged the premise for content.

[26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) The Simulated Prop Trading Trap

Digging deeper into the mechanics of these day-trading creators, the hosts point out a fundamental shift in how people lose money to financial schemes. Historically, the classic degenerate gambling scenario involved a parent squandering their family nest egg, rainy day fund, or kid's college tuition at a physical casino playing blackjack, poker, or slot machines. That evolved into sports betting, offshore websites, and cryptocurrency casinos, all of which still required risking actual starting capital on real assets. In the modern "prop firm" creator economy, however, participants are often not putting capital into real markets at all.

Instead, aspiring traders pay monthly subscriptions—typically around $100 a month per platform—to access simulated SaaS trading environments with virtual money hooked up to live market data feeds. These platforms promise users that if they prove their profitability in simulated "evaluations," the firm will stake them with hundreds of thousands of dollars in firm capital, effectively turning them into mini hedge funds. In reality, retail traders burn out, rack up multiple subscriptions on high-interest credit cards, and repeatedly "blow" their simulated accounts. Many of these entities operate essentially like Ponzi structures, distributing occasional small payouts funded purely by other users' subscription fees rather than genuine trading returns:

> “And turns out that the counterparty is basically a SaaS product that shows virtual trades that are completely simulated.”

The hosts reflect on the surreal contrast between the marketing rhetoric and institutional Wall Street. Content creators post dramatic Reels declaring that they are "working with a prop firm" or that they just "blew an account," giving the illusion that they are trading against titans like Citadel or Jane Street. In truth, institutional market makers are completely absent from these trades; hedge funds would gladly take the other side of these retail orders on real exchanges, but the entire ecosystem is walled off inside proprietary software collecting subscription fees. Because the trading is completely synthetic, subscribers run up credit card debt without ever owning an underlying asset that could conceivably appreciate.

| Dimension | Traditional Market Gambling | Simulated Prop Trading Model | |:--- |:--- |:--- | | **Capital Requirement** | Real money / family nest egg at risk | Monthly software subscription ($100+/mo) | | **Counterparty** | Real market / Exchange / Market makers (Citadel, Jane Street) | SaaS platform running simulated demo software | | **Asset Traded** | Real equities, derivatives, or cryptocurrencies | Virtual paper positions tied to live data | | **Downside Mode** | Total loss of invested principal | Compounding monthly SaaS fees on credit cards with zero asset ownership | | **Platform Revenue** | Spreads, commissions, bookmaking losses | Recurring subscription fees from churned evaluation attempts |

- Simulated trading firms regularly charge approximately $100 per month for entry into trading "challenges." - Some of these proprietary trading SaaS platforms have quietly generated hundreds of millions of dollars in recurring fees at near-pure software margins. - A recurring psychological trap is the "attaboy" feedback loop, where platforms issue gamified praise and occasional payouts funded by customer subscription pools rather than market profits.

[30:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1800s) Market Wrap: Alphabet’s Historic Sell-Off and Sign-Off

The hosts close out the episode by scanning the broader market wreckage, observing that institutional equity markets are facing genuine distress of their own. Tech indices are down across the board:

- The Nasdaq Composite dropped 1.3%. - The Dow Jones Industrial Average slid 0.25%. - Alphabet (Google) shed approximately half a trillion dollars in market capitalization, marking the worst single-day percentage drop in the company’s public history.

The hosts note the accelerating volatility of the modern market, pointing out how headlines heralding either the "best day ever" or the "worst day ever" have become routine as multi-hundred-billion-dollar market cap swings occur in single trading sessions. After reminding listeners to leave five-star ratings on Apple Podcasts and Spotify and giving a quick nod to sponsor Console, the hosts tease Tyler for his brief on-air stint wearing the Jordi mask before signing off until tomorrow's live broadcast at 11:00 AM PT.

## Notable quotes > [10:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=600s) as someone who has personally spent $500,000 a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term. — Max Anderson

> [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) Yeah, it seems like it's going to be incredibly strategic to be able to get frontier intelligence at 90% lower than market prices. — Jordi Hays

## Mentioned TBPN, John Coogan, Jordi Hays, Tyler, Alphabet, Google, Google Cloud, Google Search, Gemini, Gemini for Carbon, Genie, TPU, Larry Page, SpaceX, Warren Buffett, Berkshire Hathaway, Max Anderson, Eric Seufert, Microsoft, AWS, OpenAI, Anthropic, Claude, Sora, Operator, Demis Hassabis, Ashlee Vance, Elon Musk, NVIDIA, Jensen Huang, Alice Wong, Megaspeed, Donald Trump, Supermicro, DJI, Citadel, Jane Street, Nasdaq, Dow Jones Industrial Average, Ramp, Console

## Video description Sign up for TBPN’s daily newsletter at TBPN.com

Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after.

Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella.

TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/

Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive

Chapters: 00:00 Tyler’s Jordi Halloween Mask + The Impersonation Reveal 02:35 Google Stock Plunges 12.5% + Its $561B Cloud Backlog 05:39 Google Search’s AI Ad Problem + The $200B Compute Gamble 12:56 Can Video Games Train AGI? + The AI Simulation Breakthrough 14:52 Jensen Huang’s Alleged Chinese Spy Connection + The Nvidia Chip War 21:33 Viral Day Trader “Loses House” + The Rage-Bait Trading Course Funnel 26:34 Inside Simulated Prop Trading + The $100-a-Month Trading Trap

[Source](https://www.youtube.com/watch?v=1oxs8nUVurI)

中文简报

Alphabet暴跌12.5%：2000亿美元AI资本支出对广告模式构成压力 Alphabet创纪录的市场跌幅反映出华尔街日益加剧的担忧：在生成式AI搜索摘要对其高利润率的广告竞价机制产生蚕食的同时，激进的2000亿美元基础设施扩建可能难以为继。 节目主持人深入分析了Alphabet因每股收益不及预期以及AI搜索变现隐忧而导致的12.5%历史性股价暴跌，并将这一下跌与Google Cloud高达5610亿美元的巨额积压订单及其内部垂直整合进行了对比。他们探讨了基于游戏模拟训练的新兴AI架构，包括Google的Genie以及马斯克的“数字擎天柱”（Digital Optimist）计划。对话审视了一项据称针对黄仁勋的中国情报行动，该行动涉及Megaspeed高管Alice Wong，并分析了国际GPU转运与地缘政治套利的运作机制。最后，他们解剖了通过病毒式传播“亏掉房子”噱头的日内交易漏斗，以及向散户吸纳经常性订阅费却不在真实市场执行交易的模拟自营交易公司构成的掠夺性网络经济。 Alphabet在公布第三季度财报后股价暴跌12.5%，市值蒸发近5000亿美元；由于对其持有的SpaceX非上市股份进行了87亿美元的减记，其每股收益仅为2.12美元，低于预期的3.01美元。 (2:35) Google Cloud的合同积压订单扩大至5610亿美元——其中55%预计将在未来两年内转化为收入——同时其云业务营业利润率达到了36.2%。 (2:35) 美国情报报告揭露，总部位于新加坡的Megaspeed公司的一名高管据称与英伟达CEO黄仁勋建立了密切的个人关系，同时向北京方面输送情报，以协助转运受出口管制的芯片。 (14:52) 面向散户的算法驱动型社交媒体正日益被制造愤怒情绪的交易漏斗和模拟自营公司SaaS平台所充斥，这些平台模仿真实的机构交易，每月收取100美元的循环评估费。 (26:34)

