Meta, Google Undercut OpenAI as Ad Cash Flow Subsidizes Free AI

广告现金流补贴免费AI，Meta与Google低价施压OpenAI 大型科技巨头将使独立实验室相形见绌，因为自有资金储备和原生分发渠道让Google与Meta能够免费提供前沿AI，而沉重的债务迫使OpenAI和Anthropic提高价格。 Carlson认为，由于庞大的现有现金流、无债务负担的资产负债表以及触达数十亿日活跃用户的无摩擦全球分发能力，Google和Meta最终将在AI竞赛中战胜OpenAI与Anthropic。独立AI实验室面临数千亿美元的债务负担，被迫激进提价并限制功能使用，而Google和Meta则可以通过交叉补贴Gemini 4和Muse等前沿级工具，在价格和用户体验上压制竞争对手。此外，本期内容还认为Netflix近期遭遇的抛售是对联席CEO Ted Sarandos关于参与度的言论被断章取义后做出的不合理反应，并剖析了FICO股价暴跌60%的事件，指出这是其掠夺性垄断提价和疏远合作伙伴所导致的自食恶果。 Google和Meta在结构和现金流方面对OpenAI及Anthropic占据决定性优势，后两者背负着沉重的债务负担，被迫大幅提高订阅价格并减少使用额度。 (9:01) Google推出了Gemini 4（代号“Argon”），其定价约为竞争对手前沿模型的一半，而Meta则利用来自Facebook和Instagram的现金流来拓展其无摩擦的消费级助手Muse。 (7:04) OpenAI将开发者工具和Agent功能塞入单一的ChatGPT界面，面临引起用户疲劳的风险，而Google和Meta则依托数十亿日活跃用户进行无缝的原生分发。 (19:40) 在极端提价引发监管审查并导致征信机构合作伙伴大力推广竞争对手VantageScore后，费埃哲公司（FICO）股价从约2400美元崩跌至600美元。 (26:57)

Joseph Carlson argues that Google and Meta are structurally positioned to surpass early AI leaders OpenAI and Anthropic due to superior balance sheets, native distribution ecosystems, and user-experience advantages. Carlson also analyzes Netflix's misunderstood engagement metrics and examines FICO's stock collapse as an avoidable failure of monopoly stewardship. Balance-Sheet Asymmetry and AI Monetization Pressures (9:01) While public attention centers on benchmark competition among frontier models, Carlson contends the broader competitive outcome will be decided by balance sheets. Startups OpenAI and Anthropic carry contractual commitments and debt obligations nearing hundreds of billions of dollars. Lacking secondary revenue streams, both firms are forced to monetize their models directly by cutting developer plan allowances, raising subscription fees, and restricting free access. Investor Michael Burry has similarly warned that these startups risk faltering under commitments they cannot independently service without financial backing from hardware suppliers. In contrast, Meta and Google fund AI development through established operating cash flows from advertising, cloud services, and search. Mark Zuckerberg noted that Meta faces no existential solvency risk if AI investments underperform, allowing the company to distribute its assistant Muse with uncapped usage. Similarly, Google introduced Gemini 4 at an introductory price roughly half that of Anthropic's competing tier, utilizing AI as an ecosystem waterfall that enhances margins and user retention across Search, YouTube, Workspace, and cloud infrastructure. Ecosystem Integration Versus Bloated Standalone Applications (17:26) Beyond capital resources, Carlson highlights distribution channels as Big Tech's decisive structural advantage. Google embeds Gemini directly across Android, Gmail, Docs, and Drive, eliminating third-party authentication friction. By contrast, the most popular integrations within ChatGPT are Gmail and Google Drive, requiring users to repeatedly complete external logins and two-factor verification steps that introduce friction and privacy hesitations. Additionally, while Meta surfaces Muse across Instagram, WhatsApp, and Facebook, OpenAI must channel all product experiments into a single interface. Carlson notes industry critiques pointing out that adding canvases, mode toggles, and paywalled tools like "dots" has turned ChatGPT into a cognitively complex, bloated super-app. This complexity provides an opening for Meta's streamlined, zero-configuration consumer experience to gain rapid adoption. Decoupling Netflix's Watch Time from Revenue Power (23:05) Carlson argues that the selloff in Netflix shares misinterprets remarks by co-CEO Ted Sarandos, who stated that aggregate engagement growth had slowed to 2% across 200 billion viewing hours. Although financial headlines treated the statement as a sign of business deterioration, Sarandos clarified that Netflix deliberately directs approximately 5% of its content budget to live broadcasts and marquee events that generate only 1% of total viewing hours. While live programming produces fewer aggregate watch hours, it functions as an acquisition driver, reduces subscriber churn, and provides premium inventory for advertising tiers. Carlson emphasizes that Netflix grew trailing revenue by 16%—with at least 10% growth in every geographic region—demonstrating that pricing power and per-hour monetization matter more than aggregate viewing volume. FICO's Self-Inflicted Collapse and Monopoly Mismanagement (26:57) Fair Isaac Corporation (FICO) suffered a severe market reversal, falling from roughly $2,400 to approximately $600 per share despite 25% quarterly revenue growth and 90% operating margins in its core credit scores division. Carlson characterizes this decline as entirely self-inflicted: immediately after surviving a Department of Justice antitrust inquiry, FICO instituted a 400% price hike on its scoring product, capping several years of cumulative increases that reached roughly 1,600%. This aggressive pricing triggered public pushback from lawmakers and led FICO to try bypassing its long-time credit bureau partners. In response, Equifax began aggressively promoting alternative provider VantageScore at half the price. Carlson contrasts FICO's actions with semiconductor equipment manufacturer ASML, which preserves its lithography monopoly by practicing pricing restraint and maintaining collaborative relationships with its customers to minimize regulatory intervention. The program emphasizes that early technological leads can quickly erode when specialized startups lack the diversified cash flows and pre-installed distribution networks of established tech conglomerates. However, whether regulatory scrutiny will eventually target Big Tech's bundling practices—as it did with FICO—remains an open question for Carlson's long-term thesis.

