In short: Referenced only — Talkington (16:14), alongside Anthropic, as seeking government liability cover she doubts they get.
In short: Referenced only, as the competitor. Muse dethroned ChatGPT on the App Store. He predicts OpenAI will launch its own agentic assistant during Meta Connect "to try to steal thunder," and says Meta beating it to market "must be more than frustrating for Sam Altman," since OpenAI "own[s] the consumer AI app."
9:20Now, during Meta Connect, I believe there is a chance, and I think that this is a prediction I'll be willing to make. I think that CHACBT is going to release something to try to distract people from MetaConnect. Maybe their own agentic assistant. ChatBT could release theirs the same day as Meta Connect to try to steal Thunder from it.
In short: Eisman: the whole ecosystem "is dependent upon two companies who lose — to say that they lose billions is a euphemism — and one of them I think is in trouble, which is open AI." / Noble: "Totally agree." Nvidia's $250B funding package for it is Noble's proof the ROI isn't there; the circular financing "tells me that there's probably something wrong" (Eisman) — it only works "if by some miracle anthropic and open AI became insanely profitable."
OpenAI, maker of ChatGPT, loses billions of dollars a year and depends on raising new money. Eisman thinks it is in trouble, and Noble agrees. Its main chip supplier had to offer it a huge funding package, and money flowing in circles between suppliers and customers only works if the AI labs eventually become very profitable. Because the cloud companies and NVIDIA depend so heavily on OpenAI and Anthropic, trouble at OpenAI would spread through the whole AI industry.
17:11So basically if you follow the chain down the entire ecosystem is dependent upon two companies who lose to say that they lose billions is a euphemism and one of them I think is in trouble which is open AI. — Totally agree — and so anthropic is going to go public. They I think they need to get their S1 out fast because token maxing is ending.
In short: RPK: the labs' AI-safety campaign is "a PR onslaught": a salvation narrative ("regulate us… create a monopoly of just a few of us"), cover for build-out limits they can't grow past, and a response to growing liabilities (that morning's story of Anthropic staff hacking OpenAI's emails). Sam: "regulatory capture," coordinated with METR, the NYT and politicians within three days. Referenced as industry context, not a position.
28:28So I'm not going to go with scam, but I will say PR. I'm going to go with that. It's kind of a PR onslaught. And there are a couple things it's sort of solving for. One is sort of a classic salvation narrative, which is that it could kill us all, so you really need to regulate us and support it,
In short: Cited as evidence, not a stance: part of S&P 500 earnings are non-dividend payers booking the mark-up of their OpenAI and Anthropic stakes as net income. That is his "question mark" on index earnings.
3:36Look, why would you be a bear on stocks right now? Well, you might have a question mark on broad S&P 500 earnings. I think that's generally a valid question mark because there is some element of, in the non-dividend payers, reporting in your net income the appreciation of your OpenAI and your Anthropic stock.
In short: Referenced only — Lebenthal (14:12): Disney's "OpenAI partnership… really didn't go anywhere."
In short: Passing mention — expected to release its own Muse-style agent "very soon"; ChatGPT is his example of the stage-one chatbot that recommends but leaves incumbents in control of execution.
5:04It'll actually search online for any applicable way to lower the price for you as well. And this is just a preview of the new agentic world that we've entered into. Muse is just the beginning. We know that chatbt, we know OpenAI is going to be coming out with their version of this very soon. Same with Anthropic and same with all the rest.
In short: Sam Altman agreeing the industry should slow down is part of the same "shell game": OpenAI "alone represents 300 billion of Oracle's 600 billion plus backlog," so it cannot slow down and still meet its commitments. "Business is potentially slowing because token maxing is ending and open weight models keep taking market share," while data-center and capital costs rise. "AI won't cause extinction, but these two CEOs are creating massive damage."
OpenAI, the maker of ChatGPT, has promised to spend enormous sums on computing, including about $300 billion of Oracle's roughly $600 billion of signed future contracts. A company with commitments like that cannot really slow down, Eisman argues, so when its CEO agrees the industry should slow down for safety, he doesn't believe it.
He thinks the real problem is that business is getting harder. The rush of companies buying as much AI usage as possible ("token maxing") is ending, cheaper open models are taking customers, and building data centers and borrowing money are both getting more expensive. Talking up doomsday invites regulation that could shield OpenAI and Anthropic from rivals. He says the fear it has stirred up is already hurting, with local fights over data centers.
16:06And like a side street hustler, moving the shells in a shell game, Amodore and Altman don't want anyone to know that business is slowing or getting more difficult while costs are rising and capital is scarcer. They prefer scaring everyone into creating some kind of regulation that will protect their pricing power in the US.
In short: IPO pacing is the AI-financing swing factor (15:17–16:03). Santoli: after SpaceX the worry flipped from too much equity supply to "what if we don't get OpenAI and Anthropic IPO" — "they need the money to pay for the stuff they ordered." Ethridge: the risk is that the two big spenders are "not really good for those promises."
In short: "Between the two, OpenAI is the weaker company. I think that's one reason why they postponed their IPO." Frontier labs have "no moats around their business whatsoever" as open-weight models take share, so the safety alarm is a bid for regulatory capture and a duopoly. To labs saying slow down because it's dangerous: "really postpone your IPO." Consistent with "Open AI is in trouble" (Sep 14): "I don't change my mind that quickly."
OpenAI makes ChatGPT. Eisman calls it the weaker of the two big AI labs and thinks that's partly why it delayed plans to sell shares to the public. His bigger argument is that companies are switching to free "open-weight" AI models, so labs like OpenAI have nothing that stops customers leaving (no "moat").
That, he says, is why they talk up AI dangers: scary headlines lead to regulation, and regulation written with their help could lock out cheaper rivals. His retort to labs saying they must slow down for safety: then postpone your stock market listing.
3:39happen to one of those two companies, then the chain would really fall apart. And between the two, OpenAI is the weaker company. I think that's one reason why they postponed their IPO. So I think that's what you have to focus on. — That's what you said the last time you were here. — I still I don't change my mind that quickly.
In short: OpenAI "saying it may not even" IPO is a late-cycle signal. He adds the bailout posture: "Oracle and OpenAI — hey, hand me 500 billion here and there." Too big to fail "doesn't mean they're not going to fail"; a lot of people lost 100% before the 2008 bailouts came.
OpenAI hinting it may not go public at all is, to him, another late-cycle sign. He also sees it, alongside Oracle, leaning on government for support ("hand me 500 billion here and there"). Even if AI is "too big to fail" and gets bailed out, he reminds investors that in 2008 many shareholders were wiped out before the rescue arrived. Privately held.
35:21You see the IPOs being pulled. Anthropic pushing it back. OpenAI saying it may not even do it. So that to me suggests we're closer to the end. Unless we can reinvent the narrative, right? Maybe it's something about healthcare. Maybe we have a beautiful moment where some Claude model finds a miracle cure to something.
In short: Anchor tenant that also holds the equity upside. SB Energy's largest customer — 1.2 GW in Milam County and the 8 GW PORTS-Pike campus — with NVIDIA credit support on portions of its leases. Holds warrants whose revaluation drove $2.6B of SB Energy's H1 loss ("a very real economic cost of securing its most important customer"), a board-designation right while above 5%, and a nominee (Sachin Katti). Context: Sam Altman (and Anthropic's Dario Amodei) openly discussing "a more cautious pace at the frontier" is the question hanging over the whole buildout.
OpenAI is SB Energy's largest tenant, including an 8-gigawatt campus in Ohio. It also received warrants — rights to buy SB Energy shares cheaply — which is effectively the price SB Energy paid to win it as a customer, and it gets to name a board member. So OpenAI is on both sides: it owes decades of rent and it profits if SB Energy's stock rises. If OpenAI slows its build-out, SB Energy feels it first. Analysis, not a recommendation.
In short: "OpenAI to me is the one that potentially has the most issues because from the beginning, they were focused on consumers… only 5% of those consumers actually pay. Why? Because we're all used to getting things for free from Google." So "OpenAI in particular are stuck between those two guys. Anthropic on enterprise, Google on consumer." Anthropic's run rate (~$65B) now exceeds OpenAI's (~$40B), and "a lot of roads lead to OpenAI" in the circular financing.
OpenAI makes ChatGPT and bet mainly on ordinary consumers. The problem, Niles says, is that only about 5% of ChatGPT users pay, because we're all used to getting answers from Google for free — and Google is now putting AI directly into free search. Meanwhile Anthropic has taken the business customers who actually pay. That leaves OpenAI squeezed in the middle.
It also matters beyond OpenAI itself: many of the circular financing deals in AI (companies investing in each other and then buying from each other) lead back to OpenAI, so its weakness can ripple through the whole sector.
15:34I think you're going to have some losers within that group. And for me, I've said this now for a long period of time, I think OpenAI to me is the one that potentially has the most issues because from the beginning, they were focused on consumers, right? ChatGPT. Well, only 5% of those consumers actually pay. Why? Because we're all used to getting things for free from Google.
In short: Needs ~$600B "to get through the next two years" (low end) with $400B committed to neoclouds, after a ~$960B last round, and "the private money is out." The regulation push is Altman's "panic button." With too many eating a small pie, "they got to get rid of one. Guess who that is? … it's OpenAI," possibly via bankruptcy that wipes out investors while the tech survives.
OpenAI (ChatGPT) last raised money at roughly a $960 billion valuation and, by Taylor's estimate, needs at least $600 billion more over two years, having promised $400 billion to data-center builders. He thinks private investors are tapped out. He reads the sudden wave of "AI must be regulated" headlines as an attempt to lock in a small club of US players who could keep prices high.
If that fails, AI pricing falls and not everyone survives. Taylor expects OpenAI to be the one squeezed out. That doesn't mean the technology disappears: companies can keep operating through bankruptcy and be bought. But the current investors would be wiped out.
17:53And then I think the next part, right after the deal gets done, pricing. Because if they can't regulate it and kick everyone out. There's too many people eating the pie and the pie isn't big enough. And so they got to get rid of one. Guess who that is? Sam Altman. Yeah, OpenAI. Yeah, it's OpenAI. And that's the one that's...
In short: Backs the slowdown, and says an IPO now would be ill-advised. Rooney: Sam Altman tweeted support for Amodei, but "told Fortune over the weekend an IPO right now would be ill advised. He did say 2027 for their own listing" (CNBC had already reported next year). Amodei's essay cited "an example of OpenAI agents hacking into the startup Hugging Face." Rooney's framing of the labs' bind: they must impress Wall Street on revenue growth — "a big way to do that is to have the best and most expensive model" — while holding back for liability and regulatory reasons.
In short: Same answer on an OpenAI IPO — "there has to be a price for everything." Paused "Astra" over security risk, then launched "Astron" as the start of AGI: the national-security race with China now overrides safety; its $500B data-center plan feeds the AI debt wave; falling token prices squeeze margins.
OpenAI, the maker of ChatGPT, is also a possible listing candidate, and Woo's answer is the same: "there has to be a price for everything." He points out that OpenAI paused a model over security fears, then weeks later launched one it called the start of AGI (human-level AI). He reads that as the government deciding that beating China matters more than the safety risk.
He also notes that AI usage is growing fast partly because prices are collapsing, which squeezes profit margins for companies like OpenAI.
35:01I mean that is the greatest irony of the situation we're in and it's also the biggest single tail risk for the entire stock market and the global economy in my view at this point. — Would you buy the anthropic IPO if it comes out, or even OpenAI's IPO for that matter? — I mean there has to be a price for everything. Not at two trillion dollars. Forget it.