Reading summary

TBPN examines the financial and strategic strains facing major tech players, highlighted by Alphabet's severe post-earnings market drop and reported Chinese espionage targeting Nvidia, alongside an exploration of modern financial traps in the social media creator economy. Alphabet's Earnings Crash and Search Monetization Friction (2:35) Alphabet suffered a sharp 12.5% market drop—its worst single-day percentage decline in public history—after quarterly earnings per share missed expectations at $2.12 against an anticipated $3.01. While a non-operating $8.7 billion write-down of its private SpaceX stake knocked 55 cents off EPS, normalized figures still missed consensus. More critically, search ad growth slowed from 17% in the prior quarter to 12%. Wall Street expressed heightened concern that while AI overviews and Gemini modes drive query and token volume, plain-text AI summaries struggle to monetize user screen space as effectively as traditional high-intent, sponsored link auctions. Simultaneously, Alphabet is pursuing a massive AI capital expenditure ramp, targeting roughly $200 billion for 2026 after posting $51.4 billion in Q3 alone. While physical infrastructure bottlenecks like power constraints and un-racked GPUs led both Google and Microsoft to modestly undershoot quarterly CapEx estimates, Google Cloud remains a strong counterweight. The cloud division delivered $11.1 billion in operating profit with a 36.2% margin and holds a $561 billion backlog, with approximately 55% contracted to convert into recognized revenue within two years. Reinforcement Learning and Virtual Environments for AGI (12:56) The discussion highlights an architectural shift in model development toward reinforcement learning inside simulated worlds and video games. Recent research and industry rumblings indicate that agents trained within physics engines transfer capabilities effectively to complex mathematics and Navier-Stokes fluid equations. Initiatives like Elon Musk's Digital Optimist program and Anthropic's internal Claude checkpoints reflect this pivot, potentially vindicating Demis Hassabis's early hypothesis that gaming sandboxes, rather than brute-force LLM scaling alone, offer a viable path to artificial general intelligence. Google's continued investment in its Genie world model project exemplifies this sandbox approach. Though compute-heavy experimental programs like OpenAI's early Operator or video model Sora initially appeared as costly detours, virtual computer navigation and gaming environments are increasingly viewed as fundamental training beds for generalized autonomous agents. Espionage and Hardware Arbitrage Around Nvidia (14:52) According to reporting from The New York Times, Nvidia CEO Jensen Huang was the target of a Chinese intelligence operation. American agencies revealed that Alice Wong, an executive at Singapore-based Megaspeed and a former Chinese state media reporter, cultivated a close personal relationship with Huang while relaying intelligence to Beijing. Megaspeed faced scrutiny for allegedly diverting export-controlled AI accelerators to mainland China. Intelligence officials briefed President Donald Trump on the findings roughly a year prior, though no executive action was taken at that time, and Huang had not commented publicly. The operation underscores the lucrative geopolitical arbitrage created by trade controls. Because high-end computing components can command massive markups in restricted jurisdictions—analogous to consumer drones fetching ten times their Western retail price in conflict zones—bad actors are incentivized to strip server labels or establish Southeast Asian data centers in hubs like Malaysia and Singapore to route remote inference workloads directly to Chinese entities. The Rise of Synthetic Prop Trading Funnels (21:33) On social media platforms, a new breed of financial marketing exploits algorithmically boosted tragedy narratives to funnel users into recurring-revenue schemes. Analyzing viral videos such as a staged family eviction hook, the hosts note that creators convert negative audience sentiment into paid tiered educational products ranging from free introductory courses to $100 monthly mastermind groups. Beneath this content lies the simulated proprietary trading industry. Unlike traditional retail gambling where individuals risk their own principal on real market assets, users pay recurring SaaS subscriptions (often around $100 per month) to trade paper balances in simulated environments hooked up to live feeds. Promised large funded accounts if they pass evaluation challenges, most retail participants repeatedly blow their simulated balances and incur credit card debt. The platforms effectively operate on subscription pools that distribute occasional payouts to winners from software fee revenue rather than legitimate market profits, isolating participants entirely from institutional market makers. The source notes a broader acceleration of market volatility, where mega-cap technology firms routinely see multi-hundred-billion-dollar valuation swings in single trading sessions, while supply constraints and physical infrastructure delays continue to govern the pace of the global AI build-out.

精炼阅读摘要

本期节目重点聚焦Alphabet财报暴跌背后的云端增长与搜索变现困境、英伟达首席执行官黄仁勋遭中国间谍嫌疑人渗透的报道与芯片走私链条，以及社交媒体上利用悲情营销与模拟交易SaaS收割散户的金融套路。 谷歌财报震荡：云业务与搜索变现阵痛 (2:35) Alphabet在公布最新财报后股价暴跌12.5%，单日市值蒸发约5000亿美元，创下公司历史上最大单日跌幅。尽管其营收达到1270多亿美元，仅略微低于预期的1280亿美元，但每股收益（EPS）仅为2.12美元，较预期的3.01美元大幅下滑近30%。其中，公司对SpaceX等非上市股权投资进行了8.7亿美元的资产减记，直接削减了0.55美元的EPS；但即便排除该项减记，其正常化EPS（约2.67美元）依然不及华尔街预期。更令市场担忧的是核心搜索广告增速从上季度的17%放缓至12%。节目指出，Google传统搜索页面历经十余年优化已将竞价广告位利用至极限，而新推出的AI概览模式以文本摘要为主，尚未建立起高密度的商业链接变现机制。 与之形成鲜明对比的是Google Cloud的强势表现，该业务单季贡献11.1亿美元营业利润，营业利润率达36.2%，积压订单总额达到5610亿美元（其中55%约2750亿美元将在未来两年内转化为确认收入）。尽管Alphabet计划在2026年将AI资本支出推高至约2000亿美元，但受电力基础设施、施工许可及GPU上架延迟制约，Q3实际支出为51.4亿美元，与微软类似均低于预期。节目认为，Alphabet依靠自研TPU芯片、专有模型以及庞大消费级入口构成了全栈垂直整合优势，能以低于市场价90%的成本调用前沿智能，但其商业模式转型期仍需直面巨大的资本消耗与估值重估。 游戏模拟环境：攻坚AGI的新路径探索 (12:56) 节目探讨了前沿AI架构的演进趋势，多方行业消息表明，模型推理能力的近期跃升在很大程度上得益于游戏环境与物理仿真引擎中的强化学习（RL）。在模拟物理系统中训练习得的决策与推理策略，展现出向高等数学研究、流体方程求解及基准测试迁移的能力。Anthropic正在Claude内部版本推进类似方向，而Elon Musk旗下的Digital Optimist项目亦在重点测试从游戏操作向通用计算机操作系统控制的泛化能力。 这种技术路径被视作向DeepMind首席执行官Demis Hassabis最初技术愿景的回归——在经历基于大语言模型的暴力堆叠（brute-force scaling）后，行业正在重新审视将游戏和虚拟模拟作为通往AGI的核心基石。在此背景下，OpenAI此前开发的Operator项目以及Google坚持研发的Genie仿真系统，正从原先被视作消耗算力的边缘尝试，逐步转变为下一代智能体跨领域训练的关键数字沙盒。 黄仁勋涉谍指控与地缘硬件走私套利 (14:52) 据《纽约时报》披露，美国情报机构发现新加坡数据中心运营商Megaspeed高管Alice Wong与英伟达CEO黄仁勋建立了密切私人关系，并涉嫌直接向北京输送敏感动态。Alice Wong曾有中国官方媒体从业背景，而Megaspeed此前已因涉嫌违规将受出口管制的英伟达高端AI芯片转运至中国大陆而受到美方严密监视。情报官员大约在一年前已就相关渗透情况向唐纳德·特朗普总统进行了简报，但当时未采取行政行动。目前，马来西亚与新加坡等东南亚节点被指正成为算力中转枢纽，通过超大规模数据中心建设，使受制裁实体能够通过远程网络访问尖端推理算力。 节目分析指出，地缘政治管制设立的同时催生了巨额的跨境套利空间。以俄乌冲突期间的消费级硬件走私为例，在西方零售店以1000美元购入的大疆无人机，运入冲突区域后售价可达10000美元，单次携带十台即可获得9万美元纯利。类似的高额暴利驱使走私网络采用吹风机清除服务器标签、伪造Supermicro等硬件标识等手段规避海关监管；更为严重的是，利益驱动诱发了针对核心企业高管的直接社会工程学渗透，旨在掌握内部合规程序与出口防规避机制的底层逻辑。 情绪引流与虚拟自营交易SaaS陷阱 (21:33) 节目剖析了社交网络上针对散户的新型金融收割模式。例如Instagram上以“因日内交易导致家破人亡失去房产”为噱头的爆款悲情短视频，配以感伤音乐吸引海量播放后，创作者随即发布“如何赢回一切”的翻盘叙事，最终将流量导入个人主页的阶梯式知识付费产品（从免费试听、49美元进阶课程直至100美元核心会员圈子），利用算法的负面情绪偏好实现流量变现。 更深层次的陷阱在于所谓“自营交易公司”（Prop Firm）的软件化演变。传统的金融投机通常要求投入真实本金并承担市场风险，而当前的许多自营交易平台实际上只是提供连接实时市场数据的模拟交易SaaS软件。平台向用户每月收取约100美元的订阅费，承诺只要通过虚拟“考核”即可获得数十万美元的实盘操作资金。但现实中，普通散户往往因频繁爆仓而背负高额信用卡循环订阅债务；平台实质上依托源源不断的新用户订阅费维持极少量的考核奖励兑现，本质接近庞氏结构。用户自以为在与Citadel或Jane Street等华尔街巨头博弈，实则完全被锁在无实际资产交割的封闭软件体系中。 本期节目展现了前沿科技与资本市场交织下的多重视角：巨头在AI高额资本开支与传统业务变现效率之间面临估值重构；地缘政治对算力封锁与芯片走私渗透的博弈持续加剧；而在大众层面，生成式算法与金融模拟SaaS正在形成更为隐蔽的情绪收割生态。

