Porter Collins says Wall Street research puts 70% of the hyperscalers' AI revenue with two customers, Anthropic and OpenAI, which on his math is a quarter to a third of all the cloud revenue those companies report.
The usual read of Nvidia's revenue doubling is that the build-out is broad. Collins and Vincent Daniel read the same set of numbers as concentration, and they read the financing behind it — companies that used to buy back stock now issuing debt and equity — as the part that should worry people.
"So I think you're in a different paradigm here. And yes, it scares me, right?"
Daniel and Collins run Seawolf Capital together. They were Steve Eisman's partners at FrontPoint Partners, where the three of them made the subprime short that "The Big Short" was written about, and this is the first time the three have sat down together on the show since April of last year.
The full episode is covered here so you can skip it. 53 minutes of audio, 23 minutes of reading.
Here are the 12 takeaways that matter.
👤 Guests: Vincent Daniel and Porter Collins, co-founders of Seawolf Capital and former FrontPoint Partners colleagues of the host
🎙️ Host: Steve Eisman, the investor "The Big Short" was written around, previously a Senior Portfolio Manager at Neuberger Berman
📰 Published: 14 September 2026 on YouTube (The Real Eisman Playbook)
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 53 min | ✅ Time saved: 30 min
Key Takeaways
The Treasury is buying long-dated bonds because every other lever is politically shut
Collins: raising taxes, cutting entitlements and cutting interest are all off the table, so printing is what is left
Wall Street research puts 70% of hyperscaler AI revenue with Anthropic and OpenAI alone
Collins says that works out at roughly 25% to 35% of those companies' total cloud revenue
Nvidia's top five direct customers were 70% of its accounts receivable in the quarter
Eisman found it in note 7 of the 10-Q, against revenue that was up more than 100%
OpenAI's costs grew three times faster than its revenue last quarter, in dollars
Eisman's figures: revenue up about a billion, costs up about three
The thing that kills a bull market is new supply of stock, not bad news
Collins expects Anthropic and OpenAI to be forced public for exactly that reason
The multi-manager pods have made shorting a different and much harder business
They are levered about five to one and will cover a losing short within three weeks
A pod is happy with a 3% to 5% spread; Seawolf is trying to make 150% on a short
Carvana books most of its pre-tax income on gain on sale, and somebody is paying over 110 for the paper
Ally pays 102 to 104; the two-year average gain on sale has been 109 to 110
Seawolf's two disclosed longs are a floating LNG operator and a California cannabis grower
Gold is where most of Seawolf's capital sits, and the reason is fiscal rather than monetary
1. Bessent Is Out of Arrows
Eisman opened by asking Collins what to make of Treasury Secretary Scott Bessent and the Treasury Department buying long-term Treasuries.
Collins called it his old subject and gave a blunt verdict. "I think they're screwed." The Treasury is reaching for extraordinary measures, he said, because it has run out of ordinary ones and is trying to hold a lid on yields while inflation is still high.
His arithmetic for why nothing works: "And at this point, you know, I see it as interest expense plus entitlements is, call it 10% of tax receipts." "There's nothing you can do."
He then walked the options and closed each one. Interest cannot be cut. Entitlements could be cut in theory, but nobody is proposing it — "We've been talking about this for how long, Steve? And they haven't done anything long time." Taxes are not going up either.
What is left is the printing press, and that is why he owns metal: "I think I've had as much conviction in this precious metals trade as I've had in anything in a long, long time."
2. Why the Trade Is Gold
Eisman asked him to explain the link for viewers who do not find it obvious, because the connection between interest rates and precious metals is not intuitive.
Collins started with what a central bank owns. A government bond is someone else's liability; "And so gold is an asset." That, he said, is why central banks around the world, China included, keep buying more of it.
He thinks the crossover point is close or already passed — gold against Treasuries as a share of central bank balance sheets, depending on where the gold price is.
The mechanism he described is currency debasement rather than a gold story: gold "just sits there," everything else falls against it, and the dollar has fallen a long way over the last hundred years. "And I think that trend continues and probably, you know, accelerates at this point because they're out of arrows."
He would not put a timeframe on it. "But I think, you know, if I look back in five years, I think it's going to be a lot higher then."