In short: Sam Altman partly joins the call ("we could slow down the frontier and be a little bit more cautious"), and OpenAI shares Anthropic's incentive: "Anthropic and OpenAI have a very big incentive to reduce costs. They don't want to keep paying hundreds of billions of dollars for training." If the pacing regime is granted "they will be the duopoly that controls the world and AI, the regulatory capture will be complete. Every other company trying to catch up will be legally prohibited from catching up." Commoditization is its existential risk — "Anthropic and OpenAI go to zero" is his Colossus-sourced (exaggerated) bear case.
OpenAI, maker of ChatGPT, is Anthropic's closest rival at the frontier, and CEO Sam Altman partly endorsed the slowdown idea. Carlson lumps the two together: both face the same huge training bills and would both benefit from a pause that locks in their lead — "the duopoly that controls the world and AI." Their long-term risk is the opposite scenario: if AI models become cheap and interchangeable, a company whose whole business is selling the model is worth much less.
29:39" And now that we have world dominance, we'll slow down, too. We'll be the responsible ones. See how responsible we're being? And if this gets granted, if Daario does get his wish, then they will be the duopoly that controls the world and AI, the regulatory capture will be complete. Every other company trying to catch up will be legally prohibited from catching up to OpenAI Anthropic.
In short: Eisman: "Open AI was at 6.5 billion and it was up only 18% in 3 months. Its costs were 12 billion… their revenue went up a billion and their cost went up three. So I think Open AI is in trouble." Losing share daily and people; "your cost of capital is rising… and you can't afford that." His joke: "whoever buys OpenAI out of bankruptcy, it's going to be a fantastic deal" — the railroad/Wachovia/Bear Stearns pattern. Earlier: if it failed and the capex reversed, "the economy would go into recession almost immediately" (not yet his call).
OpenAI, the maker of ChatGPT, had about $6.5 billion of sales last quarter, up only 18%, while its costs were about $12 billion. In three months its sales rose $1 billion and its costs rose $3 billion, so the losses are getting bigger, not smaller.
A company losing money depends on investors willing to fund it. When the story around it turns sour and people leave, that funding gets more expensive. Eisman thinks OpenAI is in trouble and jokes that whoever buys it out of bankruptcy will get a bargain. The concern is wider than one company, because so much AI spending and economic growth depends on it.
23:09They didn't say anything about costs and that was up over 100% in 3 months. — Okay. — Open AI was at 6.5 billion and it was up only 18% in 3 months. Its costs were 12 billion. But the crazy thing was that if you look at forget about these the percentage in increases if you just look at the dollar changes in three months their revenue went up a billion and their cost went up three.
In short: The strategy Apple is defining itself against. Ternus' Intelligent Personal Hub walkthrough — concluding the perfect AI device "already exists. It's the iPhone" — "sounded almost like an early rebuttal to OpenAI and Jony Ive's work on a new AI device." Apple's approach "is a very different strategy from OpenAI or Anthropic. Apple does not need users spending hours inside an Apple chatbot." Referenced as the competitive foil; no stance on OpenAI itself.
In short: She'd wait for the IPO rather than buy proxies. The labs aren't threats to niche businesses — they're "interested in solving intelligence" — but the "big error of omission" is "what's the marginal demand for higher intelligence?" if open-source models are good enough: is future compute "complete malinvestment? I don't know the answer. My guess is probably not." The companies behind LLMs are "kind of weird religious organizations that are very bad at advertising," and she expects "the most insane things you've ever heard on an earnings call" once public.
In short: Cited on both sides of the safety argument and as one leg of Oracle's backlog risk. Gerstner: "Sam Altman just recently said he's telling staff that OpenAI is open to slowing AI development, that he's pushing for mandatory national AI safety requirements in the US… doesn't that make you feel great? Did you ever hear that out of Mark Zuckerberg about social media?" He also uses the unreleased model that "cracked Navier-Stokes… an Astra plus one model working inside of OpenAI" as proof that pre-release government scrutiny is real: "why doesn't the public have access to this model? Because it's undergoing scrutiny by the lab and by the government." On the other side, Lebenthal concedes the concentration critique of Oracle's backlog ("too much of that is OpenAI"), and Harrington names the dated catalyst: "I'll be curious once we see the S-1s from OpenAI and Anthropic… what if there's a little bit of pullback? I think the trickle from a marginal pullback would be pretty painful."
OpenAI is the other private lab at the centre of the argument, and also an Altimeter holding. Gerstner uses Sam Altman's stated openness to slowing development and to mandatory national safety rules as evidence the industry polices itself, contrasting it with social media's leaders, who never volunteered anything similar.
For investors the more concrete points are two. First, OpenAI is a large share of Oracle's roughly $700 billion order backlog, so one customer's fortunes sit inside another company's valuation. Second, both OpenAI and Anthropic are expected to file S-1s — the public registration document a company must publish before an IPO, containing real financials for the first time. Harrington expects those filings to be a genuine market event: if the numbers disappoint even slightly, the read-through to everything priced off the AI build-out "would be pretty painful."
In short: "I still believe that if OpenAI fails within a year, there will be a massive correction in the stock market" — the generation-one dot-com risk. Its CFO disclosed a price cut on GPT 5.6 Luna shortly after its July release, claiming "a tenfold increase in model usage" — "Is this the harbinger of a price war? We shall see." It is also ~50% of Oracle's RPO, "a company that we all know has massive negative cash flow."
OpenAI, the maker of ChatGPT, is private, loses a great deal of money, and is the single biggest customer behind a lot of today's AI spending — about half of Oracle's signed future contracts, for example. That is why Eisman says that if it failed within a year, the stock market would suffer a massive correction.
This week its finance chief said OpenAI cut the price of its GPT 5.6 Luna model shortly after launch, and usage jumped tenfold. More usage sounds good, but cutting prices is not what a company short of money wants to do, and if rival Anthropic has to match, it could start a price war that squeezes both of the labs everyone else is depending on.
15:18The first generation of dot-com companies failed miserably, and it was only the second generation companies like Google that went on to glory. I still believe that if Open AI fails within a year, there will be a massive correction in the stock market. We shall see. Moving on, summer is over and equity conference season has begun.
In short: Raised by the host as half of the AI ecosystem with an IPO expected this year. His answer: the hyperscaler link to nuclear is real, the AI IPOs will create "a lot of competition for capital," but base-load demand for data centers (plus reshoring) keeps the nuclear case intact either way.
2:10But obviously, we're seeing tremendous new sources of demand for uranium in the coming years. One of the new sources is from AI and I want to get your thoughts on this whole AI narrative because this whole ecosystem is really based on two companies, OpenAI and also Anthropic. They're going to go public sometime this year and there's all sorts of speculation out there on how these IPOs will go.
In short: The other lab named in ex-researcher Jacob Coxin's viral resignation post ("neither company is acting responsibly"), and the subject of the "Hugging Face hack" — which Carlson says was agents instructed to hack a system, not AIs acting "of their own volition," grossly misrepresented by Coxin and in Bernie Sanders' bill.
8:56language. What happened with the hugging face hack wasn't nearly as interesting as people might think. The system was told to hack a system and then the agents worked on hacking it. The monitoring systems could have easily stopped it and then the letters that the agents left at the end were logs by the
In short: Le Shrub: exhibit one of the pre-IPO narrative push — hours after his piece, OpenAI claimed to have solved Navier-Stokes, a millennium problem, then it emerged it had used human researchers' methodology, "burned, like, 22 million in tokens for a $1 million math prize… the ROI is negative" ("we spent $2 trillion to steal someone's math homework"). Paulo adds congressional letters accusing it of misleading, and data-center opposition now "four or five to one."
In short: The single counterparty the loop runs through. It supplies $24.1bn of Microsoft's $34.3bn fiscal-2026 AI revenue, is half of the pair estimated at ~70% of hyperscaler AI revenue and of the UBS 48%-of-Google-Cloud projection, and, with Anthropic, "remain[s] deeply unprofitable, having lost tens of billions of dollars between them" — while being funded by the very hyperscalers booking its payments as revenue.
OpenAI is the single node the whole chain runs through. It supplies most of Microsoft's reported AI revenue, is half of the pair estimated to account for about 70% of all hyperscaler AI revenue, and is half of the pair UBS thinks could be nearly half of Google Cloud next year.
It is also, together with Anthropic, "deeply unprofitable" — the two have lost tens of billions of dollars between them. The combination is what makes the structure fragile rather than merely unusual: the company generating the revenue that justifies a trillion dollars of chip purchases cannot yet pay for itself, and is kept going by the same companies booking its payments as sales.
In short: Same reference: one of the names banks have "drifted" into lending against off balance sheet. The risk sits with the lenders, not stated as a view on OpenAI itself.
Same role as SpaceX in his argument: one of the borrowers the banks are "trying to appease." OpenAI is privately held, so there is nothing to buy or sell here.
The takeaway is about the lenders. When banks stretch outside their normal territory to keep a marquee client, that stretch is where the next credit accident usually starts.
10:21They're trying to appease SpaceX, they're trying to appease OpenAI, they're trying to appease the Mag 7. And so yeah, they've drifted what we call style drift. They've drifted into areas that are fairly speculative for them because it's very attractive for a bank because think of like the Mag 7 companies, they have double A rated free cash flow.
In short: Named as one of the AI developers Shopify wants tapping its merchant-product catalog for recommendations — the demand side of the coming "agentic commerce" shift, where agents rather than humans search for and recommend products.
In short: Appears as an accounting object, not an investment view. Dawson: "there's a really fun accounting thing that's going to happen, because there's two different ways that you account for investments in other firms. One is called the equity method, which has to be consolidated into the income statement… Well, Microsoft owns 25% of OpenAI and so it accounts for it within the equity method. We're going to be having to see a lot more disclosures about these companies over time." The point is that a private company's economics are about to become visible inside a public company's income statement.
OpenAI is private, so there is nothing to buy — it appears here because of how it will start showing up in someone else's accounts.
When one company owns a large but non-controlling slice of another (roughly 20–50%), accounting rules make it use the equity method: instead of ignoring the stake or fully merging the other company's books into its own, it records its share of that company's profit or loss as a line in its own income statement. Microsoft's 25% stake means OpenAI's economics land inside Microsoft's reported earnings.
Dawson's forecast is simply that this forces disclosure: "we're going to be having to see a lot more disclosures about these companies over time." For anyone trying to size the AI build-out from the outside, that is one of the few places where a private AI company's numbers become publicly visible.
22:56It makes it really murky and hard to see. And there's a really fun accounting thing that's going to happen because there's two different ways that you account for investments in other firms. One is called the equity method which has to be consolidated into the income statement, and is when you own a certain percentage of a company — if you have a controlling stake, effectively not controlling stake but a certain degree of stake in a company. Well, Microsoft owns 25% of OpenAI and so it accounts for it within the
In short: Named by Nathan as the leader Meta's models have been "lagging in the performance" of — and then folded into the argument that the ranking will not matter: every lab has "a ball people, spend hundreds of billions of dollars training them… they're all going to be commoditized." A peer reference in a commoditization thesis, not a view on the company.
30:06One of the knocks on Meta is that they basically have been lagging in the performance of their models to OpenAI, Anthropic, even Gemini. And the monetization of all that spend. They don't have that big cloud business like Amazon, like Microsoft, and like Google.