[7:04] ChatGPT and Muse: the most well-rounded agentic technology in the world. [7:04] SemiAnalysis: cooked [7:04] SemiAnalysis: 0% chance of ever having a frontier model again. [9:01] Joseph Carlson: turning the screws on monetization [12:16] Joseph Carlson: waterfall [17:26] Joseph Carlson: super-app [23:05] Ted Sarandos: Yes, overall, we're not growing as fast as I want us to. And we're working on making that move faster. [26:57] Uncle Ben: with great power comes great responsibility [26:57] Joseph Carlson: Yes, we do [have a captive audience]... Our customers can't do anything. Banks can't do anything. We have a captive audience. We are a monopoly. We just wanted to affirm that and make sure it was crystal clear to regulators.

[0:00] If you ask the average investor, what is [0:02] the leading AI company, they'd probably [0:04] say OpenAI with ChatGPT or Anthropic [0:07] with Claude. After all, ChatGPT is the [0:10] go-to AI chatbot. Then you have Claude, [0:12] which is winning in enterprise. Now, [0:14] while both of these companies are the [0:16] leading AI companies today, I believe [0:18] that things will change over the next 5 [0:20] years. And I'll be making the case of [0:21] why I believe the two companies that are [0:23] best positioned to win in AI are Google [0:26] and Meta. Google's Gemini 4 was just [0:28] announced yesterday by Sundar himself. [0:30] And he also released a benchmarks [0:32] showing its relative performance with [0:34] the other flagship AIs. [0:36] We're going to be looking over it, but [0:37] also more importantly, how these [0:39] companies are structurally positioned [0:41] and how they're going down very [0:42] different paths with marketing their [0:44] products. Now, we also have a lot of [0:46] other news to get into. For example, [0:47] Netflix has been on a big sell-off. The [0:49] stock continues to go down. We have an [0:51] interview here with Ted Sarandos, the [0:53] co-CEO of Netflix. He said that Netflix [0:55] is not growing as fast as he would like. [0:57] And that's the headline that Bloomberg [0:58] is printing everywhere. But we're going [1:00] to be looking at Netflix's growth and [1:02] what Ted Sarandos really said. And then [1:04] of course, we have the fail of the week, [1:05] which in this case is FICO, which is the [1:07] worst performing company in the S&P 500 [1:09] this year. It's now down 60% [1:12] year-to-date. This is Dev Kan Tasarias' [1:14] favorite company and it's getting [1:16] destroyed. We'll be going over [1:17] everything that's happening with FICO. [1:19] So, we have a ton to get to in this [1:20] episode. Let's go ahead and jump in. [1:22] Now, if you haven't already, join [1:23] qualtrim.com. This is my own personal [1:25] membership. It's only $10 a month. You [1:28] can try it out risk-free with a 7-day [1:30] free trial and it gives you access to a [1:32] load of benefits. You get access to [1:33] Qualtrim, the entire website. You can do [1:35] analysis on any stock. We have earnings [1:37] transcripts, discounted cash flows, all [1:40] the stuff that you would expect. Plus, [1:41] we're launching a mobile app. We have it [1:43] in the iOS store in the United States. [1:45] It already is getting extremely good [1:46] feedback. People like the app more than [1:48] they like the website and we're only [1:50] making more improvements. Plus, you get [1:52] exclusive content. I just did an [1:53] hour-long exclusive on FICO and you get [1:56] access to a Discord community full of [1:58] thousands of investors. Again, you can [2:00] try this out with $10. It's the price of [2:02] a sandwich. So, try it out now. You have [2:04] nothing to lose. It's risk-free, and I [2:05] don't think you'll regret it. Now, it [2:07] seems like every day in these AI wars, [2:09] there's companies leapfrogging one [2:11] another. For example, we had the early [2:13] leader, which was ChatGPT. ChatGPT went [2:16] super viral. It was the consumer AI app. [2:19] It was the one that was supposed to [2:20] destroy Google Search. Now, ChatGPT is [2:23] extremely successful. It's probably one [2:25] of the most successful products ever. [2:27] But then you had Anthropic. Anthropic [2:30] took a different route. Instead of [2:31] focusing on the consumer use cases, like [2:34] picture editing and video, Anthropic [2:36] went straight for code. And it turns out [2:39] there's a a lot of applications for [2:41] code. Everything that businesses do in [2:43] the modern world includes code. So, [2:45] Anthropic went super viral amongst the [2:47] workforce. Amongst institutions, [2:49] automating different parts of companies. [2:51] Anthropic came out with skills and [2:53] integrations into different businesses. [2:56] And then ChatGPT turned their attention [2:58] to Claude, also going after the [3:01] enterprise applications. [3:02] You had Claude code to begin with, and [3:04] then you had ChatGPT's Codex, their own [3:07] tool for enterprise users and for [3:09] coders. And while this battle was going [3:10] on between the two unopposed leaders in [3:13] AI, it seems as though there was a [3:15] duopoly forming. It seemed like we're [3:17] going to have a Coca-Cola and Pepsi [3:19] situation, or a Moody's and S&P Global, [3:22] or a Mastercard and Visa. It looked like [3:24] for a while that it was going to be [3:26] Anthropic and OpenAI, the two undisputed [3:28] and unassailable leaders in AI. [3:31] But then things started to change. [3:33] During this time period, Mark Zuckerberg [3:35] pivoted very hard. He threw tens of [3:38] billions, if not hundreds of billions of [3:40] dollars, into creating data centers, [3:42] into buying GPUs, into training his own [3:44] models, into using his own employees, [3:47] capturing their keystrokes to learn how [3:48] they worked. He went so far as to change [3:52] the entire direction of his company and [3:54] put all the financial resources of [3:56] Facebook and Instagram into becoming an [3:59] AI leader. [4:00] And