Detailed notes — zh

## Background 本期内容由科技商业访谈节目 TBPN 于 2026 年 10 月 29 日发布在 YouTube 平台。节目属于 Diet TBPN 精简系列，主打在约 30 分钟内浓缩当日核心科技热点，涵盖谷歌最新财报异动、黄仁勋卷入涉华间谍风波指控，以及社交网络上虚假自营交易员陷阱等话题。 ## TL;DR 本期节目首先深度复盘了 Alphabet 单日暴跌 12.5% 的财报细节，对比了搜索广告因 AI 蚕食导致的增速放缓与其高达 5610 亿美元的云业务积压订单，并探讨了强化学习向视频游戏仿真演进对实现 AGI 的深远意义。随后，主持人针对《纽约时报》披露黄仁勋身边的中国间谍指控与全球芯片走私转运网络进行了全面拆解。最后，节目揭露了社交媒体上利用“炒股破产”悲情营销贩卖课程的套路，并曝光了收取每月 100 美元订阅费、以虚拟撮合收割散户的新型虚假自营交易（Prop Trading）SaaS 骗局。 ## Atmosphere & style 节目开场以团队成员佩戴逼真硅胶面具假冒主持人的万圣节整活恶作剧破冰，气氛幽默荒诞且笑点密集；但随着话题迅速切入华尔街财报大跌与大国芯片情报战，对谈节奏陡然加快，充满犀利敏锐的商业洞察与地缘政治推演。两位主持人配合默契、吐槽辛辣，在拆解社交平台金融割韭菜套路时更是字字珠玑，整期节目既有硬核前沿的深度科技分析，又兼具脱口秀般生动鲜活的临场观感。 ## Bulletin - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Alphabet 股价单日暴跌 12.5%，市值蒸发 5000 亿美元创历史最差纪录；虽然受 SpaceX 股权减记 87 亿美元拖累致 EPS 仅为 2.12 美元，但其谷歌云积压订单已膨胀至 5610 亿美元，且 55% 将在未来两年内加速确认。 - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) 谷歌 2026 年资本支出指引锚定 2000 亿美元且 2027 年将进一步扩大，而核心搜索广告增速放缓至 12%，凸显出搜索 AI 化带来的底层 Token 消耗剧增与商业化广告位转化困难之间的矛盾。 - [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) 《纽约时报》披露英伟达 CEO 黄仁勋成为中国间谍行动目标，新加坡 Megaspeed 算力高管 Alice Wong 涉嫌利用与其密切私交及马来西亚中转通道规避美国先进 AI 芯片出口管制。 - [21:33](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1293s) 社交平台涌现以“炒股破产失去房产”为噱头的愤怒诱饵营销，背后普遍导向分层付费课程以及每月约 100 美元的虚假自营交易（Prop Trading）SaaS 模拟盘订阅陷阱。 ## Key takeaways - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) Alphabet 季度每股收益遭遇大幅下滑，即便剔除 SpaceX 股权投资造成的 87 亿美元账面亏损，调整后 EPS 仍仅有 2.67 美元，显著不及华尔街 3.01 美元的预期。 - [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) 谷歌云业务展现强劲爆发力，营业利润率达 36.2%，总额 5610 亿美元的云积压订单中有约 2750 亿美元（55%）将在 24 个月内转化为实际营收。 - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) AI 交互正蚕食谷歌传统搜索广告的基本盘，搜索广告收入增速从 17% 回落至 12%，主因在于大模型对话页面尚未形成成熟的高毛利竞价广告变现链路。 - [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) 谷歌依托自研 TPU 硬件与前沿大模型构建了深度垂直整合闭环，能够以比市场采购价格低 90% 的超低成本为 YouTube 和搜索生态供给智能算力。 - [10:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=600s) 微软与谷歌的单季 CapEx 均因数据中心带电外壳（Powered shell）施工滞后与审批缓慢等物理世界瓶颈而略低于市场预期，但这在客观上保护了算力代币供给端的利润率。 - [12:56](https://www.youtube.com/watch?v=1oxs8nUVurI&t=776s) 强化学习研究范式正加速迈向游戏仿真环境，行业探索将虚拟游戏经验向纯数学前沿研究（如 Navier-Stokes 方程）及通用计算机操作迁移，戴密斯·哈萨比斯的 AGI 仿真愿景重获重视。 - [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) 美方调查聚焦于将马来西亚作为先进英伟达 GPU 的关键中转跳板，通过在第三方国家极速扩建算力池，迂回为受禁令限制的外部实体提供庞大的 AI 推理算力。 - [20:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1200s) 地缘政治出口管制制造了极高的算力套利空间，促使走私网络通过物理撕标及主动接近科技巨头 CEO 探查合规盲区等手段开展情报与转运博弈。 - [21:33](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1293s) 社交平台利用负面情绪偏差打造“日内交易输光房产”的愤怒诱饵视频，其真实变现路径是兜售 49 美元至 100 美元不等的阶梯式交易大师课信息产品。 - [26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) 新兴虚假自营交易（Prop Trading）已沦为高毛利 SaaS 订阅陷阱，散户按月支付约 100 美元考核费，其对手方并非真实华尔街做市商，而是完全由软件拟真的虚拟盘。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=0s) 开场整活：Tyler 的 Jordi 万圣节面具与替身揭秘 节目刚开场就上演了一出令人忍俊不禁的恶作剧。演播室里的“Jordi”声音听起来有些发闷、反应迟钝，当 John 询问他周末过得如何、是不是去马里布冲浪太累了时，对方含糊地回答自己昨晚在玩赛车模拟器可能有点困，甚至顺便生硬地口播了一段 Ramp 的广告赞助。然而很快真相大白：坐在演播室里的根本不是 Jordi 本人，而是团队成员 Tyler 戴上了一张制作精良却略显诡异的 Jordi 脸皮面具，并配上了标志性的墨镜。 由于面具过于逼真且僵硬，Tyler 在面具下几乎无法自如活动嘴部，也无法用鼻子呼吸，场面既怪异又滑稽。John 调侃道，如果把墨镜摘下来露出假眼睛，绝对会变成“跳杀”惊吓全场，聊天室里的观众更是起哄说 Jordi 像是刚刚打了肉毒杆菌。在把真正的 Jordi 换回演播台之前，John 笑着评价这一幕恶搞完全可以拿满分，但同时坦言：“我们之前测试过这些面具，戴着真的极其难受，感觉整个人都在窒息，简直太要命了。”随后二人将话题拉回当日重头戏：从芯片战中的间谍风波、黄仁勋遭中国间谍刺探的指控，一路延伸到当日让华尔街震荡的谷歌财报。 ### [02:35](https://www.youtube.com/watch?v=1oxs8nUVurI&t=155s) 谷歌股价重挫 12.5% 与价值 5610 亿美元的云业务积压订单 随着财报发布，Alphabet（谷歌母公司）的股价单日大幅跳水 12.5%。虽然季度营收达到 127X 亿美元，仅比市场预期的 1280 亿美元低了约 0.3%，但每股收益（EPS）却遭遇了罕见的重大暴跌，预期为 3.01 美元，实际仅为 2.12 美元，降幅接近 30%。这其中存在一定的会计特殊因素：谷歌持有的 SpaceX 股权在统计周期内出现估值下调，导致其承担了 87 亿美元的股权投资账面亏损，直接拉低了 55 美分的每股收益。然而即便将这部分投资亏损剔除，调整后的 EPS 也仅为 2.67 美元，依然明显低于华尔街最初的期望。 在暴跌的阴影之下，谷歌云（Google Cloud）却展现出了惊人的增长爆发力。早在 2017 年至 2018 年期间，曾有报道披露拉里·佩奇（Larry Page）一度考虑直接关闭谷歌云业务，当时谷歌不仅面临 AWS 的巨大竞争压力，内部两位高管也在为此明争暗斗。而如今，谷歌云已彻底蜕变为一台超级印钞机，营业利润率高达 36.2%。更令市场瞩目的是，随着各大 AI 实验室对算力与 TPU 的疯狂采购，谷歌目前的云业务积压订单（Cloud Backlog）已膨胀到惊人的 5610 亿美元。与部分实验室签署的长达十年的远期合约不同，这笔巨额订单中有 55% 将在未来两年内直接确认为收入，这意味着将有约 2750 亿美元的高毛利营收迅速到账。 - **财报核心数字**：华尔街营收预期 1280 亿美元，实际录得 127X 亿美元（不及预期约 0.3%）；每股收益预期 3.01 美元，实际公布仅为 2.12 美元。 - **SpaceX 投资账面拖累**：股权投资产生 87 亿美元亏损，拉低 EPS 达 0.55 美元；剔除该影响后 EPS 仍仅有 2.67 美元。 - **云业务积压变现速度**：5610 亿美元积压订单中，高达 55%（约 2750 亿美元）将于未来 24 个月内转化为实际营收。 ```mermaid flowchart TD accTitle: Google Cloud Backlog Realization accDescr: 谷歌云 5610 亿美元积压订单中 55% 在未来两年内快速确认 subgraph OrderBook["谷歌云总积压订单：5610 亿美元"] FastTrack["未来两年内确认 (55%)\n约 2750 亿美元"] LongTerm["两年以上长期履约 (45%)\n约 2860 亿美元"] end subgraph Impact["业绩与估值驱动"] Revenue["大幅加速实际营收入账"] Margins["36.2% 营业利润率释放现金流"] end FastTrack --> Revenue Revenue --> Margins ``` ### [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) 搜索 AI 化带来的变现难题与 2000 亿美元的算力豪赌 尽管云业务表现强劲，但谷歌的核心基本盘——搜索广告业务的增速却明显放缓，从上季度的 17% 回落至本季度的 12%。