In short: Named only alongside Anthropic as proof the US leads frontier models: "the US is obviously at the lead with Anthropic, OpenAI in terms of the frontier models." No repeat of his squeezed-between-Anthropic-and-Google argument here.
26:31It's going to be pretty darn good in terms of low-cost token production as well. And so I don't think you're going to see the US companies fall that far behind. It's just the US is obviously at the lead with Anthropic, OpenAI in terms of the frontier models, but I think you're going to have models at the lower end that are pretty darn good, too.
In short: The name he worries about most, on three counts. Structurally it is squeezed — "Anthropic is going to win on the enterprise side and then Google on the consumer side are going to squeeze OpenAI between them" — and consumers won't pay: "we've been all trained that hey, we get answers from Google for free. So why on earth would you pay for it?" Financially the gap is widening: Anthropic "turned profitable in Q2. And OpenAI, I believe, lost even more money in Q2 relative to Q1." And on the base rate, "just because ChatGPT is what started all this doesn't mean that OpenAI is going to be ultimately the biggest winner."
OpenAI, the maker of ChatGPT, is the name he says he worries about most, and the argument has three independent legs.
First, position. He expects the market to end with a business-facing winner (Anthropic) and a consumer-facing winner (Google), leaving OpenAI "squeezed between them." Its original home turf, consumers, is the hardest place in technology to charge money, because Google spent two decades teaching everyone that answers are free.
Second, the financials are moving the wrong way relative to its rival: Anthropic turned profitable in the June quarter while OpenAI, he believes, lost even more than in the March quarter.
Third, the base rate. Technology markets historically converge on one dominant winner — one in e-commerce, one in social, one in streaming — and the company that starts a category is frequently not the one that ends up owning it. AOL, Yahoo, Netscape and Lycos all looked inevitable once. "Just because ChatGPT is what started all this doesn't mean that OpenAI is going to be ultimately the biggest winner."
18:23Do you think that changes and maybe the value starts to accrue down away from them? — Yes, I do. And I worry more about OpenAI than I do Anthropic, because if you look at the genesis of both companies, ChatGPT was launched for consumers. Anthropic started focusing on enterprise, and I've been saying this now for I don't know a year or two at least, that I think Anthropic is going to win on the enterprise side and then Google on the consumer side are going to squeeze OpenAI between them.
In short: The named exception to his one bullish concession about this cycle versus 1999: "there are a lot of analogies to the dot-com bubble 25 years ago… but the one thing that's different is there are profits. I mean, except for maybe in OpenAI and Anthropic, but there are profits." Farley pushes back that the labs' revenue growth has been "among the best ever for history of companies," and Dillian defers — "you know more than me on that."
OpenAI is the private company behind ChatGPT, and one of the two large AI labs named in the episode.
It appears in the closing valuation argument. Dillian's one concession that this cycle is not 1999 is that today's leaders actually earn money — "the one thing that's different is there are profits. I mean, except for maybe in OpenAI and Anthropic." So the labs are the exception that proves the pattern, not a company he has an opinion on.
Farley pushes back that their revenue growth has been "among the best ever for history of companies," and Dillian concedes the ground rather than arguing: "you know more than me on that."
31:31dot-com bubble 25 years ago, 26 years ago but the one thing that's different is there are profits. I mean, except for maybe in OpenAI and Anthropic, but there are profits. So yes, but that revenue growth in Anthropic and OpenAI has been tremendous. So the bear argument was like where's the revenue in the labs, OpenAI, Anthropic. And I just want to say as someone who's kind of calling balls and strikes, the revenue growth has been quite quite good, like among the best ever for history of companies.
In short: The show's other main character, in three roles. The soundbite that names the segment — Sam Altman: "I'm not worried about our compute build out plans. I am worried about the world's compute build out plans… I am seeing the first signs of what feels to me like unsustainable silliness of random new neoclouds popping up, people claiming that they're going to build gigantic massive compute next year that I think they don't have the revenue to support or a buyer." Weiss discounts it on incentives: "you have to disregard somewhat what Sam Altman said, because… from a competitive standpoint, he doesn't want more going up because he's already locked in what he needs." Liz Thomas agrees on the tell: "spoken like a true CEO. We're going to do it right, but everybody else is who I'm worried about." Also: OpenAI's cyber-tool releases are part of Weiss's bear case on the security vendors, and the news update reports the administration filed in support of OpenAI in the New York Times training-data suit, arguing AI training is generally fair use.
OpenAI is private, but it drives the episode's central segment. Sam Altman says he is comfortable with his own company's compute plans and worried about everyone else's — specifically "random new neoclouds" (small firms renting out AI computing power) announcing enormous build-outs without the customers or revenue to pay for them.
Weiss's response is the one to remember, because it applies to every CEO comment about an industry: check the incentive. OpenAI has already locked in the capacity it needs. A wave of new supply would lower the price of the thing it has bought, and slow the build-out of rivals it competes with. That does not make Altman wrong — a lot of unfunded capacity really is being announced — but it does mean his warning is not disinterested, and Liz Thomas independently reaches the same read: "spoken like a true CEO."
OpenAI is also the reason two other trades move here: its cyber-tool releases are part of Weiss's bear case on the security vendors, and the administration filed a brief supporting it in the New York Times copyright case, arguing that training AI models is generally fair use.
In short: Its revenue is the evidence the AI build-out isn't a scam — "real revenue from real customers because of real use cases," growing because usage keeps expanding.
In short: (Private.) Raised only as the loser of Talkington's Apple argument, on hardware: "no one's switching to an Android, no one's switching to any other device. We don't want an OpenAI device, we want the Apple device." Sigalos separately notes that if Apple monetizes a Siri AI tier, third-party models plugged into the App Store are one of the routes — which cuts both ways. No committee position.
In short: With Anthropic, the ARR the whole complex trades off: "people are latching on to every ARR leak of Anthropic or OpenAI" in an information vacuum. Both are also building their own power and data centers — part of why the profit pool's landing spot in the compute chain is unclear.
19:46It's not like June was the blowoff top and we're going to be, semiconductors are going to be bleeding for years. I think that this couple months is a little bit more of an information vacuum. People are latching on to every ARR leak of Anthropic or OpenAI. I think that'll get clearer when they become public.
In short: The credible challenge to Nvidia's silicon monopoly, and a live financing question. Custom chip: "OpenAI did some internal testing showing that its Broadcom-built Jalapeño inference chip outperformed Nvidia's GB300 in throughput per watt and latency at 700 watts, with deployment targeted for later this year" — restated Tuesday as "Palomino." Scale: ClickHouse's ARR is being driven by "OpenAI scaling to 30 petabytes a day, which is unbelievable" (~30 trillion events daily, with log management shifted off Datadog). People: "OpenAI's data center head, Chris Malone, left the company last week" — weeks after the chief revenue officer was replaced. Funding: SoftBank's $6.3B bond is "basically to put more money, I would think, into OpenAI," with cumulative commitments past $60 billion. And the reflexive link back to the AI trade: Nvidia's guide depends partly on "OpenAI securing financing."
Three OpenAI items this week, pointing in different directions.
The first is a real technical threat to Nvidia. OpenAI's own chip, designed with Broadcom, beat Nvidia's GB300 on performance per watt and on response speed in internal testing. It is built only for inference — running a finished model — not for training, which remains Nvidia's fortress. The caveat is that the comparison was against GB300 rather than Nvidia's newer generation. But the largest single buyer of AI chips now has a credible alternative for the larger half of the workload.
The second is scale: OpenAI is generating 30 petabytes of data a day, roughly 30 trillion events, and that alone is driving the growth of the database company ClickHouse.
The third is people and money. Its data-centre head left last week, weeks after the chief revenue officer was replaced — turnover at the top of the function responsible for building out capacity. And SoftBank is raising $6.3 billion from Japanese retail bond buyers to put more money in, with cumulative commitments past $60 billion.
Put together: a company whose engineering is strong enough to challenge Nvidia's product, whose costs are large enough to need continual outside financing, and whose ability to secure that financing is one of the stated conditions on Nvidia's own forecast.
Full passage: premium transcript (PDF).
In short: Both NVIDIA's customer and the sharpest named threat to it. OpenAI "just published the first results for Jalapeño, its custom inference chip," claiming it "delivered 1.5x–1.9x more throughput per watt and materially lower latency than the NVIDIA systems tested across several models." App Economy's read is measured: "OpenAI still plans to use NVIDIA broadly, but Jalapeño shows that NVIDIA's largest customers have a growing incentive to move specialized inference workloads onto their own silicon." Jensen Huang's rebuttal on the call is aimed squarely at it — "whereas many of these XPUs are inference-specific chips for one cloud or one service, NVIDIA is a platform, an entire AI factory platform that spans the entire AI life cycle that you can use in any cloud" — with App Economy adding that "his argument isn't that customers won't build their own chips. It's that those chips tend to optimize specific workloads, while NVIDIA's advantage is a fungible platform… The question is whether that breadth remains valuable enough to justify NVIDIA's premium economics." Listed first among "what I'm watching." Named as a competitive/strategic factor, not as a stance on OpenAI itself.
OpenAI is both NVIDIA's biggest kind of customer and, increasingly, a competitor. It has designed its own chip, Jalapeño, built for one specific job: inference — running an already-trained AI model to produce answers. That is different from training, the enormously expensive process of building the model in the first place. Training is varied and experimental; inference is the same operation repeated billions of times, which is exactly the sort of work a purpose-built chip can do far more efficiently than a general-purpose one.
OpenAI's published results claim 1.5 to 1.9 times more throughput per watt than the NVIDIA systems it tested, with lower delay. Throughput per watt is the metric that matters at data-centre scale, because electricity — not the hardware price — is the binding constraint and the largest running cost. If you can serve the same number of answers on half the power, you can serve twice as many customers from the same building.
OpenAI says it will keep using NVIDIA broadly, and that is credible: designing chips is slow, and most workloads still benefit from flexibility. But the direction of travel is the point. As the author puts it, Jalapeño shows "NVIDIA's largest customers have a growing incentive to move specialized inference workloads onto their own silicon." Inference is also the part of AI demand expected to grow the most, so losing share there matters more over time than losing training share would.
NVIDIA's counterargument, in Jensen Huang's own words, is that these custom chips are "inference-specific chips for one cloud or one service," while NVIDIA sells "an entire AI factory platform" usable anywhere across the whole lifecycle. Whether that breadth stays worth paying a premium for is, in the author's framing, the central open question. Named here as a strategic factor in the NVIDIA read, not as a view on OpenAI itself. Analysis, not a recommendation.
In short: The frontier-model economics are the problem, not the technology: with token prices deflating ~80% a year, standing still requires 400% unit growth — "OpenAI apparently in its last quarter did 18% quarter after quarter, and that was deemed a disappointment." Wile E. Coyote over thin air: "I need to grow at this speed just to stand still," now with external debt service on top.
OpenAI sells access to AI models, priced by the "token" — roughly, per chunk of text processed. The price of a token is falling about 70–80% every year, faster than any commodity in modern economic history.
Do the arithmetic and the problem is stark: if your price drops 80%, you need five times the volume — 400% growth — just to earn the same revenue as last year. Not to grow. To stand still. Kedrosky's image is Wile E. Coyote over the canyon, legs spinning: "I need to grow at this speed just to stand still."