3. In Bessent's Shoes
Eisman asked the two of them to argue the problem from the Treasury Secretary's side of the desk before proposing a fix. Daniel took it.
"So the problem is that right now I have as Porter said expenses that are higher than receipts." He put the debt at $40 trillion and said, "That's one of the issues to the point where 40 trillion times an interest rate is a big number and it's really eating into all the other things the government can do and going higher."
The first move in his version is not financial at all. "So the first thing I would do is go up to my boss which is President Trump and say respectfully Mr. President can you end this war because if you can end the war maybe we could get oil down from say 86 to 60." "Inflation expectations will come crashing down."
The president in his scenario says no, and so do entitlement cuts and tax rises. "So then you start with theatricality and deception by playing twist games" — the sort of thing, he said, the three of them would have laughed at across the office.
On the size of the buyback: "Well he said at least 4 billion which means it's infinite."
The bet he thinks is being made is a timing bet: "So I think he is going to keep trying to get rates down because that's what he needs to do over the next six months to nine months." Inflation expectations need time to come down, and growth has to slow a little but not a lot.
On the war itself: "And the war was just a massive strategic error from, you know, a interest rate and debt perspective." The expectation, he said, had been that it would be over in about two weeks.
4. Eisman Is Less Worried
Eisman then put his own position on the record, and it is not theirs.
"So, you know, I'm not nearly as hung up on the deficit as you guys," he said, describing a routine he does about people who talk about the deficit for twenty-five years. He does not think doom is imminent, though he thinks the direction is wrong.
His objection is one of degree: rates 50 basis points higher is not, to him, the worst thing in the world.
Where he agrees is on tactics: "I think the mistake that Bessent made is basically putting a line in the sand. Now he has to defend it."
Collins pressed him for a concession and got a partial one. Eisman wanted it on the record, and Collins allowed that he is "not saying it as passionately" as he once did.
5. The Calm on the Screen
Collins made the case that the placid market screen is manufactured, and Eisman said he is more nervous than he has been.
The picture he described is the one a viewer gets every morning — benign rates, Nvidia up, Apple up — and the work he says goes into producing it, which he called incredible.
"And this is the first time that people are starting to see what we see." "It was just so out in the open that it couldn't be hidden anymore."
Collins then explained the Fed's side for the audience: since about 2010 the Fed has run quantitative easing repeatedly, buying Treasuries to push rates down, which in his view never helped the economy but did lift the stock market. "So Warsh as a new head of the Fed has basically said I don't want to play that game anymore."
The catch, in his telling, is that the balance sheet is still growing. "So and so he's stepped aside and they're still doing QE every day." "They're printing money every day."
His theory of the double act: "they're playing a great good cop bad cop." The Fed chair steps back from a tool and the Treasury Secretary picks it up.
Daniel added why that matters for positioning. "I think the majority of the market has been conditioned to look at the Fed as dovish or hawkish and that's their signal" — and nobody has built the same reflex for the Treasury, which is why what it is doing reads as wide open.
6. The Budget Nobody Cuts
The conversation turned to whether anything forces a fiscal correction, and all three agreed on the answer.
Collins said the only thing that produces action is getting close to the edge, and Eisman finished the sentence for him.
Daniel widened the target beyond the usual one. It is not just entitlements, he said, it is defense, and it is the whole budget.
Eisman's version of a realistic goal is not a cut at all — slow the growth rate of federal spending by half — and Collins' answer was that even that needs a crisis: "So, it has to get bad."
On the efficiency drive that was supposed to do it, Daniel was brief. "Doge lasted a month." "And what they cut was nothing."
Collins made the point that fraud is not the explanation. There probably is fraud in Medicare and Social Security, he said, but the spending itself is legitimate and large. "It's just they spend a lot of money."
7. Nvidia's Top Five Customers
Eisman brought in what Ed Zitron, a guest on an earlier episode, had shown him in Nvidia's filings, and asked the two for their read on AI.
The number came out of the footnotes rather than the press release. "So note 7 says that the top five direct customers of Nvidia accounted for 70% of accounts receivable in the quarter," Eisman said, describing it as the kind of work the three of them used to do on financial stocks.