then it emerged with one product, [4:02] which is Muse. Muse did a couple things [4:05] right. Muse was not as confusing as [4:08] ChatGPT. ChatGPT, when you enter into [4:11] it, it has different toggles that you [4:13] have to switch on for what model you [4:14] want. For most people that aren't AI AI [4:17] people, they don't know what model is [4:19] even the one that they should be using. [4:20] So, this is confusion for the normal [4:22] person using Open AI's tools, but for [4:25] Muse, you don't have one. You just type [4:26] in and it gives you a great answer. Muse [4:28] also has this nice little whimsical [4:30] mascot, this friend that talks to you [4:32] and creates a little bit of a [4:33] relationship as you're using the [4:34] product. He's cute and he can change and [4:37] be customized in any way that you want. [4:39] These were striking adaptations that [4:41] Muse did over ChatGPT or Anthropic. [4:44] ChatGPT and Anthropic at the time have [4:46] been so lost in enterprise world, [4:49] looking for coding applications and [4:50] different enterprise workflows, that [4:52] they forgot about the consumer. And Meta [4:54] was so focused on the consumer that they [4:56] made a streamlined product that's very [4:58] appealing for the average user. Muse's [5:00] growth has been rapid. In fact, reports [5:02] show that Muse's growth today is at a [5:05] pace faster than when ChatGPT first [5:07] launched. But then we also have Google [5:09] with Gemini. Google is the leader in [5:12] artificial intelligence. They had [5:13] invested in this for a long period of [5:14] time, and Google, by every metric, [5:17] deserves to be the leader. They are the [5:19] ones that have invested the most money. [5:21] They're the ones with the full stack [5:22] team. They have the silicon. They have [5:25] the GPUs. They have all the server [5:26] capacity. They even have the [5:28] intelligence in DeepMind. They had [5:30] everything in order, but like in many [5:32] cases, as a bigger, larger, bloated [5:35] institution, not as streamlined as a [5:38] younger startup, they're not as nimble. [5:40] Google got caught up in bureaucratic [5:42] bloat. They couldn't release products [5:44] because their products are too big. They [5:46] have to scale too far. There's too many [5:48] safeguards and bureaucratic red tape to [5:50] step through to be able to push their [5:52] product out. They initially released [5:53] Bard and then they upgraded that to [5:55] Gemini and they've been chasing the tail [5:57] of ChatGPT ever since. But Google's done [6:00] a couple things that have surprised [6:01] different investors. First of all, [6:03] Google has shown that ChatGPT really [6:05] isn't destroying the Google search [6:06] business. They've been able to protect [6:07] that somehow. Second of all, Gemini has [6:11] grown rapidly and this shows the power [6:13] of Google's distribution. Gemini isn't [6:16] just a app on the phone. It is installed [6:19] on every Android. It's used in every [6:21] part of Gmail and YouTube. It's [6:23] integrated into Google Drive. It's [6:25] integrated into every service like Maps [6:28] that people use billions of times per [6:30] day. [6:31] So Google has been able to catch up at a [6:34] rate that normally they should not. So [6:36] Google is in a situation where they have [6:38] all the right ingredients, but they have [6:39] been chasing ChatGPT. They've always [6:42] been a little bit behind. Now in my [6:45] previous episode, I said that Google's [6:47] behind, but I'm not going to sell the [6:49] stock. I own a lot of Google. It's my [6:50] largest position. I have a $200,000 [6:53] position in it and I'm not selling a [6:55] share specifically because even though [6:57] Google's behind, I believe that things [7:00] can change quickly. And it looked like [7:02] they changed quickly once again [7:04] yesterday. Yesterday Sundar posted on X [7:07] that he is announcing Gemini 4. This is [7:09] called Argon and it's Google's next [7:11] flagship model. Now they also gave their [7:14] benchmarks for Gemini 4. And when you do [7:16] a more full analysis on where they [7:17] actually lead and they don't lead, this [7:19] shows that it's incredibly well-rounded. [7:22] In fact, I plugged these benchmarks into [7:24] ChatGPT and into Muse to get their [7:26] unbiased opinions as competitors and [7:28] both of them noted that if these [7:30] benchmarks are real, that Gemini 4 would [7:32] be the most well-rounded agentic [7:34] technology in the world. It's not the [7:35] best at everything, but it's the best at [7:37] many things and very good at everything. [7:40] They say that it has frontier [7:41] performance in complex workflows across [7:43] real-world software engineering, [7:45] enterprise knowledge work like legal and [7:47] finance and cybersecurity defense. Now, [7:50] you may scoff at these benchmarks, but [7:51] this is important. A lot of companies [7:54] and users and organizations are deciding [7:56] which model to use based on its [7:58] capabilities. Which one does the task at [8:00] hand competently and in a reasonable [8:02] amount of time? And which one is priced [8:04] the best? And many people doubted Google [8:07] once again. Even SemiAnalysis wrote an [8:10] article. This was August 6th saying that [8:12] Gemini is cooked. Part of that article, [8:14] they say Google has 0% chance of ever [8:17] having a frontier model again. And here [8:20] we have with Gemini 4, according to the [8:21] benchmarks, it is quite possibly the [8:24] best or one of the best models, [8:25] definitely competitive with the other [8:27] frontier models. Google is officially no [8:29] longer behind. Google has caught up. [8:31] They are right there with the other [8:32] frontier models. So, that's important in [8:35] and of itself. But, there's something [8:36] else going on that makes me more bullish [8:38] on these two companies in particular, [8:40] Google and Meta. Now, I'm fully