That's the lens he says you should use on the headline numbers. OpenAI reportedly grew 18% in a quarter and that was treated as a disappointment — and now there's external debt in the structure that has to be serviced regardless. Very useful technology; brutal economics for whoever owns the frontier model.
20:37still," let alone producing. What is Wall Street going to want from these guys? 50% year-over-year growth, 100% year-over-year growth. OpenAI apparently in its last quarter did 18% quarter after quarter, and that was deemed a disappointment. So, this tells you what we're up against here. Not just the expectations but structurally the economics of riding this deflating exponential are so consequential in terms of both the kinds of top-line growth you can deliver but also in terms of paying the fixed obligation that comes from having more
In short: "The weak sister" / "the problem child," and the sequential math is the case: June-quarter revenue ~$6.5B, "up only 18% versus the March quarter. Worse, its costs went to 12 and 1/2 billion, up 3 billion in 3 months… revenue in 3 months was up a billion dollars sequentially and its costs were up 3 billion sequentially. That's not the right direction." Growth "has obviously slowed dramatically. It better start reaccelerating soon," and with token maxing over "it's going to be harder for these guys to raise capital." The stakes: "if tomorrow OpenAI failed, the US economy I think would go into an immediate recession, and the market would have a massive correction."
OpenAI builds the models behind ChatGPT. It rents its computing power from other companies, and it is losing money, so it stays alive by raising fresh capital. Eisman now separates it from Anthropic and calls it the weaker of the two — "the weak sister," "the problem child."
The reason is arithmetic done quarter to quarter rather than year to year, which shows a turn much earlier. In the June quarter OpenAI took in roughly $6.5bn, only 18% more than the previous quarter — while its costs rose to $12.5bn, an increase of $3bn in three months. Revenue up $1bn, costs up $3bn: "that's not the right direction." A company whose spending is pulling away from its income has to keep persuading investors to fund the gap, and Eisman thinks that persuasion is about to get harder.
Why anyone outside Silicon Valley should care: OpenAI's promised future spending is other companies' promised future revenue — half of Oracle's $600bn order book, a large share of Microsoft's, and, through the hyperscalers, a share of Nvidia's. Add that AI capital spending is roughly half of all US economic growth this year and the conclusion follows: "if tomorrow OpenAI failed, the US economy I think would go into an immediate recession."
8:33— Mhm. — And it was up over 100% versus the March quarter. Open AI, which like I said is the problem child, had around 6 and 1/2 billion in revenue and it was up only 18% versus the March quarter. Worse, its costs went to 12 and 1/2 billion, up 3 billion in 3 months. So, simple math, Open AI's revenue in 3 months was up a billion dollars sequentially and its costs were up 3 billion sequentially.
In short: Signpost number one: Altman and CFO Sarah Friar floating government financing at the end of last year "because the engines of financing were starting to crack." Since then "OpenAI has a lot of people on the inside leaving… that suggests things aren't going so well internally." Its Stargate $500B was also an unfunded MOU.
OpenAI is the private company behind ChatGPT and the biggest single driver of the AI spending story. Dowd doesn't value it — he reads its behaviour as the first crack.
Late last year Altman and CFO Sarah Friar floated the idea of government financing. Companies with easy access to capital never ask taxpayers for a backstop, so to Dowd that was "salvo number one": the private funding channels — especially private credit, the main lender to the AI build-out — were starting to freeze. Since then, senior people have been leaving, which he reads as insiders knowing something the outside doesn't.
The Stargate project fits the pattern: a headline $500 billion announced at the start of the Trump administration that was only a memorandum of understanding and is still unfunded.
2:05So, yeah, what are you currently seeing there? — So, there's a couple signposts that I watch. I watch what the players are doing. So, at the end of last year, we're going to remember OpenAI, Sam Altman and the CFO, Sarah Friar, floated the idea of government financing. And why did they do that? That's because the engines of financing were starting to crack a little bit.
In short: The deceleration datapoint of the week: "OpenAI said revenue reached 6.7 billion at the end of Q2, up only 18%, and the market was very worried that OpenAI's growth rate is decelerating given hundreds of billions of dollars it needs to spend on CAPEX." Also the tenant behind Nvidia's $1.5B SB Energy investment for 4.25 GW at the Ohio campus — "because energy is the main bottleneck."
OpenAI, the maker of ChatGPT, reported second-quarter revenue of $6.7 billion — up only 18%. For a company that needs to spend hundreds of billions of dollars on computing capacity, a growth rate that starts with a one is a problem, because the spending is committed years in advance while the revenue has to keep compounding to justify it.
The scale of that commitment appeared in the same week from the other direction: Nvidia invested $1.5 billion in an energy developer to secure 4.25 gigawatts of electricity for OpenAI's campus in Ohio, "because energy is the main bottleneck." Getting power, not chips, is now the binding constraint.
Singh does not take a directional view — OpenAI is not listed — but treats the slowing growth as evidence for his broader argument that the market has stopped rewarding AI spending automatically and started asking when it pays back.
Full passage: premium transcript (PDF).
In short: Guest: a new Zeta activation channel — enterprises launch ads/brand placement inside ChatGPT, and Zeta resolves identity for the ~30% of logged-out users; OpenAI is "incentivized to have this robust ad revenue" ahead of an IPO.
In short: (Private.) Raised twice, both times as a soft spot in the AI complex. Raskin, on why NVIDIA's print is not a foregone conclusion for the momentum trade: "they want to see OpenAI. OpenAI's growth is a little disappointing. There's a lot of things in there." And Harrington names it among the businesses for which NVIDIA's guidance genuinely matters — "it might matter more to CoreWeave or Microsoft or OpenAI" — the customers whose economics move on GPU supply and pricing rather than the supplier itself. No committee position.
OpenAI is private, so there is nothing to buy — but it appears twice as a soft spot in the AI story.
Amy Raskin raises it when explaining why Nvidia's results are not an automatic win for the momentum trade: investors "want to see OpenAI," and its growth is "a little disappointing." If the most visible customer of AI infrastructure is growing more slowly than expected, that is a question mark over the demand everyone is extrapolating.
Jenny Harrington names it from the other direction — as one of the businesses for which Nvidia's guidance genuinely matters, along with CoreWeave and Microsoft. Her point is that the buyers of chips, not the seller, are where the uncertainty actually lives.
In short: (Private, on file with the SEC.) The unfavourable half of the AI-lab comparison, and Wapner frames the session around "the disparity between the way the street is starting to view these two names." Per the Wall Street Journal overnight, OpenAI's Q2 sales showed only tepid growth versus Anthropic's, and Kate Rooney's numbers are the story: "operating loss grew to $12.3 billion in the quarter, up from about $9 billion in Q1. Losses also outpaced revenue growth — revenue for the quarter just under $7 billion, up about 18% from the first quarter." Set against Anthropic's EBITDA profitability and $65B annualised run-rate, OpenAI's is "around $40 billion." The mitigants: the data is backward-looking (through end-June), and — per Jim Cramer's sources and confirmed on air by president Greg Brockman — July revenue grew 20%, enterprise 32%. Rooney's caution: "it doesn't speak to profitability, and that's the thing that's really getting attention." Also in the background: "OpenAI has been lately in the news about top names leaving." No committee position.
OpenAI is the one being marked down in this comparison. Per the Wall Street Journal, its operating loss grew to $12.3 billion in the second quarter, up from about $9 billion in the first — and, crucially, the losses grew faster than the revenue, which was just under $7 billion (up about 18% on the quarter). Growth that costs more to buy each quarter is the pattern investors punish.
The fair counterweights, both stated on air: the data is backward-looking, covering the three months to the end of June, and things appear to have improved since — July revenue grew 20% and enterprise revenue 32%, confirmed on the network by president Greg Brockman. But as Rooney notes, revenue growth "doesn't speak to profitability, and that's the thing that's really getting attention."
Why it matters now: both OpenAI and Anthropic are on file with the SEC to go public, so the two will be compared directly by public investors, and the audited numbers in their prospectuses will settle it. The steady drip of senior departures is an added overhang.
In short: The circularity in the cloud numbers: hyperscalers "have doubled their backlogs, their cloud backlogs and the risky part is a lot of it's circular financing — a lot of the backlogs are OpenAI and Anthropic, and that's money that's being raised in the public markets and from VCs."
22:12And you can see that in the numbers, right? You can see that Microsoft, Amazon, all these companies have doubled their backlogs, their cloud backlogs and the risky part is a lot of it's circular financing and that a lot of the backlogs are OpenAI and Anthropic and that's money that's being raised in the public markets and from VCs.
In short: The other half of the same dependency. Trennert: "totally negative cash flow," Chinese competition, and together with Anthropic roughly 70% of hyperscaler AI revenue — "if Open AI and Anthropic ever get in trouble, the ecosystem is in trouble." Independent confirmation of the map Eisman laid out solo on the Aug 14 wrap.
The other half of the same dependency, and the same characterisation: heavily negative cash flow, funded by capital raises, facing cheaper Chinese competition, and jointly responsible for about 70% of hyperscaler AI revenue.
What makes this worth recording is the corroboration. Two strategists who cover the entire market, with no stake in Eisman's thesis, arrived at the same single point of failure from their own client work. When independent analyses converge on the same node, the node is real even if the timing is unknowable.
24:15They're totally negative cash flow. Right. Right. Right. The Chinese are competing with them and I read this report by a firm where they basically said that something like 70% of the AI hyperscaler revenue is from just those two companies. So if this is the big if, if Open AI and Anthropic ever get in trouble, the ecosystem is in trouble.
In short: He hasn't reported on it directly, but relays the FT house view: "Anthropic looks financially a lot healthier than OpenAI." His own scepticism is about the shape of private revenue — "how much of that is actually cash, like free cash?" — and he is openly waiting for the filing: "I'm really looking forward to the S-1s for OpenAI and Anthropic. That's going to be a popcorn moment."
53:13So I only know what my colleagues have reported in the paper. I think it's broadly understood that Anthropic looks financially a lot healthier than OpenAI. And that's one of the reasons why they're probably going a little bit more aggressively for an IPO now. But I'd question with private companies how real sometimes revenue is — not like fending numbers but just, if you just look at the net income of some of the hyperscalers, the public companies now, look at how much is actually classified as other income,
In short: Referenced — OpenAI became an Adyen customer, "reinforcing Adyen's positioning around AI-native businesses and agentic commerce." A demand-side datapoint for the payments platform, not a view on OpenAI.
In short: "Much more problematic… there just don't seem to be any moats, or at best, the moats are shallow… the future for these large LLM providers is very questionable." And the transmission channel is its order book: "of Oracle's 600-plus billion backlog, around half is from OpenAI."
OpenAI builds the AI models behind ChatGPT. It does not own the data centres it runs on — it rents them, in enormous quantity, from hyperscalers like Microsoft and Oracle.
Eisman's problem is that there is nothing stopping a customer from leaving. "There just don't seem to be any moats, or at best, the moats are shallow. Enterprises are switching between models and using cheaper open-source Chinese models in order to control costs." A business with no lock-in and cheaper substitutes appearing is a business that competes on price, and price competition against a rival with a far lower cost base rarely ends well: "the Chinese models are much cheaper, and this could eventually cause a price war."
The reason this matters beyond OpenAI is that its spending commitments are other people's revenue — roughly half of Oracle's $600bn-plus backlog. So its financial health is not a private-company curiosity; it is the load-bearing assumption underneath a large slice of listed technology.