造成这种分化的核心原因在于 AI 模式对传统搜索的蚕食。随着越来越多用户在搜索中直接触发 AI Overview 或切换到 AI 对话模式，底层的 Token 消耗量呈现出爆发式增长，然而商业化效率却大幅滑落。过去二十年里，谷歌将传统搜索结果页面的每一个像素都打造成了极致的竞价广告位；而在 AI 问答界面中，谷歌尚未摸索出如何将每一个生成的句子和段落都转化为竞价广告位的成熟链路。如何让 AI 呈现的内容重新被链接、商业赞助和竞价机制填满，成为了谷歌当下最严峻的商业化攻坚战。 面对激烈的 AI 军备竞赛，谷歌在资本支出（CapEx）上表现得毫不手软。2026 年全年的资本支出指引直接锚定在 2000 亿美元级别，仅第三季度的实际支出就高达 514 亿美元，年化折算已稳稳站在 2000 亿美元的运行轨道上，且官方明确释放出 2027 年开支将进一步扩大的信号。与此同时，Gemini 的月活跃用户数（MAU）已跨越 10 亿大关，第一方 API 处理量达到每分钟 320 亿 tokens（环比激增 40%）。虽然二级市场对庞大开支与短期利润下滑反应剧烈，但两位主持人强调，谷歌凭借从 TPU 芯片、底层基础大模型到全套消费级产品生态的完整闭环，具备行业罕见的垂直整合壁垒。正如 Jordi 在讨论中所指出的：“以比市场价格低 90% 的成本获取前沿智能，这看起来将具有不可思议的战略意义。”这种低成本前沿算力不仅能无缝赋能 YouTube、AI 搜索等内部场景，还能为谷歌在漫长的 AI 消耗战中筑起坚固的护城河。 - **搜索广告增速放缓**：从上一季度的 17% 下滑至本季度的 12%，AI 交互模式对传统高转化广告位造成挤压。 - **算力资本开支激增**：Q3 单季度 CapEx 达到 514 亿美元，2026 年全年指引奔向 2000 亿美元，并预告 2027 年将进一步大幅增加。 - **Gemini 生态数据**：月活跃用户数突破 10 亿大关，第一方模型 API 处理量达每分钟 320 亿 tokens，季度环比上升 40%。 - **垂直整合护城河**：依托自研 TPU 硬件与自研前沿模型（最新公布 Gemini 4 Carbon），能以远低于公开市场采购达 90% 的成本为 YouTube、内部搜索提供超低延迟定制化服务。 ### [10:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=600s) 谷歌股价下挫、搜索广告的“恶化”与云厂商资本支出延迟 主持人讨论了 Alphabet 的股价走势。Alphabet 股价从大约每股 350 美元（拆分后分属 Class A 和 Class C 股票）一路下跌至大约 280 美元，在短短六个月内出现了相当戏剧性的回调。主持人指出，这表明最初做空谷歌的逻辑（OG Google Bear case）如今又重新抬头。网络上出现了一篇广为流传的帖子，Max Anderson 在 7 月 23 日发文称：“作为一个多年来每月在 Google Ads 上亲自花费 50 万美元的人，我可以肯定地告诉你，搜索领域的这种收入增长是人为制造的，对谷歌的长期业务极不健康。”（"As someone who has personally spent $500,000 a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term."） 搜索广告体验的持续恶化已成为老生常谈。如今在谷歌上进行特定搜索时，用户很难再找到真正的自然搜索结果（Organic result）：顶部前三到四个结果全是付费广告，中间插入一两个自然结果，紧接着又是更多的广告。以“买床垫（Mattress）”为例，页面首先弹出付费的 Google Shopping 橱窗，接着是三到四个针对具体品牌的赞助商关键词广告；若往下滑动试图寻找评测，会发现那些床垫评测网站其实全都被各大床垫公司收购了，用户根本找不到维基百科或中立客观的第三方信息。不过，由于用户的搜索使用习惯极具黏性（Sticky），习惯路径依旧固化，因此谷歌目前看到的仅是轻微的业务减速，而不是直接跌落悬崖。 另一方面，微软与谷歌的最新财报披露了各自巨额的资本支出（CapEx）： - **微软**：资本支出为 383.7 亿美元，低于市场预期的 459 亿美元。 - **谷歌**：资本支出为 513.6 亿美元，低于市场预期的 557 亿美元。 主持人强调，两家巨头资本支出低于预期，并不是因为他们不愿意花这笔钱，而是因为遭遇了外部交付延期。包括数据中心带电外壳（Powered shell）施工滞后、施工许可审批缓慢等物理世界中的种种壁垒，导致大量英伟达 GPU 只能等待上架。但这对于算力代币（Token）的卖家来说并不算世界末日，因为在供给受限的情况下，他们的利润率反而得到了保护。 --- ### [12:56](https://www.youtube.com/watch?v=1oxs8nUVurI&t=776s) 强化学习新前沿：通过游戏仿真迈向 AGI 的 Hassabis 愿景 在 AI 模型训练与架构演进方面，业内近期传出消息：前沿大语言模型内部推理能力的跨越式提升，很大程度上归功于更激进的强化学习（Reinforcement Learning, RL），尤其是通过视频游戏和仿真环境进行训练。仿真假说（Simulation hypothesis）正在加速落地，模型在游戏与仿真环境中所习得的策略，正在被成功迁移到纯数学研究（例如 Navier-Stokes 方程与 ETH 相关问题）等尖端领域。Anthropic 内部的 Claude 模型也在探索类似的突破方向，而埃隆·马斯克旗下的“数字擎天柱”（Digital Optimus）招聘需求中，也明确要求研究如何将游戏中的操作经验迁移至通用计算机操作（General computer use）。 科技作家 Ashley Vance 在其专栏 Core Memory 中提出疑问：这是否意味着行业在经历了短暂的蛮力穷举（Brute force）之后，又重新回到了戴密斯·哈萨比斯（Demis Hassabis）最初关于实现通用人工智能（AGI）的愿景——即“游戏才是通往 AGI 的康庄大道”？谷歌在这一方向上始终坚守自研的通用世界模型 Genie，而与此同时其他实验室则采取了不同的侧重点： - **Anthropic**：战略极度克制，专注于代码生成，甚至刻意不做图像生成。 - **OpenAI**：尝试了大量被戏称为“支线任务（Side quests）”的项目，部分任务如 Sora 消耗了极其巨大的算力，而另一些项目（如早期的 Operator 虚拟桌面点击自动化工具）当时虽然粗糙笨拙，如今却已接近解决，被证明是具有战略价值的探索。 主持人们讨论指出，谷歌对 Genie 的投入究竟会沦为一个昂贵的炫技玩具，还是会成为其内部极具价值的训练飞轮资产，是当下 AI 圈讨论的核心焦点。 ```mermaid flowchart TD accTitle: AI训练范式转移与跨领域能力迁移 accDescr: 展示强化学习从仿真环境与游戏训练向通用科学与计算机操作迁移的路径架构 subgraph Simulation["环境与训练输入"] G["视频游戏 / 仿真环境"] Genie["Google Genie 动态世界模型"] end subgraph RL_Core["强化学习推理跳跃"] RL["更激进的强化学习 (Aggressive RL)"] end subgraph Transfer["跨领域策略迁移 (Cross-domain Transfer)"] Math["数学与物理突破 (Navier-Stokes / ETH 研究)"] OS["通用计算机操作 (Computer Use / Operator)"] Robotics["具身制造与物理迁移 (Optimus 制造落地)"] end G --> RL Genie --> RL RL --> Math RL --> OS RL --> Robotics ``` --- ### [14:52](https://www.youtube.com/watch?v=1oxs8nUVurI&t=892s) 纽约时报曝英伟达间谍风波：黄仁勋身边的 Alice Wong 与马来西亚算力谜团 《纽约时报》当天刊发的一篇重磅调查报道引爆了科技圈：英伟达首席执行官黄仁勋（Jensen Huang）被指成为一桩中国间谍行动的目标。报道称，新加坡某数据中心公司的女性高管 Alice Wong（中国公民）与黄仁勋建立了极其密切的私人关系。美国情报机构经调查认定，她实际上是一名向北京输送情报的特工，且正受到美方审查，怀疑其暗中协助将受禁令管制的先进英伟达 AI 芯片转运至中国。 面对这一发现，白宫警铃大作——因为北京居然能够从一个与 AI 革命最具权势的领袖拥有直接接触渠道的人那里获取第一手汇报。报道披露，大约一年前，唐纳德·特朗普就已听取了该情报的简报，但在获知后并未采取任何后续行动。 TBPN 主持人回顾了他们在 10 月 10 日的节目，当时他们就深入拆解过这家名为“Megaspeed”的公司及其掌门人 Alice Wong。主持人当时就对该公司的背景提出过质疑： - Alice Wong 在创立这家业务横跨马来西亚和新加坡的超大规模算力企业之前，曾是中国官方媒体的记者。 - Megaspeed 成立后以极快的速度崛起并获取英伟达芯片配额。 主持人利用 ChatGPT 对比了自己两周前的言论与最新报道，调侃自己的洞察力经受住了考验：当时他们明确指出了 Alice Wong 的官媒背景及其接触黄仁勋的罕见通道，甚至质疑过 Megaspeed 的真实商业图谋，只是当时缺乏细节，没有直接点明她是“间谍”。 这一事件的深层核心在于全球算力走私与转运网络。美方的控诉重点在于她涉嫌打通渠道，将大批先进英伟达芯片运往马来西亚等中转站；因为进驻马来西亚的算力越多，实际上能够迂回为中国所用的 AI 推理（Inference）算力池就越大。主持人引用了乔治·W·布什的名言打趣道：“在田纳西州有句老话，得克萨斯大概也有：欺骗我一次，是你可耻；欺骗我两次……好吧，你休想再骗我一次。”（"There's an old saying in Tennessee. I know it's in Texas, probably in Tennessee. That says, fool me once. Shame on. Shame on you. If you fool me, we can't get fooled again."） 截至节目播出时，黄仁勋本人对此事尚未作出任何公开回应。