"So on the one hand your revenue is up over 100%. On the other hand it's totally concentrated in a handful of names and a couple of those handful of names are not looking so good."
Daniel answered with a live example from a friend at a public company he would identify only as "Mr. X". "Mr. X told me that his company's queries are up, I'm forgetting the percentage, 30, 40%." "But his costs are down something like 60%."
The mechanism is routing. The company centralized its AI queries and sorts them by how much model they need. "So in other words, they're very careful now about what they send to Anthropic and OpenAI" — the rest goes to cheaper open-weight models, including Chinese ones.
Daniel's question from that is the one that decides the trade: "How do we get to the returns on invested capital of all the money that we're spending on artificial intelligence?" Funding can keep coming, he said; the pattern he expects is the ordinary one of a capital-spending boom, then a bust, then good businesses afterward.
The show's own earlier reporting was cited alongside it: Torsten Slok, a previous guest, had said GDP would grow about 2% this year and at least half of that is AI capital spending. Eisman's follow-up was that if OpenAI failed, "the economy would go into recession almost immediately."
8. New Supply Kills Markets
Collins would not make the recession call, but he described the change in the hyperscalers' balance sheets that has him positioned defensively.
"No, you know I'm not making the call." "This is not our expertise, right?"
The change he pointed to is in how the spending is funded. Companies that generated enormous free cash flow and bought back their own stock are now issuing debt, and in some cases equity. "Google issues $85 billion in equity, right?" — a sale, he noted, that was not a distressed one.
That is why he expects the private AI companies to be pushed into the public market. They need the money, and Anthropic will be allowed to come public because it has to be.
His objection is not to the companies but to the supply: "My contention always is that just like SpaceX, it's bad for the market because it's supply, right?" "And that's what usually has killed markets, right?" "That's what killed the market was new supply of stock," he said, naming 2000 and 1929.
The demand side he described as mechanical and thoughtless. "They talk about the S&P. It's the dumbest investment philosophy ever, right?" "If I get more money in, I buy. If I get money out, I sell. There's no other thought to anything, right?"
The digression that followed was about SpaceX's filing. Daniel raised asteroid mining; Collins confirmed it is in the S-1 as a future activity, and said he had noted on his own weekly podcast that Apple has a science-fiction show, For All Mankind, in which asteroid mining is a major theme. On Musk's forecasting record: "You know, full self-driving is about a decade behind and it's at level three." "So, you know, he makes a lot of good comments that don't quite come true."
9. OpenAI Against Anthropic
Coming out of the break, Collins gave the research number that frames the rest of the section, and Eisman supplied the two companies' quarterly figures.
"So my thought about the whole AI thing, which is what makes me nervous right now, is that there were some very good reports by some Wall Street firms that basically said that 70% of hyperscaler AI revenue is from just Anthropic and OpenAI," Collins said, relaying the research rather than his own work.
Run through to the top line, he said, that equates to about 25% to 35% of the hyperscalers' total cloud revenue.
"The problem is that the OpenAI numbers that have just come out are actually quite poor."
Eisman gave the comparison. "So, Anthropic put out like 11 and a half billion in revenue for the June quarter. They didn't say anything about costs and that was up over 100% in 3 months."
"OpenAI was at 6.5 billion and it was up only 18% in 3 months. Its costs were 12 billion." Set aside the percentages, he said, and look at the dollar change over three months: revenue up a billion, costs up three.
Collins' point is that an unprofitable company cannot absorb a bad story. A profitable company shrugs off a bad newspaper article and buys back stock. "But when you're OpenAI and you're bleeding cash every single day and all of a sudden the narrative has really started to change" — and people are leaving — "Your cost of capital is rising."
He made the joke that whoever buys OpenAI out of bankruptcy will probably get a fantastic deal, and said that is the history of it: the railroads, Wachovia bought for nothing by Wells Fargo, Bear Stearns.
10. Why Shorting Got Harder
Eisman asked Collins about an article he had written on shorting in Seawolf's newsletter. The answer began with Sean Connery and ended with an explanation of how the multi-manager funds work.
Collins built the piece around The Untouchables — an actor who outshines everyone in every film, and a character who knows better than to go in and goes in anyway. "Fighting crazy is hard."