4:53This is Steve talking about the LLM providers, both OpenAI and Anthropic. — If the lack of moats begins to cause them problems, then the entire AI ecosystem could go through a correction phase because so much of the hyperscaler backlogs are from these two companies. For example, of Oracle's 600-plus billion backlog, around half is from OpenAI.
In short: The demand side of two of the three stories. Named first among the customers CRWV rents compute to (part of the 98% committed-contract base), and again at Cerebras, where "OpenAI remains a major customer" inside $25.4B of remaining performance obligations — a backlog the issue notes "requires substantially more infrastructure" to convert. Referenced; not a stance call.
In short: The other half — and the one with the visible counterparty exposure: "Oracle has a $600 billion backlog and half that backlog is from OpenAI alone." Loses billions, depends on capital raises, no moat around the model, and is "racing to IPO." Until that IPO, "there is still no data really to allow us to accurately measure the risk, but we all know it's there."
OpenAI is the other half of the dependency. The concrete number he keeps returning to: Oracle reports a $600 billion backlog of future cloud revenue, and about half of it is OpenAI alone. A backlog is a promise to buy; a backlog concentrated in one loss-making customer is a very different asset from one spread across a thousand.
Like Anthropic, OpenAI funds itself with capital raises rather than profits, sells a product with no obvious lock-in, and faces much cheaper Chinese competition. Eisman isn't predicting failure — he is naming the exact thing to watch, and pointing out that the only honest verdict today is "we don't know," because the financials are private.
His practical conclusion: the AI trade keeps working "until this Anthropic and OpenAI risk metastasizes." And when they IPO, the supposition finally becomes data.
8:45These reports state that 70% of hyperscaler AI revenue is from anthropic and open AI and 25 to 35% of total cloud revenue. Also, Oracle has a $600 billion backlog and half that backlog is from open AI alone. The dependency of the hyperscalers on anthropic and open AI is just huge and quite scary given that both companies lose billions and are reliant at this point on raising capital for their survival.
In short: Half of the single point of failure: "a lot of this AI trade is predicated on the two frontier models succeeding, OpenAI and Anthropic" — yet there are $1.5 trillion of unfunded commitments tied to the two, and "those companies don't have the income to fund the commitments of capex that they've put out there which the entire finance industry is relying upon." Open-weight models at a tenth to a hundredth of the cost undercut them.
Hayes' concern is concentration risk. "A lot of this AI trade is predicated on the two frontier models succeeding" — OpenAI and Anthropic — and roughly $1.5 trillion of spending commitments across the industry are tied to those two companies continuing to buy. But neither generates anywhere near the income needed to honour those commitments; they're funded by expectations. "Those companies don't have the income to fund the commitments of capex that they've put out there, which the entire finance industry is relying upon."
What makes it fragile is competition from below. Large companies got their first real AI bills and were shocked, which pushed them toward "open weight" models — free or near-free AI systems anyone can run — at a tenth or even a hundredth of the cost. If good-enough AI is that much cheaper, the revenue that was supposed to back $1.5 trillion of commitments never shows up.
31:07But a lot of this AI trade is predicated on the two frontier models succeeding, open AI and anthropic. So if there are models that you can get for onetenth or 1/100th of the cost, they've got 1.5 trillion in unfunded commitments related to those two companies. And then you look at other earnings from all the people holding the stock in anthropic and open AAI.
In short: (Private.) Kate Rooney with a leadership-shake-up alert: Brad Lightcap, one of the longest-tenured employees (joined 2018), former COO, most recently leading special projects and a close confidant of Sam Altman from their Y Combinator days, is leaving to start something new — staying a few weeks. It follows Fiji Simo (head of AGI and CEO of Applications, formerly Instacart CEO) leaving earlier this year due to chronic illness. Her read: power is consolidating under Altman and co-founder Greg Brockman, and the company is bringing in "a more institutional group… from outside of tech, coming from software" (she names Sarah Friar and Denise Dresser as examples of that profile) — "you see this with tech companies that mature and grow ahead of an IPO. Some of the original leadership isn't always the right group to take the company public." No committee stance.
Brad Lightcap — OpenAI's former chief operating officer, one of its longest-serving employees and a close ally of Sam Altman since their Y Combinator days — is leaving to start something new. It follows the earlier departure of Fiji Simo, who ran the applications business.
Kate Rooney's interpretation matters more than the news itself: power is consolidating under Altman and co-founder Greg Brockman, while OpenAI recruits more institutional executives from established software companies. That is a familiar pattern for a start-up preparing to list: "some of the original leadership isn't always the right group to necessarily take the company public and lead a public company."
In short: (Private.) Named in the same imminent-IPO trio (with Anthropic and Databricks) that Terranova says will disintermediate incumbent software. Firestone cites ChatGPT alongside Claude as the everyday AI use driving Taiwan Semi's 45% sales growth. No committee stance.
In short: The other half of the same duopoly framing — the two model leaders Meta is spending to avoid depending on, because a supplier "can determine how you use it, how much you use, what type of guardrails. They can discontinue your use at any time." Cited alongside Anthropic as the frontier bar Meta hasn't cleared ("they don't have a flagship model, they're not state-of-the-art") and as the models Gemini is being ranked below by SemiAnalysis. No investment call.
15:09What is possibly even more important than the phone and the App Stores? AI. Now, there's two companies that are leading the AI race, Anthropic and OpenAI. They're almost like the Apple and Google of AI. And Mark Zuckerberg is looking at this unfold. That these two companies, OpenAI and Anthropic, might end up in the exact same situation that Apple and Google ended up with the phone.
In short: The other frontier lab in the framework — the layer whose unit economics compress when open weights arrive. Enterprises "don't have to pay $10 per million input tokens for anthropic or ChatGPT" any more, creating "an air pocket in demand for the frontier models because they make up such a large bulk," with that demand re-hosted on the hyperscalers instead.
3:15It's not purely due to innovation, architectural improvements. But yeah, we believe that there's still a case to be made for frontier models as well. — Right. So Kimmy is the model Kimmy 3 that was released by Chinese AI startup Moonshot AI recently which is now number three in the world. So it's kind of a fear that this opensource model is going to severely reduce the pricing power of the AI labs, anthropic and open AI.
In short: Named with Meta as one of the first large Helios deployments — the anchor customers whose racks have to translate into the "tens of billions" of Data Center AI revenue AMD has promised for 2027. Implicitly also on the other side of Palantir's "sovereign AI" pitch: the frontier model providers customers are said not to want to become "vassal states" of. (Referenced; not a stance call.)
In short: Same conclusion, same mechanism: in a world where "90% of the companies aren't using the leading edge models and they're switching to using open source models," the frontier labs get routed around for routine work — "Anthropic and OpenAI become more commoditized over time." He also notes not every task needs a frontier model at all: "you don't need them for what's three plus three."
OpenAI, the maker of ChatGPT, gets the identical verdict by the identical mechanism. In a world where "90% of the companies aren't using the leading edge models and they're switching to using open source models," the volume that would justify a frontier lab's valuation goes to somebody cheaper for everything that is not genuinely hard.
The frontier keeps the hard end — "if you want to design a new database, then yes, you need their highest end models for that." But the trivial end, which is most of the requests, drains away to free or near-free alternatives: "you don't need them for what's three plus three."
What makes the drain automatic rather than a customer decision is the routing done by Azure and Google Cloud. Nobody has to actively switch away from OpenAI; the platform simply stops sending it the easy work. That is what commoditization looks like in practice — you keep the prestige jobs and lose the volume, and the margin lives in the volume.
2:33AND IF YOU GO AHEAD AND YOU'RE ON AZURE OR GOOGLE CLOUD PLATFORMS OR, YOU KNOW, WHATEVER, THOSE SERVICES WILL ROUTE WHATEVER YOU'RE TRYING TO DO TO THE BEST MODEL AVAILABLE FOR THE TASK. AND I THINK IN THAT SCENARIO, YES, ANTHROPIC AND OPENAI BECOME MORE COMMODITIZED OVER TIME, AND THE VALUE ACCRUES TO MORE OF THE INFRASTRUCTURE PLAYERS, WHICH INCLUDES THE CLOUD PLATFORMS AND SEMICONDUCTORS.
In short: The loser by construction in a two-winner category: "I think OpenAI is stuck between the two of them. So for me, those are the two winners. I think OpenAI's got the problem because they're jammed between the other two guys." Asked directly whether OpenAI and Anthropic interest him as public-market stocks, he takes only Anthropic.
OpenAI, the maker of ChatGPT, is the name Niles leaves out — and the reason is positional, not a criticism of the technology. Once Google is the consumer winner and Anthropic is the corporate winner, there is no uncontested customer left: "I think OpenAI is stuck between the two of them… they're jammed between the other two guys."
The squeeze is on both flanks. Against Google in consumer it fights a company that owns everything from the chip to the browser and can give AI away inside products people already use. Against Anthropic in corporate it fights a business that has already reached profitability with enterprise customers. Being second-best on both sides of a two-winner market is a worse place than being first in a smaller one.
Worth noting what he does not claim: nothing about OpenAI's models, users or revenue. This is a market-structure argument — a category that supports two winners has already allocated both slots.
1:11I think OpenAI is stuck between the two of them. So for me, those are the two winners. I think OpenAI's got the problem because they're jammed between the other two guys. And that's kind of how I'm thinking about things going forward.
1:29Tim, always love to get your thoughts on the latest from Tesla. Let's start with one of the major concerns. I mean delivering record volume but profitability collapse. So at what point does volume growth stop mattering if say the core auto business is barely generating earnings? — It's good to see you Jenny. Absolutely great point. Uh I we knew the business was not doing as well and especially when you stopped selling the Model S and and and the Model X which were their highest margin vehicles.
In short: Same bucket, plus the systemic link: "so much of the hyperscaler backlogs are from these two companies. For example, of Oracle's 600 plus billion backlog, around half is from OpenAI." If the missing moat starts to bite, "then the entire AI ecosystem could go through a correction phase." Also named in the regulatory-capture charge alongside Anthropic.
OpenAI is the other closed-model provider, and the one that makes the concentration concrete: roughly half of Oracle's $600-billion-plus order book is a commitment from OpenAI. That means a large share of one listed company's future revenue depends on the finances of a private one that most investors cannot examine.
The same moat problem applies — cheap open-source competition, customers switching to control costs, and no established side businesses to cushion it. Which is why Eisman's answer to "what would cause a real sell-off?" is not a Fed decision or an earnings season but "the health of Anthropic and OpenAI."
7:50If the lack of moes begins to cause them problems, then the entire AI ecosystem could go through a correction phase because so much of the hyperscaler backlogs are from these two companies. For example, of Oracle's 600 plus billion backlog, around half is from open AI. On the other hand, AI is allowing the creation of software and other tech that is much cheaper than existing software and tech.
In short: Both Microsoft's biggest partner and its biggest concentration risk. The closed-model archetype (weights locked behind an API, with the provider controlling guardrails, infrastructure and pricing); initially absent from the open-weight letter, it joined after Sam Altman said he wanted the US to lead in both open and closed models — but stayed out of NVIDIA's Open Secure AI Alliance. Commercially it is the largest disclosed component of Microsoft's $678B commercial RPO, and the article's key diagnostic is that RPO grew 25% excluding OpenAI with all sequential growth coming from non-frontier customers — the concentration isn't worsening at the margin. GPT-5.4 is also the escalation target in Microsoft's new cyber router (the hardest ~10% of tasks); and Microsoft excluded a $480M OpenAI investment gain from adjusted EPS. (Referenced; not a stance call.)