主持人认为，随着调查公开，美国中情局（CIA）势必会介入干预，切断高管与情报目标之间的私人接触。同时，全球各国的算力分布图表（Compute by country）俨然已成为“煤矿里的金丝雀”：如果马来西亚这样一个本地经济体量并不算巨大的国家，其机房计算能力却在反常地极速膨胀，就如同“在运动员身上测出了离谱的血糖指标”，哪怕尚未正式确诊，外界也已经清楚其中必有蹊跷。这预示着美国围绕先进制程芯片的出口管制与制裁法案很可能会迎来新一轮的大幅修补与收紧。 ### [20:00](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1200s) 芯片走私暴利、地缘政治套利与针对英伟达的间谍战 主持人讨论了芯片走私分子为了绕过出口管制而采取的极端手段，包括使用吹风机撕掉服务器标签以便跨境运输。有人指出涉事公司正是超微电脑（Supermicro）。这种行为的核心驱动力在于巨大的地缘政治套利空间。主持人用俄乌冲突中的真实案例做类比：有人买头等舱机票飞往西方国家，在 Best Buy 以 1,000 美元一台的价格扫货大疆（DJI）无人机塞满行李箱，然后飞往邻国搭乘火车或巴士进入冲突区，转手就能以每台 10,000 美元的高价卖出。跑一趟带 10 台无人机，就能净赚 90,000 美元暴利。 英伟达（NVIDIA）的 GPU 同样存在这种暴利套利空间。因此，各方不仅有极大的动力去窃取或转运货物，更有极强的动机去接近其首席执行官（黄仁勋），试图探查英伟达内部防止芯片转运的真实流程，从而找到漏洞加以规避。正如主持人所指出的：“So there's going to be a huge incentive, not just to steal them and divert the cargo and do whatever you can, but also get close to the CEO, see what you can do, learn what is their actual process for making sure that chips aren't diverted, right?” 这种高风险博弈印证了 AI 2027 报告中所描绘的地缘政治紧张局势——大国早已全面苏醒，更激进地开展情报战与间谍活动。 - 超微电脑（Supermicro）服务器被曝通过吹风机撕标等手法规避芯片出口禁令。 - 走私套利模式：以 1,000 美元购入消费级无人机，运往冲突地区后以 10,000 美元转手，单趟净利达 9 万美元。 - 高端 AI 算力芯片由于高额价差，面临着类似的跨国走私与情报渗透风险。 --- ### [21:33](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1293s) 社交媒体“炒股破产”悲情营销与信息产品引流套路 话题随后转向社交平台上泛滥的金融博主营销陷阱。两位主持人谈到他们的 Instagram 推荐算法完全同频，几乎同时刷到了同一个爆款短视频。该视频配着低沉哀伤的萨克斯背景乐，标题带有明显的语法错误——“Losing Our House Because of Day Trader”（因为日内交易者而失去我们的房子），画面展示了一家人被迫搬离家园的惨状。这种本应是家庭毁灭的人间悲剧，却被包装成了极具视觉冲击力的“愤怒诱饵”（Rage Bait）。 然而，一旦深入挖掘该账号的主页，就会发现紧随其后的下一条视频口风突变，变成了“我日内交易输光了房子，以下是我打算如何全部赢回来的计划”。主页链接直接导向所谓的“交易大师训练营”（Trading Mastermind），里面设置了精巧的变现漏斗：先是一个免费课程，接着是 49 美元的进阶课，再到 100 美元的系统课。主持人一针见血地指出，这完全是利用社交媒体负面情绪偏差（negativity bias）制造爆款，进而收割受众的信息产品漏斗。该博主究竟有没有赔掉房子根本无从查证，甚至可能因为卖课赚得盆满钵满而过得比以往任何时候都好。 ```mermaid flowchart TD accTitle: 社交媒体炒股悲情营销漏斗 accDescr: 展示博主利用破产悲情视频引流并变现信息产品的完整转化路径 subgraph Traffic["流量捕获阶段"] A["爆款悲情视频<br/>'因日内交易输掉房子'"] --> B["算法推荐与情绪煽动<br/>利用负面情绪偏差引流"] end subgraph Retargeting["人设反转阶段"] B --> C["励志翻盘视频<br/>'看我如何把房子赢回来'"] C --> D["个人主页外链引流<br/>Trading Mastermind"] end subgraph Monetization["分层变现阶段"] D --> E["免费试听课程"] E --> F["49 美元初阶课程"] F --> G["100 美元高阶大师班"] end ``` - 视频博主以失去房产的人间惨剧作为视觉诱饵，依靠算法机制实现病毒式传播。 - 转化漏斗设计：悲情视频引流 $\rightarrow$ “绝地反击”励志人设 $\rightarrow$ 主页挂课（免费体验 $\rightarrow$ 49 美元 $\rightarrow$ 100 美元大师课）。 - 这种反常识的营销手段利用了大众“想看失败者翻盘”以及“不想在成功前夕放弃”的赌徒心理。 --- ### [26:34](https://www.youtube.com/watch?v=1oxs8nUVurI&t=1594s) 虚假自营交易公司（Prop Trading）与每月 100 美元的订阅陷阱 主持人深入剖析了这类日内交易陷阱背后的底层机制。在过去，赌博败家通常意味着父亲在实体赌场输光了养老金或孩子的大学学费；后来演变成智能手机上的体育博彩和加密货币赌场。但无论哪种，参与者都必须拿出真金白银作为本金去冒风险。如今在社交平台上泛滥的所谓“自营交易”（Prop Trading）则完全不同，普通人并没有把钱投入真实的股票市场，而是掉入了一种基于虚构市场的 SaaS 订阅陷阱。 大量的平台打着“模拟自营交易挑战”的旗号，宣称用户只需每月支付约 100 美元的订阅费，在接盘真实行情数据的模拟盘里证明自己的交易水平。平台承诺，只要交易员考核达标，就会提供真实的自营资金，把用户打造成管理巨资的“小型对冲基金”。然而，许多这类型平台本质上是庞氏骗局：它们靠收取每月 100 美元的考核订阅费赚取了数亿美元的收入，利润率远超普通软件公司。更有甚者，这些平台会随机挑出部分用户给予虚假的“盈利返佣”奖励，通过间歇性强化让人上瘾，促使受害者绑定信用卡开通多个高额订阅账户。 许多短视频创作者声称自己“打爆了自营交易账户（blew an account）”，受众往往误以为他们在与华尔街顶级对冲基金（如 Citadel 或 Jane Street）对弈。主持人强调，真正的华尔街对冲基金恨不得这些业余散户进入真实市场，好作为做市商赚取交易差价；但这些受害者的真实交易对手方只是一款提供虚拟撮合的 SaaS 软件。正如主持人剖析的那样：“And it turns out that the counterparty is basically a SaaS product that shows virtual trades that are completely simulated.” 用户把成百上千美元的订阅费刷在信用卡上，本金根本没有进入真实股市，连蒙中暴涨股票的理论机会都没有，最终落得负债累累。 - 传统赌博与虚假交易对比：传统赌博损失的是真实储蓄与本金，而新型虚假交易是通过持续支付软件订阅费被持续放血。 - 虚假自营交易平台商业模式：依靠每月 100 美元的“模拟考核订阅费”聚敛数亿美元，毛利率极高，且多数具有庞氏骗局特征。 - 虚假撮合机制：普通散户以为自己在与 Citadel 等量化巨头交手，实际对手方仅为一个完全拟真的虚拟 SaaS 平台。 - 宏观市场速览：纳斯达克指数下跌 1.3%，道琼斯工业指数微跌 0.25%，谷歌母公司 Alphabet 市值单日蒸发 5000 亿美元，创下该股历史上最差单日表现。 ## Notable quotes > [05:39](https://www.youtube.com/watch?v=1oxs8nUVurI&t=339s) 以比市场价格低 90% 的成本获取前沿智能，这看起来将具有不可思议的战略意义。 — Jordi Hays ## Mentioned Alphabet, Google, Google Cloud, Gemini, Genie, YouTube, SpaceX, NVIDIA, Jensen Huang, Alice Wong, Megaspeed, Microsoft, Anthropic, Claude, OpenAI, Sora, Operator, Supermicro, DJI, Elon Musk, Demis Hassabis, Larry Page, Max Anderson, Ashley Vance, Donald Trump, George W. Bush, Citadel, Jane Street, Ramp, Tyler, John Coogan, Jordi Hays ## Video description Sign up for TBPN’s daily newsletter at TBPN.com Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive Chapters: 00:00 Tyler’s Jordi Halloween Mask + The Impersonation Reveal 02:35 Google Stock Plunges 12.5% + Its $561B Cloud Backlog 05:39 Google Search’s AI Ad Problem + The $200B Compute Gamble 12:56 Can Video Games Train AGI? + The AI Simulation Breakthrough 14:52 Jensen Huang’s Alleged Chinese Spy Connection + The Nvidia Chip War 21:33 Viral Day Trader “Loses House” + The Rage-Bait Trading Course Funnel 26:34 Inside Simulated Prop Trading + The $100-a-Month Trading Trap [Source](https://www.youtube.com/watch?v=1oxs8nUVurI)