invested [8:43] in these companies. I have a combined [8:44] $400,000 [8:46] between Google and Meta. So, this shows [8:48] it's important to me in getting this [8:50] right. But, I believe that Google and [8:52] Meta both are set up to be the two [8:54] biggest winners in AI even more than [8:56] Anthropic and OpenAI. Now, again, [8:59] there's big arguments against this [9:00] because everybody will say that [9:01] everyone's using Anthropic at their job [9:03] or at their work. Everybody will say [9:05] that they have ChatGPT installed on [9:07] their phone. And that's true. But, [9:09] there's something different going on. [9:10] One of them is the financial realities [9:13] for these companies. Even though these [9:15] companies are competing with different [9:16] frontier models and they're kind of [9:17] going back and forth. One day Muse feels [9:19] like it's the best. The next it's [9:21] ChatGPT or Anthropic. The next it's [9:23] Gemini coming up with Gemini 4. They're [9:25] all just leapfrogging each other with [9:26] this technology. And while they battle [9:29] that out, that's only part of the game. [9:31] The other part is the financial reality. [9:34] You see, OpenAI and Anthropic are not [9:36] publicly traded, so we can't see their [9:37] financials quite yet. But, what we do [9:40] know is that they have a lot of debt. [9:42] And a lot of debt is probably the [9:44] understatement of a century. They have [9:46] enormous mountains of debt. Debt that [9:49] we've never seen in different companies. [9:51] Obligations so big that it's almost [9:53] incomprehensible. [9:55] These are the amounts of debt that [9:57] people like Michael Burry have been [9:58] writing about. It's just too much debt. [10:00] In his mind, he thinks these companies [10:01] will collapse under the weight of their [10:03] enormous obligations. He doesn't think [10:05] they can pay anybody back. He also talks [10:07] about how Nvidia is funding these [10:09] companies trying to prop them up because [10:11] they can't financially fund themselves. [10:14] And we've seen that with these [10:15] companies. We've seen Anthropic recently [10:18] turn net positive, so they're at least [10:20] making a profit. And we see OpenAI [10:22] driving towards that point as well. Both [10:24] of these companies are under financial [10:26] pressures that don't exist with Google [10:29] or Meta. [10:30] Not at all. And the pressure must be [10:32] immense. For example, the only way that [10:35] Anthropic has to pay back on all their [10:37] obligations, all the agreements they [10:39] signed, all these deals, hundreds of [10:40] billions worth of debt, nearly half a [10:42] trillion dollars in obligations, [10:45] they can only pay that back through AI. [10:48] That's how they monetize. The one tool [10:50] they have to pay back all these lenders [10:52] is with the product itself. Anthropic, [10:55] Claude, Opus, their models. They have to [10:58] monetize them enough and make people pay [11:00] enough so that they can pay back on [11:02] their lenders. [11:03] They have to do it. They have no other [11:05] revenue stream. And you have OpenAI with [11:07] ChatGPT. [11:08] ChatGPT is in a very similar situation. [11:11] They've said that they're healthy [11:12] business. They said that they're making [11:13] money and they they can become really [11:15] profitable. But, what OpenAI has been [11:16] doing has been signaling that they're [11:18] really turning the screws on [11:20] monetization. They know they have [11:22] enormous obligations. They're feeling [11:24] the weight of that. They have to [11:26] generate positive income and a lot of it [11:28] to be able to pay back. The only way [11:30] that OpenAI, like Anthropic, can pay [11:32] back on all these obligations, this [11:34] mountain of debt, is by charging for [11:37] their AI, by directly monetizing it, by [11:40] making it ever increasingly expensive [11:42] for the consumer. Now, they'll do this [11:44] either by loading ads and making so you [11:47] have sponsored listings within your chat [11:48] application, or they'll just directly [11:51] charge you, like they did with Codex. [11:53] Basically, doubling the price of the [11:55] $200 plan. Open AI made it so that the [11:58] $200 plan now gives you half the [12:00] credits. And they introduced a new $500 [12:02] plan that only gives you 20% more [12:04] credits than the 200. This is another [12:06] way of having a massive price hike. Open [12:09] AI is raising prices, Anthropic is [12:12] raising prices, specifically to make it [12:14] so that they can pay back on their debt. [12:16] Now, compare that to the situation of [12:18] Google and Meta. [12:19] Google and Meta's are free. [12:22] In fact, when you look at Muse, it gives [12:24] you a crazy amount of usage for free. [12:27] I have my sister, who's not a tech [12:29] person, barely knows how to use a [12:30] computer, but she's used ChatGPT for all [12:33] the basic stuff. For searches and [12:35] creating birthday invitations and lots [12:37] of fun stuff like that. She continually [12:39] ran in to limits on the free ChatGPT. On [12:43] Muse, [snorts] she has not. Not even [12:45] close. In fact, she says that it's it's [12:47] so great, she can use it all day long [12:49] for every application. She's told [12:51] friends about it because it's so much [12:52] usage without the caps that ChatGPT has. [12:55] And that's the difference. With Meta, [12:57] they don't need to monetize with Muse. [12:59] Meta's not in a situation where they [13:01] have a mountain of debt to pay back on. [13:03] Meta's not even funding this whole thing [13:05] by debt. They've taken out very little [13:07] debt, and almost all of their ambitions [13:10] in AI are being funded by their cash [13:12] flows from Facebook and Instagram. See, [13:14] Meta has the advantage of having these [13:16] other businesses that gush cash every [13:18] single day. Mountains of cash being used [13:20] to fund