OpenAI is simultaneously Microsoft's most important partner and its biggest single risk. It is the archetype of a "closed" AI company: the model lives behind an internet connection it controls, and it sets the guardrails, the hardware and the price. It initially skipped the industry letter defending open models, then signed on once Sam Altman said America should lead in both approaches.
Financially, OpenAI is the largest identified chunk of Microsoft's $678 billion of contracted future revenue — wonderful visibility, uncomfortable concentration. The article's useful check: strip OpenAI out and the rest of that backlog still grew 25%, and every dollar of this quarter's increase came from ordinary corporate customers rather than AI labs. So the dependence is real but no longer deepening. OpenAI's GPT-5.4 also plays a supporting role in Microsoft's new security product, handling only the hardest 10% of tasks that Microsoft's own cheaper model can't. Referenced, not a stance call.
In short: Named twice: as the counterparty in the Nvidia circular deal ("they're going to provide money for OpenAI to buy the chips") and, with Anthropic, as one of the two labs "responsible for a lot of the spending" whose closed models are "much more expensive than the competition… a potential bottleneck."
OpenAI appears twice, both times as plumbing rather than as a company he rates. First it is the other side of the Nvidia deal: Nvidia backs the financing, and OpenAI uses it to buy Nvidia chips — the circularity he calls "incestuous" and "totally convoluted."
Second, with Anthropic, it is one of the two labs "responsible for a lot of the spending that people are doing," running closed models "much more expensive than the competition." Same conclusion as for Anthropic: the cost of the leading closed models is "a potential bottleneck" on how much AI actually gets used at full price.
4:43But look, again, I think one of the bottlenecks is that Anthropic and OpenAI are responsible for a lot of the spending that people are doing. And Anthropic and OpenAI have models that are now much more expensive than the competition. And I think that is a potential bottleneck. — A much better cook because of AI. — And you are.
In short: Two-sided. The bear case: closed-source pricing "five to seven times" Kimi K3, and on Sep-11-2025 it "did not have money… very little revenue" behind $1.4T of commitments. The repair: "$122 billion, the largest fund raise in history… they went into code red… they narrowed their focus to only the things that really matter, which is really compute" — so Oracle gets paid. Still, Eisman doubts the moat, and Luria puts it with Anthropic in the "pulling the ladder" regulatory-capture charge.
OpenAI is the other big closed-model provider, and the episode tells its financial story as a cautionary tale that partly resolved. In September 2025 it committed to roughly $300 billion of computing from Oracle and around $1.1 trillion more elsewhere while having "no capital" and "very little revenue" — which is what eventually took Oracle's stock from 350 to 140.
Since then it has raised $122 billion, described the trillion-plus commitments as flexible rather than fixed, and gone into "code red" — cutting everything except compute. That is enough, Luria argues, for it to pay Oracle, which is his reason for liking Oracle.
What hasn't been fixed is the same moat problem as Anthropic: cheap open-source competition, enterprises switching between models to control costs, and no established side businesses to fall back on. Eisman's follow-up wrap makes the health of these two companies the single thing to watch as a trigger for a broader AI sell-off.
30:43The only thing is by that point as we enter this year Open AI raised $122 billion. The largest fund raise in history. They had the capital. They took that 1.4 trillion and they made it clear that they actually didn't make that many commitments. These are all flexible arrangements. Therefore the actual commitments they have they will be able to pay and they went into code red which is say they narrowed their focus a lot to only the things that really matter which is really compute.
In short: "The whole OpenAI business model I think is coming under tremendous suspicion and questions, and it should, because the capex is ginormous" (~$700B this year toward ~$1T next). On the >$1T IPO: "I don't know about that. I think there's a good chance that IPO gets pulled and if they try to ram it through, it could be quite a fiasco" — and he doubts the ROI is achievable ("really really tough to do, and I think the market's starting to wake up to that").
OpenAI is the private company behind ChatGPT, and Hay's problem is arithmetic rather than technology. Industry-wide AI capital spending is heading toward roughly $700 billion this year and perhaps $1 trillion next. On numbers that large, "how are you going to get an ROI on that?" — a return on investment big enough to justify the outlay. His answer: "really really tough to do, and I think the market's starting to wake up to that."
That matters beyond OpenAI itself, because the cloud giants' reported order books lean heavily on a handful of AI customers. If those customers' business models are shaky, the backlogs everyone is valuing the hyperscalers on are shakier than they look: "are some of these customers like OpenAI really that solid?"
On the widely expected trillion-dollar-plus listing, he is blunt: "there's a good chance that IPO gets pulled, and if they try to ram it through it could be quite a fiasco." (The host notes reports of a slip to 2027.)
43:00it was going to be over a trillion dollar valuation. I don't know about that. I think there's a good chance that IPO gets pulled and if they try to ram it through, it could be quite a fiasco. But just the point that a lot of these backlogs you were asking about, Google, there's anthropic is the big part of their there.
In short: The beneficiary of the $250B Nvidia-guaranteed, 10-gigawatt SoftBank build in Southern Ohio — and the reason Niles stays negative on Microsoft's 27% stake: OpenAI is "caught between Google in consumer AI and Anthropic in enterprise." The financing is what makes it work; the competitive position is what makes it a risk.
OpenAI is the subject of the week's biggest headline — Nvidia may guarantee $250 billion of financing for a giant 10-gigawatt data centre being built for it in southern Ohio. Note what that actually says: the company needs someone else's balance sheet to fund its compute.
The competitive read is the negative one. Dan Niles remains bearish on OpenAI because it is "caught between Google in consumer AI and Anthropic in enterprise" — squeezed on both flanks by rivals with either a bigger consumer distribution or a stronger corporate foothold. That matters to public-market investors mainly through Microsoft, which owns 27% of it.
Full passage: premium transcript (PDF).
In short: Reference, as the source of ORCL's leverage risk: Oracle is "the hyperscaler most levered to OpenAI given its $300bn+ compute/Stargate commitments," underpinning the ≥$90bn FY2027 capex that is inflating ORCL's debt (per Ed Zitron's note). Cited to size the counterparty risk behind the credit call, not a fresh stance on OpenAI.
In short: One of the three big private names at 20–70× revenue vs Google's 8.5× IPO — all the best-case optimism priced in, "doesn't leave a lot of upside." A pass; expected to list later this year / early next.
OpenAI (ChatGPT) is one of three giant private AI companies expected to go public soon. Finucane's objection is valuation: at 20 to 70 times revenue, versus the 8.5 times at which Google went public, every optimistic scenario is already baked into the price.
Since Oxbow doesn't ride momentum and doesn't short stocks, the response is simply to wait — over any five-year stretch a patient buyer usually gets the valuation they want.
22:12And right now valuations like what we see in the private marketplace for these three big companies that have everyone's attention, if they're trading at 20 to 70 times revenue when by comparison Google went public at eight and a half times revenue. So that was a more reasonable, great IPO to try to invest in during the first year that it became public.
In short: At the center of the "circular financing" and litigation — Apple's trade-secrets suit, a coming Microsoft fight; the OpenAI/AMD mega-deal that spiked then unwound Oracle.
38:44— On the Apple front, I thought that lawsuit, Apple suing OpenAI, that's kind of the run-of-the-mill stealing trade secrets. We see it a lot, but do we see so many researchers flipping around between Anthropic and OpenAI and Microsoft? It's been a huge brain drain, right? From some of the traditional like Microsoft.
In short: Named among the three mega private IPOs at "20 to 70 times revenue" vs Google's 8.5× at its 2004 IPO — all optimism priced in. Also cited as a source of the one-time investment gains artificially boosting hyperscaler earnings.
OpenAI (maker of ChatGPT) is one of the three giant private AI companies drawing enormous investor interest. Finucane groups it with SpaceX and Anthropic as trading at "20 to 70 times revenue" — far above the ~8.5× at which Google went public.
He also flags it as a source of accounting distortion: the big tech companies that own stakes in OpenAI book one-time paper gains on those stakes, which flatters their reported earnings even though it isn't real operating profit. His conclusion on the stock itself: pass — all the good news is already in the price.
22:09So that's boosting their earnings artificially, right? Those aren't necessarily operating earnings. But then also you've got all this CapEx spending that's going on right now. And the depreciation cost of all this CapEx is coming in the future. It's not right now, right? So when you really factor in what the ongoing depreciation costs are going to be, yeah, profit margins are actually going to be coming down a fair amount because of that.
In short: (Private.) Named as Anthropic's likely IPO rival — it also filed confidentially with the SEC ~six weeks ago. Sechan: OpenAI and Anthropic have "been raising non-stop" and eventually "need the IPO market" — and once they have public stock it "drives more M&A" (a tailwind he ties back to Goldman). They "may have second thoughts after looking at SpaceX," but he doesn't think they'll wait. No committee stance.
In short: The loss-maker exemplar: US profits are inflated by "massive non-cash gains on investments in loss-generating entities like Open AI," whose "red ink… is astronomical" (with peers "vectoring to go public"). Private; cited to illustrate the earnings-augmentation problem, not an investable idea.
OpenAI is the private company behind ChatGPT. Hay names it as the poster child for the earnings problem: it loses enormous amounts of money ("astronomical" red ink), yet the big companies that hold stakes in it get to report gains as its private valuation climbs. So a business that is deeply unprofitable in cash terms actually boosts other companies' reported profits on paper. Several such loss-makers are heading toward IPOs, which would add a wave of new stock supply. You can't buy OpenAI directly (it's private), so it's here purely to explain why the market's overall earnings look better than the underlying cash reality — the core of Hay's caution.
In short: Still losing money in Q1 even amid the token surge; wants an IPO because it needs lots of cash. With Anthropic it represents ~50% of the hyperscalers' backlogs — at "great risk" as its pricing gets undercut; he expects some such names to go bankrupt.
OpenAI (maker of ChatGPT) was still losing money in the first quarter even during the token-spending boom, and it wants to go public precisely because it constantly needs cash. The bigger risk: OpenAI and Anthropic together account for about half of the hyperscalers' order backlogs. So if cheaper Chinese models keep undercutting their pricing and they run into trouble — Fred thinks some AI companies go bankrupt — the damage feeds straight back into Microsoft, Amazon and the others counting on those orders.
19:07they require lots of cash, yet the hyperscalers' backlogs, they represent 50% of the hyperscalers' backlogs, if they end up with a lot of trouble, if they go bankrupt. And I believe some of these companies will end up going bankrupt. I mean, Oracle has, their debt is two and a half times their sales from last year.
In short: (Private.) News/context (Kate Rooney): Apple's Friday lawsuit alleges OpenAI stole trade secrets after Altman's ~$6.5B acquisition of Jony Ive's startup + a bench of ex-Apple execs to build an AI device; OpenAI says it isn't interested in others' trade secrets. Reignited the Musk–Altman feud (Musk's "scam Altman" jab; Altman's retort). A fresh legal headwind as OpenAI eyes going public. No committee stance.
In short: Fail of the week — a former Apple executive who joined OpenAI "systematically, with full intent… over weeks of time stole incredible documents from Apple," detailed hardware designs; Apple's "thermonuclear" suit aims to jail OpenAI legally and block its hardware. Framework-wise it's also on the losing (premium-model) side of Carlson's tug-of-war. (Plus the childish Musk "scam Altman" feud.)