Source quotes

[5:39] Jordi Hays: 以比市场价格低 90% 的成本获取前沿智能，这看起来将具有不可思议的战略意义。

Source quotes

[10:00] Max Anderson: as someone who has personally spent $500,000 a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term. [5:39] Jordi Hays: Yeah, it seems like it's going to be incredibly strategic to be able to get frontier intelligence at 90% lower than market prices.

Original transcript

[0:00] Why don't you sound weird today? Are you okay? What are you doing this weekend? Were you surfing a lot? Oh, I was surfing this weekend down in Malibu, having a good time down there. That's good. Yeah. What do you mean? It seems like you're not sharp. Do you want to sound different or something? Yeah, something's a little different. I don't know where to place it, but let me tell you about RAMP. I'll try and figure it out. Time is money saved. Both is easy as corporate cards, bill pay, accounting, and a whole lot more. I was playing on my racing sim last night, so maybe I'm a bit tired. Okay. Yeah, that must be it. That must be it. How's your phone doing? Oh yeah, my phone, it's feeling pretty good. What the heck? Who, what the? Who is that? It looks like Tyler is wearing a perfect Geordie mask. Because look at Tyler today. He looks exactly like Jordy. This is crazy. I'm so confused by what's going on right now. No, of course... Of course, Tyler is wearing the the Geordie mask, which is extremely freaky. When I ran into him this morning, I thought-- I didn't clock it as Geordie, actually, initially. I just thought it was a guest who was here to hang out or say hi. But it is bizarre. You can't really move your mouth that much, right? That's the main thing. - I can't really move my mouth. I can't breathe through my nose, either. It's a bit hard. It's pretty uncomfortable to wear. - But do a little spin, because the back is pretty solid. They nailed the hair. I think they got it pretty good back there. And I think the sunglasses do a lot of the work. If you take out the sunglasses, you'll give everyone a jump scare. Yeah. The eyes are. Not good. I've got to keep those. Not good. You've got to keep the sunglasses. But it's helpful because Jordy wears the sunglasses a lot on the show. So I think we're going to sub the real Jordy back in. But this is an A+ bit. I thoroughly enjoy this. Tyler forgot your laptop. Keep it on for the whole show. Cool? Do not do that. No, keep it on for the whole show. We've tested these masks before and they are so uncomfortable. It feels like it's suffocating. It's absolutely, absolutely brutal. But... It's a lot of fun. The chat's saying you got Botox over there. Anyway, chip wars going on. There's another spy. There was a spy at OpenAI. Now there's a spy tracking Jensen over at NVIDIA. The wars are waging. Google earnings is today. We have Ice Cube joining. We got lots of stuff going on today. Can we start a little Google earnings recap? And then we'll go into the NVIDIA story. What do you think, Jordan? Yeah, let's do it. Cool. So Google stock is cratering. It's down 12 and a half percent. Earnings miss rare. And here are the results. So revenue consensus was one hundred and twenty eight billion came in at one hundred and twenty seven point. Like only a third of a percent miss down, but to the negative, you usually want to be beaten these consensus estimates. Earnings per share, not nearly as good. 30% miss. They were expecting $3 a share, $3.01, came in at $2.12. And so there was a little bit of nuance there because some of the earnings miss was related to owning SpaceX stock, which traded down in the interval, and so they had to take a little bit of a write down on that. But EVEN IF YOU FACTOR THAT OUT, SO THAT WAS AN 8.7 BILLION LOSS THAT THEY TOOK ON THE EQUITY INVESTMENTS. THAT REDUCED EPS BY 55 CENTS. SO EVEN IF YOU EXCLUDE THAT IMPACT, EPS STILL WOULD HAVE BEEN $2.67, still below the $3.01 that Wall Street was expecting. So you can factor out that nuance and still get to a place where something went, slightly awry. There's a lot of good stuff happening, a lot of other worries, and that's why the market's reacting the way it is. So the first key takeaway is that cloud is just a monster business. There was very interesting reporting years ago in 2018 that back in 2017, Larry Page had considered pulling the plug on Google Cloud. He was like, we're competing with AWS. There's two executives that were sort of fighting at the time. There's too much going on. Maybe we shouldn't even be in this business. Maybe we shouldn't even do cloud and have a cloud platform. Well, it's a completely different world today because cloud is so important to the business. to $11.1 billion. Operating margin is at 36.2%. It's a fantastic business. If it were independent, it would be worth it. an incredible amount of money, hundreds of billions, maybe a trillion dollars right there, just in that value. And so Google and Alphabet truly has become this diversified portfolio of bets. Search is just one of them. They got YouTube. They also have cloud, obviously. So the cloud backlog now is $561 billion of companies, labs that have gone to them. We just saw this with OpenAI buying cloud and TPUs. [5:00] There is a lot of demand for their services. And what's interesting is that you can look at that number 561 billion. That's a ton. Is that over 10 years? You know, we saw in the Anthropic S1, some of these contracts go out, you know, to 2033, 2036. But 55% of that cloud backlog is expected to be recognized as revenue within two years. So even though you have this big number, you're getting half of that 275 billion, something like that, over the next two years. That's pretty quick for a business that did $30.7 billion in revenue. You're getting $250 coming really fast. So that should be driving valuation upwards. Search advertising, though. Growth slowed. Last quarter was 17%. This quarter came in at 12%. AI is accelerating query volume. So they are seeing a ton of token production. You see that when Sundar steps on stage and shows the progress of their models, the diffusion of Gemini models. query AI mode is actually taking share. People are using it in place of the default search, not really by their choice a lot. It just sort of pops up a lot of the times. It monetizes less effectively than traditional searches right now What interesting is that at the end of the day it a person asking a question and getting a bunch of pixels on a screen And if you on your phone it a vertical screen that you know nine by 16 right Square for some. Square for some if you're on the duo. You're Mark Garman. You're Mark Garman. But at the end of the day, if you're Instagram, you want to monetize every pixel that you're showing the person while they're on Instagram. If you're Google, when someone asks a question, you want to monetize every pixel with ads. And that's what has happened with Google search over the past decade, right? There was a time when it used to be a very clear call out for one sponsored post at the top, and now it's been, you scroll the whole first page and it's all ads. AI mode isn't that. AI mode might have some monetization going on, but it's not every single sentence is a sponsored ad just yet. They gotta get there. They're not gonna talk about that because that's rough and that's complicated, but it's gonna be a hard road to figure out exactly how do we monetize the full response, Everything that we're showing to the viewer should have a lot of things, a lot of links and offers and brand messages that are monetizable, paid for, bidded auction style to really maximize every view that they get. And so that is the coming challenge, but they been doing this for 20 plus years And so that is the path that they on The last thing Google spending billion on AI infrastructure this year 2026 CapEx guidance to billion right in that billion range They are also signaling that they expect a significant increase in 2027. Q3 spending alone was 51.4. So they're already on their $200 billion run rate. and next quarter should be higher, following quarter should be higher, et cetera. So let's go. And then the last thing is that Gemini crossed a billion MAO, monthly active users. The first party AI model, the API is doing 32 billion tokens a minute. Hard to say what that means in the modern era because there's so many models that are more token efficient and really just getting stuff done, getting answers, satisfying users is what matters. But it's up 40% quarter on quarter. Clearly they're serving their models more people are using them. And then of course they announced Gemini for carbon, the newest frontier model, the day or before earnings, two days before earnings, sort of trying to reassure investors that they are still on the frontier, that they that the benchmarks look good and that the adoption will come And more importantly that they are vertically integrated So even if they don show up on the ramp data and the open router data just having the frontier model internally is super valuable because it will drive YouTube, it will drive AI search, it will drive internal workloads. There will be all these things where they aren't going outside for costs and they will be vertically integrated. Yeah, it seems like it's going to be incredibly strategic to be able to get frontier intelligence at 90% lower than market prices. Exactly, exactly. And then there's a whole bunch of other interesting things. They can fine-tune their own models. They can optimize them for their infrastructure. The model that works in Google search, it does feel different than a Claude model or a GPT model because it needs to be really fast because people expect that from the AI search results. And then on YouTube, you expect something different. You expect something that can piece, can crawl over all the transcripts of YouTube with no, you know, robot block up, no captchas and whatnot if you're asking for extra context. So lots of optimism there, but rough day in the market at least. And a little bit rough for Berkshire, who, their sort of blended price per share of their recent $10 billion investment is around $3. [10:00] $350 per share. They did some part into class A, some into class C, down pretty dramatically from that point. Alphabet now sitting at $280 a share in just around six months. Got it. And yeah, it's