this massive AI opportunity. And [13:23] on a recent podcast, just a couple days [13:24] ago, Mark Zuckerberg talked about just [13:26] that. He said that there are going to be [13:28] companies that get really hurt in AI. [13:31] But he says with Meta there's basically [13:32] no risk. If AI doesn't work out, they're [13:35] not in debt, the company's not in [13:36] trouble, they're not going to go like [13:37] bankrupt or anything. There's really no [13:40] stress in the situation. And meanwhile, [13:42] they can give users enormous amount of [13:44] free usage. They can make really [13:46] streamline product to gain massive user [13:48] adoption, and Meta is very good at [13:50] monetizing free products. That's kind of [13:52] their thing with Facebook and Instagram. [13:54] They've shown that they have the ability [13:56] to make a lot of money with free [13:57] products. But at the same time, they can [14:00] afford to lose money on AI for a very [14:03] long period of time, much longer than [14:05] OpenAI or Anthropic. See, Meta has the [14:07] advantage of an enormous piggy bank [14:09] that's ever increasing, always flowing. [14:12] Instagram and Facebook can fund their AI [14:14] efforts until they make ChatGPT run out [14:17] of money. Until they make them not [14:18] competitive with the value proposition [14:20] that they offer. And consumers are [14:22] already starting to notice the [14:23] difference. [14:24] One of these companies offers a product [14:26] that is virtually uncapped for the [14:28] average consumer. The other one has very [14:30] quick caps for the free ChatGPT. Muse in [14:33] and of itself is an agent. That's what [14:35] it does. It opens up browsers, and it [14:37] does work behind the scenes that helps [14:39] you accomplish things you want to do. [14:41] OpenAI's response to Muse was to create [14:44] dots. Their version of an agent. Now, I [14:46] think the mascot in and of itself is not [14:48] quite as compelling as Muse. That's just [14:50] an opinion thing. But one thing that's [14:51] not an opinion is the way that they've [14:53] monetized it. You see, Muse is free for [14:56] everyone. Everybody has access to this [14:58] agentic technology today for free. [15:00] ChatGPT has gated dots behind the plus [15:03] plans. They're not offering this for [15:05] free. Only paid users, which is a tiny [15:08] minority of users, have access to dots. [15:11] So, most of them are not going to see [15:13] this new agentic technology. And why [15:15] doesn't OpenAI include dots for free in [15:18] ChatGPT like Muse is for their users? [15:21] Well, of course, because they need money [15:22] and they need it right now. If they want [15:25] to IPO, they need to show that they're [15:26] growing in profits. [15:28] So, you have this distinction between [15:29] these companies that already have this [15:31] massive cash flow that can indefinitely [15:33] fund their new efforts competing with [15:35] companies that are strapped for cash. [15:37] They have to continually charge for [15:38] products and limit usage and increase [15:40] prices. [15:41] And the value proposition is getting [15:42] wider and wider between the different [15:44] product sets. When Sundar announced [15:46] Gemini 4, he said the introductory price [15:48] for it is going to essentially be half [15:50] the price of the equivalent from [15:51] Anthropic. A 50% discount on an [15:54] introductory price. Now, we don't know [15:55] when that introductory price discount's [15:58] going to end, but that's going to [15:59] attract a lot of developers in the [16:01] meantime. If you're getting essentially [16:03] the same performance for half the cost, [16:05] uh that can save a lot of money for a [16:06] lot of people. And how can Google afford [16:09] this? Well, see with Google, they don't [16:11] need to monetize Gemini directly. When [16:13] investors look at Google, they notice [16:14] that the company has many assets that [16:16] benefit indirectly from their AI [16:18] leadership. Google looks at Gemini as [16:20] just one more part of Google. Google has [16:23] cloud hosting. They have an entire chip [16:25] business. They have their AI and [16:28] DeepMind and then they have YouTube and [16:29] Waymo and search and enterprise [16:31] applications, Gmail, on and on and on. [16:35] Google has like nine different apps that [16:37] are used by a billion plus users. It is [16:39] a sprawling business that is incredibly [16:42] profitable and they have no pressure to [16:44] generate enormous profits in the short [16:46] term from Gemini. Google knows that [16:48] Gemini is not the end product. It's not [16:51] what they need to monetize through. If [16:52] Gemini helps YouTube grow by better [16:55] recommendations of video, if Gemini [16:57] helps Gmail maintain its market share [16:59] and better advertise, if Gemini helps [17:01] other aspects of the business like [17:02] search, which it's integrated into [17:04] search, then Gemini will be like a [17:05] waterfall monetizing all parts of the [17:07] business. The exact same thing that Muse [17:09] is doing for Meta. So, these two [17:11] companies, Google and Meta, are in no [17:12] situation to rush to monetize their [17:14] applications. Instead, they can go for [17:16] massive land grab market share. They can [17:19] undercut Anthropic and OpenAI every [17:22] turn. And pricing is just one aspect of [17:25] why I think they'll be winners. Now, [17:26] aside from the differences in financial [17:28] realities that these companies face, [17:30] there's also a different reason that I [17:31] believe that Google and Meta have a very [17:33] good chance of becoming the biggest AI [17:35] companies and biggest winners. And part [17:37] of that is simply friction and trust. [17:40] Now, you may say trust is not an [17:42] advantage of Meta. In fact, a lot of [17:44] people are distrusting of the company. [17:45] I'll address that in a minute. But, I [17:47] want