OpenAI is Carlson's "fail of the week" on two counts. First, Apple's lawsuit: a former Apple executive who joined OpenAI is accused of systematically stealing detailed Apple hardware designs over weeks, and Apple is now trying to trap OpenAI in years of litigation to keep it from shipping devices. Second, in his AI framework OpenAI is on the losing side — it needs its model to stay premium and unique so it can charge for it directly, but Carlson believes the deep-pocketed commoditizers (Meta, Google, Amazon) grind that advantage away over time.
The rest is theater — Elon Musk calling him "scam Altman," Altman firing back about Musk's "space data centers" — which Carlson enjoys but treats as billionaires behaving badly, not investment substance. OpenAI is private, so there's no stock call; the stance reflects his skeptical read of its position.
23:38Now finally we get to the fail of the week which there's a lot of failing going on in this one. We have first of all OpenAI stealing documents from Apple. Now this isn't one of those cases where they took like a couple documents and Apple's just freaking out and overreacting. No. An employee that worked at Apple that moved to OpenAI systematically with full intent and motivation over weeks of time stole incredible documents from Apple.
In short: Same grouping — a headline IPO candidate cited only as an example of the market's appetite for loss-making listings: "Companies like OpenAI, SpaceX, and Anthropic lose money every single month."
In short: Jain: lost ~$39B last year on $13B of revenue — and the loss percentage is increasing; with Anthropic it funds a significant share of the AI capex boom, so if closed-source revenue can't outrun open-source models (now ~two-thirds of token consumption per OpenRouter data, much of it Chinese), the enormous capital-raising treadmill breaks. Ellenbogen: the secondary (behind Anthropic) driver of AI spending.
In short: (Private.) Macro AI-cost color via Boorstin: in her prior-day Sam Altman interview he touted OpenAI's new models being much more efficient — part of an industry-wide "moment" where every AI player (Grok 4.5 too) is focused on cost efficiency as component costs rise. No committee stance.
In short: The concentration that spooked the market — "50% of that backlog was solely from OpenAI" at Oracle; also among the mega-caps' customers as they raced to buy Nvidia chips.
10:12Oracle reported thirdarter numbers in October and showed a massive increase in backlog. The stock went from 230 to 330 in a few days. But then analysts figured out that 50% of that backlog was solely from OpenAI and that made the market nervous and the stock corrected to below where it was when it reported.
In short: (Private.) Named as context: Oracle is "tied to OpenAI" (Bryn's enterprise question mark), and Brown's consumer point — no one wants "a walnut-shaped box that Sam Altman built" in the living room, which is why Apple's device gateway wins. No stance on OpenAI itself.
In short: Wears three hats: marquee customer (a multi-year deal for >$20B of Cerebras compute), a $1B working-capital lender, and a warrant holder — the warrants count as a discount so part of what OpenAI pays never shows up as revenue, meaning the reported figure understates the true OpenAI business and the gap widens as the deal ramps (back-half-2026 loaded). Referenced/neutral.
OpenAI (the maker of ChatGPT) is tangled up with Cerebras in three different ways at once, which makes the numbers tricky to read. First, it's a huge customer — it agreed to buy more than $20 billion of Cerebras computing over several years. Second, it lent Cerebras $1 billion in working capital, so the customer is literally helping fund the supplier. Third, to win the deal Cerebras gave OpenAI warrants — the right to buy Cerebras shares cheaply later.
The accounting wrinkle: those warrants are treated as a discount, so a chunk of what OpenAI actually pays never gets counted as revenue. That means Cerebras's reported sales understate how big the OpenAI relationship really is, and that gap grows as the deal ramps up (mostly in late 2026). Good to understand when judging the headline revenue number — not a recommendation.
In short: (Private.) Per the NYT, leaning toward a 2027 IPO (Altman focused on a $1T+ valuation); pre-IPO pricing/demand meetings not yet held, timing "not decided." Its delay rippled through the AI ecosystem — pressuring Goldman (IPO-cycle leader) and SoftBank (−double digits). Also the on-again copilot vendor for Microsoft.
In short: Passing reference — Codex named alongside Claude Code and GitHub Copilot as competition for Cursor's developer workflow.
In short: Body blow from Microsoft: Copilot moving to usage-based pricing and diversifying to DeepSeek/others. Though <5% of pro-forma revenue, the risk before its ~$1T IPO is that others follow suit; ChatGPT 5.5 "priced itself into a corner" vs cheaper open-source models.
Full passage: premium transcript (PDF).
In short: Watch-list, not impulse-buy — filed confidentially alongside Anthropic; same caution: none of the three mega-IPOs is profitable, and unprofitable issuers historically underperform the most. Put it on the watch list and wait for public-market data.
OpenAI (maker of ChatGPT) has also filed confidentially to go public. It gets the identical treatment in this piece: a likely watch-list name, not a day-one buy. It's huge, unprofitable, and arriving at a euphoric valuation — the exact profile the data says tends to disappoint early buyers.
The practical move the author suggests is to let it list, let the first wave of enthusiasm fade, wait for at least the second earnings report, and only then consider a small starter position anchored to a sensible valuation. Missing the first 20% is fine; overpaying for perfection is the real risk.
In short: IPO up next in the fall. Per a WSJ story it's considering cutting token prices — the commoditization tell: "trillions are being spent for a product with no moats and prices are already being cut." Also the OpenAI investor whose backer (Amazon) reportedly snitched on Anthropic.
OpenAI (maker of ChatGPT) is the other marquee AI IPO due this fall — and Eisman's skeptical tell is pricing. A Wall Street Journal report says OpenAI is considering cutting the prices it charges for "tokens" (the metered units of AI usage). If the industry is spending trillions yet the flagship product already has to discount, that's the signature of a commodity with "no moats," not a durable franchise — exactly what you don't want when the valuation assumes pricing power.
He also ties OpenAI to the Anthropic episode: Amazon, an OpenAI investor, is the party that reportedly tipped off the government about Anthropic's jailbreak.
9:05No moats. Trillions are being spent for what looks increasingly like a commodity. Number three, and speaking of the commoditization of AI, last week an article appeared in the Wall Street Journal stating that OpenAI is considering lowering the prices it charges customers. The company is considering cutting what it charges for tokens. This is pretty astonishing news.
In short: Named as one of the external AI suppliers (with Gemini / Anthropic) Meta refuses to "rent" its tools from — staying full-stack to keep pricing power.
22:59Zuckerberg describes it as a company that builds new applications, new experiences, new features. To be able to build all of this stuff, you need to have the best tools possible. If Meta is reliant on renting those tools from Gemini, renting them from OpenAI or Anthropic, then now they're no longer full stack. They're beholden to someone else.
In short: The other pure-play frontier owner — filed confidentially to IPO, could list above $1T. Same structural exposure as Anthropic: the model is the whole business, so a regulatory shutdown has no other revenue to absorb it. (Free GPT-5.5 cited as a model that can find the same minor flaws Anthropic's was faulted for.)
OpenAI (maker of ChatGPT) is the other giant pure-play AI lab, and it has filed confidentially to go public — potentially worth over $1 trillion. The article groups it with Anthropic because it carries the exact same structural risk: the model is the whole company, so a government recall or export restriction would hit it with no other revenue to fall back on.
The piece's bottom line is that for investors weighing these labs, the old question ("who has the best model?") matters less than a new one: "who can keep their model online?" Owners with broad businesses (Google) or renters (Apple) are more insulated than a standalone lab like OpenAI.
In short: Cited alongside Anthropic as additional equity that "needs to be sapped up" — capital has to come from somewhere, a risk the melt-up may "bulldoze right over" until it can't.
7:24We've never seen so much equity ownership. Feels like kind of like a housing crisis again when your barber's talking to you about Nvidia call options. There's a problem generally. So, but it's not a hollow market. Earnings are real. And I think we can bring it all back to this whole AI thing. Is it real? And it is.
In short: "We've also obviously got Anthropic coming up and OpenAI coming up" — more late-life-cycle, maximum-valuation supply headed for the passive indexes in what he calls "literally the largest IPO cycle we've ever had."
Same argument as Anthropic. OpenAI is part of what he calls "literally the largest IPO cycle we've ever had" — companies coming public "much later at much higher valuations," with the gains already captured privately and the remaining risk sold into index funds.
11:32How does all this play out, do you think? — Well, the good news is there's a lot of millionaires that have been created and I love the stories of the welder that's now a millionaire at SpaceX. I mean, there's wonderful stories. But it's similar to Facebook when it came public.
In short: "Could be the most overvalued of the AI names": banks refused SoftBank a $6B margin loan against its OpenAI stock (marked ~$800B) — the first crack that may cap what OpenAI can raise in its IPO.
Full passage: premium transcript (PDF).
In short: Named as one of DocuSign's new IAM integration partners for its agentic agreement workflows.
In short: Same as Anthropic — named as upcoming overvalued IPO supply absorbing market liquidity, "not investing."
14:32Again, just more of the strip mining and stealing and taking advantage of the average person. I'm totally against it, but nobody cares what I think. So, hey, if you did well, Allah be with you. Take the money and run, I guess. But this isn't investing, make no mistake. So, this is going to continue because we have other things coming to market too. Anthropic at some point, OpenAI, other things, these things are all overvalued relative to historical norm. And again, I don't
In short: Part of the issuance wave (SpaceX + OpenAI + Anthropic) that may be sapping market liquidity. Won't show its books to anybody — asked for a private-placement loan, told a willing lender "we're not showing you our numbers," which "might possibly mean they are dressing up their numbers" to propel a higher IPO price.
OpenAI shows up as a warning, not a pick. It's part of a wave of huge stock sales (with SpaceX and Anthropic) all soaking up the same pool of investor cash. More damning: when a lender asked to see OpenAI's actual financials before making a loan, the answer was effectively "no." To Gundlach that secrecy suggests the numbers may be "dressed up" to justify a richer IPO price — exactly the kind of opacity that thrives in private markets where outsiders can't check the books.
42:55So they're obfuscating which is always the case when you have private markets. Private markets they attract people who like to obfuscate because by being private people can't really look at what's happening. Two companies OpenAI and Anthropic they do not show their books to anybody. One of them wanted to get a loan recently, a private placement loan to get some capital and a potentially willing lender said, "Let me take a look at your numbers.
In short: Filed a confidential S1 (~$100B expected; last round $122B raised at an $852B valuation). The commoditization tell: a WSJ report says it's considering cutting token prices — "trillions spent for a product with no moats and prices already being cut." Also sued by Florida over ChatGPT addiction.
OpenAI (maker of ChatGPT) filed paperwork to go public, likely raising around $100 billion. Eisman's skeptical tell: a Wall Street Journal report says OpenAI is considering cutting the prices it charges for "tokens" (the metered units of AI usage). Trillions are being spent industry-wide, yet the flagship product is a near-commodity ("no moats") already discounting — exactly what you don't want to see if the valuation assumes pricing power.
He also notes Florida is suing OpenAI claiming ChatGPT fosters addiction — tying it to his broader "addiction business model" theme.
10:16Trillions are being spent for what looks increasingly like a commodity. China is highly competitive as well. Something for equity holders to think about while being asked to fund future growth. And speaking of the commoditization of AI, on Thursday, an article appeared in the Wall Street Journal stating that OpenAI is considering lowering the prices it charges customers.
In short: SoftBank is "overpaying for OpenAI shares"; it sits at the center of the circular graphic (investing in Oracle while Oracle funds its data centers), and its ~$60B IPO (~Sept; 88% announce-odds) drains the same liquidity pool as SpaceX.