basically like the OG Google Bear case is like rearing its head again. Yeah, that's right. And, you know, they've been... It became sort of a boring thing. There was a viral thread. I'll try to find it. I think you're talking about from Levels.io, maybe? No, this was basically this guy, July 23rd, Max Anderson, says, as someone who has personally spent $500,000 a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term. He goes on to make a bunch of points, which have been refuted in different ways, but Satrini brought it up yesterday. The Soufinator at the time pushed back. on a lot of this, but either way, it just felt like search was already perfectly optimized. I think I was bringing this up. probably, you know, over a year ago, like, have you actually paid attention to a Google search recently, where it's like, for certain queries, it's hard to find, you know, an organic result, right? Where it's like, ads at the top, first three results are ads, then you have maybe one or two organic results and then a bunch more ads. I mean, do you understand the structure of like, finding a mattress on Google, it got so crazy that, first off, there would be Google Shopping call-outs that are also paid for, there would be a Google Shopping widget, then there would be three or four sponsored results. So the links would be, you know, specifically this brand, specifically this brand, just directly brand keyword advertising. But then you'd go down and you'd get to mattress review sites, And they'd all been purchased by the various mattress companies. So there was actually nowhere to end up. There was nothing that you could go to. There was no Wikipedia site. And yeah, it's been It's been crazy, but this stuff's just sticky. People are just in their flows still. So that's why you only see slight deceleration. It's not like falling off a cliff or anything like that. Anyway. One other interesting detail, so Microsoft had earnings as well. CapEx came in at 38.37 billion. Estimate was 45.9 billion. So Google was 51.36 billion. Estimate was 55.7. This wasn't because they didn't want to spend the money, it was because of delays. So they're trying to spend the money. Powered shell delays. Yes, just a variety of different factors. Yeah, this whole rumor of like, There's a whole bunch of GPUs that are waiting to get racked because stuff is slow and there's permitting, and you're bumping up against the real world, the physical world. It doesn't really matter where the demand is. AND Everyone who's selling tokens is pretty happy to be like, yeah, we wouldn't hate if our margins were good this quarter, even though we were delayed on supply. Like, supply constraint isn't the end of the world for most people. On the AI build-out and strategy specifically, I'm just new-tied? I'm just new at AI, I guess, has a breakdown of some rumors saying that the internal model's reasoning jump comes from more aggressive RL, including training in games and simulated environments. Simulation hypothesis is... going strong. strategies. learned their transfer to research mathematics and other fields Navier Stokes and ETH. show the scale of the improvement in anthropics in some-- similar direction with its internal Claude model. Musk's digital optimist hiring explicitly calls for manufacturing, measuring transfer from games into general computer use. And so Ashley Vance over at Core Memory is saying, does this mean we're just in the midst of Demis Hassabis' original vision of how to get to AGI. After a kind of brief, brute force period, the games are the path to AGI. And it is interesting that Google has stayed the course with Genie when a lot of the other labs, I mean, Anthropic has famously been like, we're only focused on coding, we're not even doing images. OpenAI has done a bunch like, quote unquote, side quests. Some of the side quests were too compute intensive, like Sora. Some of them, like computer use, remember operator? Do you remember Operator? Like it was the original, like go and use a computer in a virtual computer to go and click on buttons on a computer. website and it was pretty janky at the time, but now it's basically solved and that was actually a good investment on the side. And it will be interesting to see which way Genie goes. It could just be, oh, it's only fun and it's really expensive or it could be like, hey, this is a really, really great next leg up for training potential and it's a really valuable asset for them to have internally. I think that's what people are debating right now. Do you want to stay on Google or move on? NVIDIA world. Yes. In the times today, NVIDIA chief seen as target of Chinese spy plot, a Chinese [15:00] Businesswoman who had a close relationship with Jensen was found to be passing intelligence to Beijing. President Trump was briefed and took no action. That's interesting. American spy agencies had gathered striking intelligence, a top executive at a Singapore-based data center, The company was under scrutiny for possibly steering advanced NVIDIA AI chips to China, and now American officials had concluded that she was a Chinese intelligence asset funneling information to Beijing. But what came next was even more alarming. Surveillance of the executive Alice Wong, a Chinese citizen, revealed that she had developed a close relationship with Jensen. For years, Washington had worked to deny China the technology needed to match or surpass the United States in the intense global competition for superiority in AI, The discovery that Beijing was getting reports from someone with direct access to Mr. Huang, one of the most powerful figures in the AI revolution, as well as an influential ally, President Trump, set off alarms at the White House. Roughly a year ago, Mr. Trump was briefed about the intelligence findings, which have not been previously reported and chose not to do anything with the information. So Good opportunity to rewind and talk about a show that we did on October 10th where we talked about this company and the CEO, Megaspeed. Megaspeed. We called it a great name. I mean, it is a great name. And I think we gave her a round of applause for being a reporter. The name is fantastic. It's always good to say something nice. Yeah. So the name is fantastic. Totally. So I asked Chad to kind of summarize our discussion. Yes. How did our takes hold up? And I did a question and a finding. discuss Alice Wong, the CEO accused by. Yes. Did you discuss her access to Jensen? Yes. Did you highlight her Chinese state media background? Yes. So she had been a reporter for Chinese media. state media prior to starting a sort of Malaysian Singaporean hyper scale. Is there non-Chinese state media? Did we question Megaspeed's activities? They don't have independent media, right? So if you do anything in the media, it's by definition you're working for the government? No, I mean, I don't think that's like actually how they're defining it. There's like actual state media. Like being like a press secretary. Yeah, you can make the argument that all their media is going to come under the influence of the state. Did you discuss Alice Wong? Yes. Did you discuss her access to Jensen? Yes. Did you highlight her Chinese... State media background? Yes. Did you question Megaspeed's activities? Yes. Did you explicitly suggest she was a spy? No. Could the discussion be interpreted as raising suspicions? Yes. It was too early to go out and say we didn't have any... real details around the situation, but the whole situation was so suspicious. I'm surprised it's taken this long for it to kind of bubble up. Yeah, yeah, no, it's very, very tricky. Yeah, I mean, it seems like at the very least, it's like, she is being accused of, you know, being deployed to figure out how to gain access to advanced AI chips. So how can more NVIDIA chips land in Malaysia, right? Because the more chips that get to Malaysia, the more inference that will be available for China. essentially. Matthew's in the chat pretty much saying what we're all thinking. Fool me once, shame on you. Fool me twice, you get fooled. Can't get fooled again. It's a great quote. George W. Bush. One of the best quotes in White House history. Yeah. There's an old saying in Tennessee. I know it's in Texas, probably in Tennessee. That says, fool me once. Shame on. Shame on you. If you fool me, we can't get fooled again. Well, Jensen has not responded to this. He has not actually posted since October 21st. But here's the thing. October 26th. Here's the thing. Have they had interactions over the last... We've been pictured together, right? Over the last year, though. Oh, interesting. This might be like kind of old. I think it might be old news. Yeah, yeah, yeah. You'd imagine if like the New York Times is on the hunt that like the CIA would also be like, "Hey, you gotta stop talking to this person. They are trying to get information out of you." Something like that. I don't know. It'll be interesting to see what happens with mega speed with Malaysia in general. Like it's kind of a canary in the coal mine, that chart of compute by country. if Malaysia continues to ramp compute capacity, it's like... We don't know the cause, but we know that there's a problem. If you're measuring like an athlete and you're noticing like insane blood glucose levels or something like you might notice that there's a problem before you can diagnose it Malaysia seems to be in that camp of like, it's, I don't know. if they're scaling dramatically. Yeah. given the local economy. It's suspicious. Yeah. And so I don't know. I mean, there's there's a whole bunch of chip sanctioned chip control legislation that could be expanded, reworked, changed. There's also I mean, the incentive is just so high. We've heard about people taking [20:00] hair dryers and removing labels from servers to get them across the across borders. It's an extremely valuable-- That was called Supermicro? I think it was Supermicro. But anytime that there's these geopolitical there's just such an incentive for arbitrage. I remember hearing about people THAT, YOU KNOW, IT'S NOT in the Russia-Ukraine war would take basically like a first class fight to a Western democracy, buy a DJI drone at like a Best Buy for $1,000 or buy a bunch of them, put them in their suitcase, fly them into the closest