to go first to friction. [17:49] You see, I have a paid ChatGPT account. [17:52] I think it's great. But, when I look at [17:53] integrating it and using it agentically, [17:56] you have to use connectors or plugins. [17:58] They have all different names, but it's [17:59] basically connecting things to your [18:01] ChatGPT account. In this case, when I go [18:04] to the most popular plugins across all [18:06] of ChatGPT, number one is Gmail, number [18:10] two is Google Drive. [18:12] Then going down the list, you have all [18:13] different plugins. Some from a different [18:15] companies, but many of them from Google. [18:18] But, overall, the most popular plugin in [18:20] the world for ChatGPT, the most used [18:23] agentic use cases are Gmail and Google [18:26] Drive. That's what people are using [18:28] ChatGPT for. [18:30] Now, you may say that many people trust [18:32] ChatGPT, so they'll plug in with Gmail [18:34] and Google. [18:35] But, there is no trust deficit that [18:38] needs to be bridged when you're using [18:40] Gemini. You're literally using a Google [18:42] product. Google will still likely have a [18:44] prompt where you can enable it, but [18:46] it'll be much simpler. You won't have to [18:48] enter in credentials. You won't have to [18:50] go through the two-factor authentication [18:51] to say, "Hey, does Gmail really have [18:53] permission to access all of this?" This [18:55] will make it so that if Gemini 4 really [18:57] is frontier, which it appears to be, [18:59] that you'll have an easier time [19:01] integrating with all the apps that you [19:02] would normally integrate with, which is [19:04] Google Drive and Gmail as a top two. So, [19:07] the trust factor is already solved by [19:09] Google. But, then you have Meta, which a [19:11] lot of polls say people don't trust [19:13] Meta. They they have a a big distaste [19:15] for it. A lot of this is media framing. [19:17] 3.7 billion people that don't trust Meta [19:20] log in or use one of their applications [19:22] every single day. They use their chat [19:24] services, they post pictures on Meta, [19:26] they they post on Instagram, they read [19:28] it, they give information every single [19:30] day. And these actions speak much louder [19:32] than a few surveys from people that [19:34] don't like Meta on the surface. When I [19:36] look at it, Meta, I believe, will gain [19:38] the trust of consumers. But the other [19:40] advantage that has, that Anthropic and [19:43] ChatGPT are trying to gain, is [19:45] distribution. Now, ChatGPT has a lot of [19:47] distribution, but this is also a [19:49] problem. See, with Meta, they have it [19:51] everywhere. Meta has WhatsApp, which [19:53] Muse is like appearing everywhere in. [19:55] Then Meta has Instagram, which Muse is [19:57] at the top of your feed. They're [19:58] integrating it into Instagram. Then you [20:01] have Facebook, which a lot of people [20:02] still use. I know young people aren't [20:04] using it as much, but a lot of people [20:05] still use Facebook, and Meta has [20:07] integrated Muse into it. Has put [20:09] advertisements of Muse intertwined with [20:11] it. Meta has the biggest distribution by [20:14] users every day of maybe any company in [20:15] the world. OpenAI has a lot of [20:17] distribution, but it's confined into one [20:20] app. Just one. ChatGPT. [20:23] So, OpenAI has a problem here. [20:26] To make an agentic assistant that [20:28] competes with Muse, they'd have to make [20:30] another app. But then they'd have to [20:31] start from zero downloads. And that [20:33] would be difficult because they don't [20:35] have an Instagram to push it or a [20:36] Facebook to push it or a [20:38] WhatsApp. They don't have all of these [20:40] different assets to get people to try it [20:42] out. They only have ChatGPT. So, what [20:44] does OpenAI do? They tuck their new [20:47] agentic assistant dots [20:49] as an item in the menu. And there's [20:51] toggles and switches and menu items and [20:53] libraries and projects and canvases. [20:55] There's all sorts of things that are [20:56] difficult to understand. And then there [20:58] is dots now tucked in the corner if [21:00] you're a paying user. What do dots even [21:03] mean? [21:04] A lot of this is becoming more [21:05] confusing. Is dots like the normal [21:07] ChatGPT? Is it different? How are they [21:09] different? You see, OpenAI putting [21:12] everything into one app, which is [21:13] ChatGPT, is making things way more [21:16] confusing. And this is being noticed by [21:18] even a lot of AI people. We have single [21:20] here saying, "The Muse hype is real. I [21:23] can definitely say that it is the [21:24] easiest to use and most focused consumer [21:26] AI product on the market. The extremely [21:29] generous limits also make it impossible [21:31] to ignore. It actually helps the average [21:34] person navigate day-to-day of personal [21:36] life." He says, "Here's what he would [21:38] have done with ChatGPT instead. I [21:40] probably would not have created a super [21:42] app. Instead, I would have created two [21:44] distinct product services. Yes, I [21:46] understand the downsides, especially [21:48] heading towards an IPO. But almost every [21:50] great tech company eventually becomes a [21:52] collection of products, either through [21:54] acquisition or by building them. Google [21:57] has like eight products with 2 billion [21:59] plus users. Facebook has lots too. The [22:02] main problem is that trying to turn [22:04] ChatGPT into a super app has created an [22:06] enormous amount of complexity, not just [22:08] technically, but cognitively. There are [22:10] so many modes and tools and agents and [22:13] surfaces, etc., that even describing the [22:15] product to a normal person is getting [22:17] difficult. ChatGPT is becoming insanely [22:20] complex because they know that they [22:21] don't want to start an entirely new app. [22:23] They're going into IPO. They know that [22:25] they don't have the distribution to push [22:27] an entirely new app. So instead of doing [22:29] that, they have shoved it into the [22:31] existing ChatGPT, making an ever [22:33] complex, more sophisticated, more [22:35] cognitively difficult app to understand. [22:37] While you have Muse, which is [22:38] streamlined, easy to use with a huge [22:40] amount of free limits. And this is all [22:42] creating a massive opening for Meta and [22:44] Google. Now, there's no way of seeing [22:45] the future, but we have every reason to [22:47] believe that Google and Meta are in very [22:49] advantage positions and have a huge [22:51] opportunity to win in the AI race. [22:54] Now, moving on, I want to go to another [22:55] company that's gone through a bit of a [22:57] sell-off here. It's one that's shown up [22:59] on my DCF as being undervalued once [23:01] again. In fact, it is the most [23:03] undervalued company in my portfolio [23:06] based on my own internal analysis. That [23:08] is Netflix. [23:09] We have a recent clip here. This is from [23:11] Ted Sarandos, the CEO of Netflix, [23:13] addressing some of the concerns about [23:15] the company. He was asked first about [23:17] engagement. [23:17] >> Uh and engagement is a is a it just is [23:20] another metric. But at the time we said [23:22] that, I think we were pretty [23:24] unsophisticated about how we talked [23:25] about engagement. Meaning in that world, [23:28] all thing all revenue is equal, all [23:30] profit is equal, all engagement is [23:32] equal. And all engagement is not equal. [23:34] So, we are first and foremost, we are [23:36] growing engagement. Uh so, we're growing [23:39] on 200 billion hours of watching. We [23:41] grew 2% in uh you know, in our last [23:43] announcement. [23:44] >> 2% is not what people are hoping for out [23:46] of you, right? They're used to [23:48] double-digit growth. In in at least in [23:50] certainly in revenue and to some extent [23:52] even in viewership, they'd expect more. [23:54] >> This is my This is my point about the [23:55] the growth in general. Yes, overall, [23:58] we're not growing as fast as I want us [23:59] to. And we're working on on making that [24:01] move faster. [24:02] >> That last sentence you just said, that [24:04] we're not growing as fast as we want to, [24:06] is the headline of many articles being [24:08] published today. That That's the [24:09] headline article. But, he also goes on [24:11] to explain the specific factors behind [24:14] this growth and how much of this is [24:15] actually misleading. [24:17] >> We are though also doing things that [24:19] create a lot of headwind to that number. [24:20] Meaning, when we do live programming on [24:23] Netflix, which is a relatively new [24:24] thing, [24:25] um we spend about 5% of our content [24:28] budget on live [24:30] events. [24:31] They generate about 1% of our watching. [24:33] >> So, Netflix has chosen to hurt their [24:35] watch time deliberately, which most [24:37] companies wouldn't want to do. [24:38] >> Now, they all but they do a very [24:40] different job than I [24:42] >> a lot of sign-ups. They're really [24:43] effective for advertising. [24:44] >> Sign-up retention, advertising, all [24:46] those things that they do, but it [24:47] creates engagement headwind uh in in how [24:50] you invest against it. [24:51] >> Investors are worried about something [24:52] that's actually a strength. Engagement [24:55] should not grow as fast as revenue [24:57] growth. Because if revenue growth was [24:59] limited to engagement growth, that means [25:01] that Netflix's total revenue is limited [25:02] by hours watched. Meaning that they'd [25:05] have massive pricing limitations. People [25:08] would only pay proportionate to the [25:09] amount of time they watch. But that's [25:11] not how people pay for content. World [25:13] Cup tickets sold for thousands of [25:15] dollars just for 1 hour or an hour and a [25:17] half of pure entertainment. That's [25:19] because it was a big meaningful event. [25:20] And that's the arena of content. People [25:22] will pay a lot of money even for a short [25:24] amount of watch time if that watch time [25:26] is really good. Every area, all of them [25:28] are growing at least 10%. The company [25:30] overall on a trailing basis grew by 16%. [25:33] It's not a bad sign when your company's [25:35] creating engagement that people enjoy. [25:37] They're willing to pay more for it. [25:39] They're growing throughout the world in [25:40] regions that they're not currently big [25:42] in. And in every territory, they're [25:44] growing above 10%. That's not the story [25:46] of a company that's struggling. Netflix [25:48] is actually doing quite well. Now, [25:50] finally, we move on to the fail of the [25:51] week. In this case, I have to highlight [25:53] FICO. This is a stock that I've been a [25:56] fan of one part of it for a long period [25:57] of time and a bit concerned about [25:59] another part. And it turns out that [26:01] concern that's caused me to stay out of [26:03] this company despite its incredible [26:05] fundamentals, [26:06] well, that concern has played out in a [26:07] really bad way for FICO shareholders. [26:10] And it's really unfortunate. This is one [26:12] of the fail of the weeks where I I just [26:14] feel bad for the whole situation. It [26:16] feels very unnecessary and it feels like [26:19] a lot of people are getting hurt with [26:20] FICO that don't need to be. It's a [26:21] company that Dev Kantesaria has made a [26:23] signature part of his portfolio. At one [26:25] point, he had over 30% of his $4 billion [26:28] equity portfolio just in FICO. It had [26:31] raised up to $2,000. It looks like [26:33] things were going good for Valley Valley [26:35] Forge Capital until we saw this sequence [26:38] of events cause this company to trade [26:40] down. Now, I went through systematically [26:42] every single thing that happened with [26:44] FICO in