8:32now in terms of size and valuation, the share price has been indicated and not confirmed at $135 per share, which was following a five for one private stock split executed in May. the raise is going to be roughly 555.6 million shares to raise about $75 billion, the biggest IPO of all time. you will see open AAI is estimated to do another 60 billion IPO and so is Enthropic.
In short: Also coming public into the same window — the smartest people in AI are "selling their equity to bag holders as fast as they can" (Erik's framing, which Larry endorses).
OpenAI is the third mega-listing in the wave. The tell, in Erik and Larry's shared framing: the smartest people in AI — the very founders who built it — are all choosing this exact moment to sell their private equity to public-market "bag holders," just as tech insiders did in 1999–2000. When the smart money sells in size, take the hint.
8:08The actual raise which is the money that we need to come up with someplace to pay for the shares being offered is 80 billion. Okay, add that 80 billion. Google's — and add to that Anthropic is coming up. We're going to have OpenAI coming up. There's about 200, 250 billion of immediate raise. But here's the thing that I'm actually focusing more on, Larry, is 6 to 12 months after that, all the insiders and the VCs and the early investors in those companies, it's not 200 billion.
In short: Considering "drastic price cuts" anticipating a war for users with Anthropic (per the WSJ) — which threatens its own IPO chance; the flagship of the circular flywheel that "works splendidly until VC flips from buyer to exiter and someone asks for their money back."
OpenAI (ChatGPT) is also in the IPO queue, and per the Wall Street Journal it is weighing drastic price cuts to win users before Anthropic does. To Paulo a price war is the tell that the exponential-growth story is over: companies with pricing power don't slash prices. Combine that with the circular money flow — VC cash raised by the AI labs, paid out to the chip-and-hardware suppliers, and relaundered back into the labs — and the whole flywheel "works splendidly until VC flips from buyer to exiter and someone asks for their money back."
In short: Won the consumer; now improving in enterprise/coding with accelerating growth — one of the two or three leaders likely to hold position.
OpenAI, the private maker of ChatGPT, is the second of his three model-layer winners. It already "won the consumer" — ordinary people's default AI app — and is now getting better in the enterprise and in coding, where growth is accelerating.
His broader point is that the leaders in a technology race tend to keep leading ("the leader goes bigger, faster, and wins"), so OpenAI is likely to hold its spot as one of the two or three survivors.
38:16you know, Open AI has, you know, they they were focused on so many different other sectors, but they're starting to do better in enterprise and their coding tools good and they're starting to see accelerating growth on that side. And then look, the consumer franchise, it it looks like enterprise right now is much better because you're, you know, you and I, we're willing to pay a lot because it's replacing human beings.
In short: Private — named (with SpaceX and Anthropic) as the IPO pipeline a rules-based Bloomberg index would admit faster than the S&P committee; investment merits "aside."
59:49Now this is a competitor in theory to the S&P 500. But instead of there being this committee that governs the S&P 500, which is shown to be fairly arbitrary at times, Bloomberg is fully rules-based. — Mhm. — And one of the inclusion criteria is that it allows these IPOs to be admitted much more quickly. And so investment merits of investing in SpaceX and Tropic, um, OpenAI, any of this pipeline of IPOs aside, exchanges are trading venues.
In short: Cited in the same IPO-liquidity-drain breath (SpaceX/Anthropic/OpenAI) — "the beginning of the end?" — a dynamic he'd welcome for cheaper entry points.
37:32I wonder what these IPOs — SpaceX, Anthropic, OpenAI — will do to liquidity, sucking capital from other names. The beginning of the end? I certainly hope so. There are a lot of names in conventional financial services and natural resources I'd like to own much more of — I'd be delighted to see Exxon Mobil, Agnico Eagle or Franco-Nevada fall by half. The only way I've found to become profoundly materially richer is to buy undervalued assets and wait until they return to value. So my hope is that higher oil prices and IPO-driven illiquidity lead to materially worse equity markets, particularly in financial services and natural resources — the markets I know best.
In short: Sam Altman is "admitting AI costs are becoming a huge issue" as overspending becomes a meme; ChatGPT is the commoditized model users "switch constantly" between. Trigger for the AI-cost-panic narrative.
4:48But I also think that SpaceX um it's $75 billion is a lot to ask the public to fund. So I think it is a big lift in terms of how you're going to fund that purchase. And so I think now when the stocks wobbled a little bit, I think people are raising cash. — While we can point blame at the Broadcom report or the jobs report or investors raising cash for SpaceX, there's also the fact that we had this news from Sam Alman admitting that AI costs are becoming a huge issue.
In short: Eisman floats the bear theory (a "shell game," the "weak sister" whose IPO opens the kimono); Rasgon doesn't take the bait — his counter is monetization evidence and use cases. Also one of AMD's two multi-gigawatt GPU deals (with warrants).
47:21That's one theory. You know, another theory is that Open AI is kind of a shell game and that the guy who runs it a liar and and it's going to go public and it's gonna and it's not going to be great. And then — when they go public, you have to open up the kimono. — They got to open the kimono. — Um and and uh they're the kind of the weak sister of the whole story and but they're a big percentage of the whole industry.
In short: Named as a Broadcom XPU customer — 1.3 GW in 2027, part of a 10 GW deal through 2029 — anchoring the AI-bookings backlog behind Broadcom's "visibility to 2028."
In short: Accessible frontier capabilities "plateauing" is the biggest problem facing OpenAI (and Anthropic) ahead of an IPO later this year; its Codex is among the tools catching up to Claude Code.
OpenAI is the private maker of ChatGPT and the Codex coding tool, also reportedly headed for an IPO this year. Woo's concern mirrors Anthropic's: the most useful, broadly available AI capabilities are leveling off, which is the central challenge for these companies as they try to justify huge valuations to public investors.
9:06I see this as being the biggest problem facing Anthropic and ChatGPT ahead of their IPO later this year. For Anthropic, there's another problem. That is competition is closing in. Cursor, GitHub Copilot, OpenAI Codex are all quickly catching up to Claude Code. This is one reason why Microsoft reportedly has canceled most of its Claude Code licenses.
In short: Same overpriced AI-IPO rush; investors are selling other parts of the market "to make room" for these listings.
OpenAI is the private maker of ChatGPT, another name in the overpriced AI-IPO rush. He notes investors are selling other parts of the market "to make room" for these listings — a forced reshuffling he sees as a sign the market is dangerously top-heavy in expensive tech.
33:02And so the AI thing and the space war and the anthropic and open AI, these companies are are just coming into the market at very expensive levels and they're being really picked off by China. — Any other places like Brazil or anything that's catching your eye as far as resource driven economies? I realize there's a lot of you know you know elections coming up um you know in some of these places so it's hard to really game that but what are you looking at there as well kind of resourceoriented emerging markets — well we we did this with Argentina so
In short: The AI-revenue benchmark App Economy uses to size xAI's lag — ChatGPT ~50M paying subs vs SuperGrok's 1.9M; OpenAI ~$24–25B annualized vs Anthropic's ~$30B. Frames why the AI segment is a $200–500B placeholder, not a proven cash engine.
In short: He'd "take the over" on OpenAI + Anthropic combined hitting $200B of revenue in maybe 12–18 months — code generation turned out to be the killer app to monetize AI.
12:20Yeah. It's fascinating. As a monopolist provider, they're rate limiting supply at some fundamental level. They dismissed Sam Altman as a podcast bro after they met with him. — [laughter] — You and we were talking about this backstage, which is you would take the over on OpenAI and Anthropic combined at 200 billion of revenue maybe 12 months, 18 months.
In short: One of the two private AI labs at the center of the loop: funded by hyperscalers at successively higher rounds, then booking enormous cloud commitments back to them. The circularity only reverses once it "becomes a public entity where the stock price at quarter end is the one true mark" — which forces it to (A) generate cash and (B) never draw down, or the hyperscaler Other Income machine runs in reverse.
OpenAI plays the same role as Anthropic on the other side of the board — funded by hyperscalers at ever-rising valuations, then handing much of that money back as commitments to buy their cloud computing. Paulo's key point is what happens when it goes public: right now its value is whatever the insiders say it is, but a public listing forces a real, unfudgeable price every quarter. That's the pin near the balloon — the company would suddenly need to (a) actually make money and (b) never let its stock fall, or the entire mark-up-your-own-investment machine unwinds.
In short: The other owned model-layer winner — "between Anthropic and OpenAI, towards year end you're going to be looking at $200 billion of revenue," with enormous incremental margin because they locked up compute early.
OpenAI, the private maker of ChatGPT, is the other model-layer name Whale Rock owns. Sacerdote lumps it with Anthropic: together he thinks they'll be doing about $200 billion of revenue by year-end, at unusually fat margins because both secured their computing supply ahead of the crowd.
The takeaway is that the two leaders in the AI-model race are turning into highly profitable franchises, not cash-burning science projects.
8:18— I think they liked your thesis of the oligopoly of LLMs.
8:41You can break it down any way you want to. You can make a guess. What's more interesting about it is the margin profile of these companies. Because they've been able to lock up compute — and they were some of the first, and they have it already locked in for the next several years — this is going to be enormous incremental margin.
In short: The other half of the IPO race; same "shell game" of dumping richly-valued equity into index funds.
OpenAI (maker of ChatGPT) is the other half of that IPO race. He treats it identically to Anthropic.
It's part of what he calls a "shell game": offloading enormous, expensive equity into index funds before any downturn — a setup he views as bearish for the crowded tech market that has to absorb it.
5:14for each of those companies. Do these historically massive IPOs plus the AI hyperscaler spending create a scenario that in effect maybe insulates the markets potentially from a collapse until we at least get through these issues? — one thing about liquidity and the Trump team, and I'll get into the whole everything you laid out is a very well laid out.
In short: Named once, as Novo Nordisk's counterparty: a "strategic partnership with OpenAI: enhancing drug discovery", listed among the developments supporting the STRONG BUY rating. No view on OpenAI itself.
In short: A WSJ story said OpenAI missed its revenue targets and that its CFO is nervous it can't meet its massive data-center commitments — the single news item that triggered a one-day correction in semis, infotech and the Nasdaq this week.
9:59On Tuesday of this week, there was an article in the Wall Street Journal that Open AI missed its revenue targets. It was also reported that Open AI's CFO is nervous that Open AI will not be able to meet its massive data center commitments. That one news story caused a one-day correction in the semi subsector, the infotech sector, and Nasdaq.
In short: Private; named as an embedded part of the Microsoft position rather than as a company with a view: "Please note that by investing in Microsoft, you are also investing ChatGPT. Microsoft currently owns 27% of OpenAI." It is also half of the risk on the other side of the ledger — "while Microsoft has made major bets on AI through its OpenAI partnership, the pace of innovation means it risks being disrupted." No valuation, no comment on OpenAI's own economics.
In short: At the center of the turn: The Information's leak of Altman's "rough vibes / economic headwinds" memo after Gemini 3's leapfrog; ~$500bn valuation but projected >$100bn cash burn while inference costs run ahead of revenue. "No culture" vs Google's institutional memory; key-man risk.
Nothing matches this filter.
Verbatim excerpts from the public transcripts (auto-pulled at each mention's timestamp, lightly cleaned). Timestamps link into the video; "source page" opens that commentator's full analysis at this row. Click a mention's header line to expand it (one open at a time). For personal study — not investment advice.