country, hop on a train or a bus, get into the conflict zone, and be able to sell them for like $10,000 a piece. And so that arbitrage is just like, okay, one trip, if I have 10 drones with me, I make $90,000 in profit. People do that. It's the same thing with NVIDIA GPUs. So there's going to be a huge incentive, not just to steal them and divert the cargo and do whatever you can, but also get close to the CEO, see what you can do, learn what is their actual process for making sure that chips aren't diverted, right? How can we mitigate the diversion as much as possible or divert as much as possible, right? And so it's an ongoing battle. But another example of the high geopolitical stakes of the AI race is this. moment that AI 2027 was talking about. China wakes up. wakes up and starts spying, like starts going more aggressively. Well, wouldn't that... They've been doing this for a while. I guess they've been awake for a couple of years then, yeah. Should we go down the rabbit hole? Yes, we should. Let's see. Simulated markets, the simulation hypothesis. So I went down this rabbit hole a couple of days ago because You got served this video too on Instagram, right? You saw it. - Well, we have the exact same feed. - We have the exact same feed. For some reason, Instagram doesn't show us any ads, and sends us the exact same videos at the same time. I mean, this is just the nature of things that go viral, everyone sees them. But let's go on-- No, but we specifically are sending posts back and forth all day. So it's like you guys like the same thing. So there's this video. Yeah, we have the-- we'll each send the same video. Yeah. to each other. I'm just like, we're just like, scroll up. Two ships passing in the night. Everyone says the same video. But there's this video-- Two reels passing in the night. --that has the title "Losing Our House Because really day really Tyler what happened? I had to take it off. Oh, you had to take it off. Wow. Sitting here in the Capitol. Of show business. Yeah. and you got a little bit uncomfortable with your character. You got the Ben Macklin behind you over your shoulder. He looks like a little angel on your shoulder. I like the way it's framed right now. It's great. Let me tell you about Console. Console builds AI agents that automate 70% of ITHR and finance support, giving employees instant resolution for access requests and password resets. So this video came up, and it says losing our house, Because of Day Trader. Did you notice this? not because of day trading, but because of day trader. So already odd grammatical error, but play this video. because it's designed to sort of rage bait you and go viral. And it has this saxophone getting louder song. And it's the whole family. This would normally be a tragedy. Like you're losing your family home. But it's super dark and you're like, this is awful. Like, what is going on? But if you dig into this person's if you dig into this person's account, the next video is like I lost my house day trading and like, Here's how I'm planning to win it all back. And then it's like, click my link and buy it. And then you click on the link in bio and it's like, here's my trading mastermind. I have a free course and then I have a $49 course and then I have a $100 course. And so all of this is actually a funnel to an info product. and this is designed to go viral. And I don't know if they lost the house or not. Like, I have no idea. It could be that they're doing better than ever because they're not actually trading at all. Very, very counterintuitive way to market an info product. Being like, I just lost my house because I was bad at something. Now follow me as I run it back up. Maybe. I don't know. It's again, it's like it's. I saw it and it made me very sad. Yeah, totally. Like that is the correct emotional response, but the nature of the algorithm and negativity bias on social media is This is something that you see and you're like, oh, this is horrific. I can't look away. And then a lot of there's just a funnel and some people click and they watch another video and they're like, oh, this is this is inspiring. She's she's working her way back up. and they're staying in the game. They're not, you know, hanging up their jersey, I guess, or whatever you want to call. They're not quitting right before they hit it big. Basically, that's the pitch. Yeah. [25:00] It's really, really dark. So I started going down this because I was like, okay, Who's on the other side of these trains? But yeah, it's interesting because it just like it sticks in your mind because it is such a tragic scenario. Totally. But then what's interesting is they're just like spinning it as, oh, yeah. Well, it's all good. I'm sticking. I'm not. We're not giving up yet. Yeah. Yeah. So it's very, very odd. But there's actually something very interesting going on, because when you see that, you think casino-style gambling, you think that somebody took their capital if they have their nest egg, their kid's college fund. Like when I was a kid, like the don't gamble meme was he gambled away his kid's college fund. Like that was the most degenerate thing you could do as a father. It would be gambling away money that was in your retirement, money that was saved, your nest egg, your rainy day fund was gambled away, usually at a physical casino because you couldn't resist the temptation of blackjack or poker or slots or something. Then we came to the era of the phone came sports betting, offshore betting casinos, crypto casinos. People were losing money on crypto. But again, crypto is a moment where you have to put capital at risk. Like you have to have the nest egg. This is different. This is people that are so there are people that are. getting fleeced and losing money But what's interesting is that it's not that they're taking a nest egg and putting it in a stock that then goes down and they pick the wrong stock. That's not exactly what's happening. Oftentimes what they're doing is they're signing up or subscriptions, TO trade. fake stocks. It's a simulated market that... gives them the feeling of, okay, maybe one day they will be able to get good enough at day trading that they will have access to capital. Well, and these platforms pitch it as do your simulated trading with us. And if you're good, then we will give you, then we will stake you. And we'll give you capital that you can trade. We'll basically turn you into a mini hedge fund. Yes, exactly. And we did a little bit of research. on these companies and their shockingly large. And there's a lot of them. And there's a lot of them. Yeah. We found one that has done hundreds of millions of fees, which for them is like clearly... like better than software margins, right? Because they're just have, they basically have like a fake trading app. Yeah. And they're like, yeah, place your trades. It's tied to real data. Yeah. But then they just promise this, pay the $100 a month, prove that you're good at trading. And if you're good, we'll give you capital, right? Because it'll be a win-win. But of course, over time, a lot of these companies have been exposed as Ponzi schemes. where they take other customers' money. And basically, if you're doing the simulated trading, they're like, you get an attaboy, like you did a good job, and then we're giving you like a reward. And so they'll pay them, which I think is... part of what makes this process addicting. which has driven some of these people to go and get a bunch of subscriptions, which is just like very, very strange because every subscription could be $100 a month. multiplying that out, it's like you could just be investing. But when you're investing, nobody says good job. If you do a little bit better, I'll give you a bunch of capital to invest, right? Because people... LPs are not sitting around being like, how do I back the next stay at home day trader. Yeah, yeah. You know, like... No, 100%. Like, these were so, like... jarring to me because I saw multiple reels from people who were like, I have a prop trading account or I'm working with a prop firm or I blew my account or a blue and account or several accounts. And I was just like you. are not... working with Citadel or Jane Street like this is just this does not compute like you're not on the phone, like your counterparty is not a hedge fund. Like something else is going on. And it turns out that the counterparty is basically a SaaS product that shows virtual trades that are completely simulated. And the crazy thing is that I originally said this. I was like, imagine, like, look at this horrific story. Imagine being at like Citadel or Jane Street watching this and being like, this is the person on the other side of the trades, right? But it's not, it's actually not the hedge fund that's making money off of this. It's the independent. The hedge funds would love to be trading. They would love to be trading against these people. Yeah. They're like, get in the real market. Go on a real, go on a real trade map. Don't be scared. every time you make a trade. We're happy with doing market making. But instead, it's these firms that are actually just taking $100 a month from people, they'll do it on credit cards. And so that's where you get people. Again, we don't know if this one's real or not, but there are actual people who will put multiple hundreds [30:00] a month subscriptions for these simulated trading challenges on their credit cards, and they're not even putting any capital in the market. So there's actually no chance of like, oh, you actually did get into a good stock and it did well. It's purely fake. It's crazy. Crazy. Well, I mean, the market overall is in a rough spot. The Nasdaq is down 1.3%. The Dow is down 0.25%. The headline alphabet sheds half a trillion in market value in the stock's worst day ever. These best day ever, worst day ever headlines are increasingly common. The swings are crazier than they've ever been. There's a lot going on. But we'll cover it here on the show. So thank you for tuning in. Leave us five stars on Apple Podcasts and Spotify. Tyler, you want to take us out with a truth nuke? You didn't get much time on the show today, honestly other than as yourself yeah sort of larping jordy over there but we got a dream we will see you tomorrow at 11 a.m.