a timeline in an exclusive [26:46] episode. It is the downfall of FICO. [26:49] What really happened? [26:51] But if we look at this company, it seems [26:53] as though it has no business to sell off [26:54] this much. In the past 5 years, FICO has [26:57] now underperformed the S&P 500. But it's [27:00] also gone from $2,400 per share to [27:02] around $600. It's a company that's [27:04] growing quickly. The revenue grew 25% in [27:07] the most recent quarter. The operating [27:09] margins are extremely high. In fact, for [27:12] the scores segment, they're at 90%. [27:14] Even their software business, which is [27:16] like the bad business for FICO, has [27:18] operating margins in line of top-tier [27:21] companies. It It's operating margins on [27:23] its software are similar to Netflix's. [27:25] This is a good business, and then the [27:27] scores business is phenomenal. There's [27:29] nothing else like it. Overall, FICO has [27:32] one of the best businesses in the world, [27:34] bar none. It was an undisputed monopoly. [27:37] It was seemingly unassailable. They had [27:39] the whole world in their hands. [27:41] But like Uncle Ben says in Spider-Man, [27:44] with great power comes great [27:45] responsibility. And that is what FICO [27:47] didn't seem to understand. They had [27:50] everything. They They really did have [27:51] everything. They had the whole world. An [27:54] extremely profitable business, almost no [27:56] real competition, regulators that didn't [27:59] seem too concerned about them. They were [28:00] poking around, but but there wasn't any [28:03] alarm. And then FICO made deliberate [28:05] decisions to torpedo this entire [28:07] situation. To just sink everything. You [28:10] can blame regulators. You can blame [28:12] different parties. But much of this was [28:14] self-inflicted. FICO had literally just [28:16] gotten through an open and closed [28:18] Department of Justice case on whether [28:20] they're a monopoly, an antitrust [28:22] investigation. This was the government [28:24] saying, "Hey, we think you might be a [28:26] monopoly. We believe you might have a [28:28] captive audience." And then FICO just [28:30] raised their hand, and they said, "Yes, [28:31] we do. We're going to raise prices by [28:33] 400%. Our customers can't do anything. [28:36] Banks can't do anything. We have a [28:38] captive audience. We are a monopoly. [28:40] We just wanted to affirm that and make [28:41] sure it was crystal clear to [28:43] regulators." [28:45] This was everything regulators needed. [28:47] Once the price hikes came into effect, [28:49] regulators amped up the pressure [28:51] enormously on FICO. And then you had [28:53] Bill Pulte come into the scene. Bill [28:55] Pulte used the price hikes by FICO, as [28:58] well as different senators like Josh [29:00] Hawley, to emphasize pressure on [29:02] regulating this company. When you'd go [29:04] looking for a home, the credit bureaus [29:05] would run your score as you get your [29:07] credit pulled. And they'd also give you [29:09] the FICO score and make a little money [29:11] off the top. So, whenever a FICO score [29:13] was ran, both FICO and the credit [29:15] bureaus were happy. They'd both make [29:17] money. FICO [snorts] decided to [29:18] circumvent the credit bureaus. Going [29:20] around them, cutting them out of the [29:21] equation. Their distributors and [29:23] partners for years, which they had this [29:25] cohesive relationship, now was under [29:28] attack by FICO. [29:30] And then you had Equifax respond [29:32] aggressively, very counter to what FICO [29:34] is doing, making Vantage Score half the [29:36] price of FICO. FICO somehow, not only [29:39] made regulators upset by raising prices, [29:42] but then they turned to their own [29:43] partners and distributors and made them [29:44] upset by circumventing them. They're [29:46] also not just their partners and [29:47] distributors, they're the people that [29:49] own the Vantage Score competitor. And [29:52] they're the people that own the data of [29:53] which they pull their scores from. This [29:55] was FICO management choosing to destroy [29:58] virtually every relationship around [29:59] them. Now, of course, none of this was [30:01] necessary. FICO could have been a better [30:04] steward over their monopoly. They could [30:06] have raised prices 10 to 20% a year. [30:08] Enough that they make good money for [30:10] their shareholders, but also they don't [30:12] raise the eyebrows of regulators to such [30:14] a huge degree. By raising prices nearly [30:17] 1,600% over just a couple year period, [30:20] it made everybody on board with fighting [30:22] against FICO, whether justified or not. [30:24] It was now a PR campaign to destroy this [30:26] company, to introduce real competition. [30:29] And that's what we see being priced into [30:30] the stock today. [30:32] So, as we see FICO today, [30:34] even though some of this was out of [30:36] their hands, much of this was [30:37] self-inflicted by management. And it's [30:39] unfortunate and unnecessary. Companies [30:41] that have monopolies need to act [30:43] responsibly, with restraint, with [30:46] partnership leadership. They need to [30:48] work cohesively with their customers. [30:50] ASML is an example of this. [30:52] ASML has likely a better monopoly than [30:54] FICO, one that's even more undisputed, [30:57] and ASML could theoretically raise [30:58] prices as much as they want. But they [31:00] won't, because they're operating with a [31:02] mature, long-term vision. They work as [31:04] partners with their customers. They want [31:07] both of them to be happy with the [31:08] situation. They don't want to attract [31:10] undue regulation. This is the type of [31:12] forward-looking management that a [31:13] monopoly should have. [31:15] In the case of FICO, unfortunately, it [31:17] didn't have it. That's all for this [31:18] episode. Hope you enjoyed. See you in [31:20] the next one.