Daniel's structural point is that the short side is now dominated by firms running a completely different return threshold. "Most of the people in the investing world like you're short, they're long."
He explained the pod model at Citadel, Millennium and Point72 for the audience: the parent raises the money, the grand portfolio manager hands a team an allocation — Daniel's illustration was Ken Griffin handing someone 250 million in capital and telling him to go do something with it — and the team runs long and short inside its sector.
The leverage comes from the low volatility that hedging produces: get annualized volatility below a certain level and the firm will lever the book about five to one, so $25 billion of capital runs $125 billion of positions
The book is matched beta-adjusted and factor-neutral, swept constantly, so a manager cannot be long growth and short value
The example of what happens otherwise was Leopold Aschenbrenner, who was long AI and short software: "And Leopold obviously wasn't as sophisticated as these big guys. And they knew it and they blew him up." Eisman: "They blew him up and Kenny G cleaned his clock."
The return thresholds are what make the difference. A pod long $100 million of JPMorgan against $100 million of Bank of America is done at a 3% to 5% spread: "If they made a spread between the two of three to 5%, they're happy and they're done." "When we're shorting stocks, we want to make like 150% or whatever."
The consequence for a fundamental short seller is the squeeze. "They know like every short that we would come up with, right? They're probably short as well in size. And if it doesn't work for them in 3 weeks, they're covering." "What we used to do for a living is brutal, extremely more difficult."
Eisman offered his own version of the factor risk. "So you're long every auto and auto parts company and you're short every horse carriage company and it's working great" in 1905 — until one bad car accident on Fifth Avenue makes the front page, sentiment reverses for two months, and a book levered four or five to one is taken out.
On Tesla, which they have been short on and off for the better part of five or six years: Daniel said the earnings have gone one way, down, and the stock the other. Collins drew the lesson. "I think the lesson is don't short a cult." Daniel: "We're just doing it with significantly less capital than we used to."
11. The Shorts: FICO, Carvana
Asked for a short he is willing to name, Daniel started with one he does not own and then gave two he does.
"Circle doesn't make any sense to me. Zero," he said — "And yet the stock has rallied enormously." Eisman put the rally down to crypto legislation.
Seawolf stays close to home. "We short a lot of stuff in our neck of the woods," Daniel said, meaning financials and fintech.
The FICO case is a pricing case. Daniel said that if the company had raised prices over five years at something like inflation plus 2%, the way Visa and Mastercard do, nobody would have noticed — and that instead it raised prices 1,600%. "It was pigish."
Eisman added the mortgage detail: "And not only that, they're pulling credit files and mortgages and charging I think it's 3 to 5x what you're doing on credit card and auto loans."
His view is that the large buyers — he named Fannie Mae and Rocket — have the leverage to demand a lower price and have not used it, because everyone treats the fee as a small piece of the loan
On Carvana, the short is about where the profit comes from: "And the majority of their profits, almost all of their profits, a good portion of their pre-tax income between 75 and 100% from gain on sale."
Carvana sells used cars and finances them, and a good portion of those loans are subprime
It securitizes the loans and sells the residual, and also sells whole loans to third parties
The arithmetic of the short is a price nobody can account for. Ally, one of the biggest buyers, pays somewhere between 102 and 104 — "That means if I originated a loan for $100, I get to sell it for $102 or $104." "Their average gain on sale over the last two years has been 109 110." Work backwards from the blend, Eisman said, and someone in the world is paying north of 110.
"So there's a mystery buyer out there and no one knows who the mystery buyer is."
Daniel said the company denied about eight years ago that it was Mark Walter, the investor now attempting to sell the Lakers and the Dodgers, and that Walter is the fifth-largest holder of the stock
"And allegedly Delaware Life owns a heck of a lot of Carvana paper" — an insurance company connected to Walter, whose disclosed related-party transactions, Daniel said, rose from about 3% to something like 30% or 40%
Walter has addressed it in public. "And also listen Mark Walter came out and said publicly there is no fraud here." "There is no fraud here and the market believes it." Eisman: "So I got squeezed."
The stock has done "a lot of nothing" for the past year, Eisman said, and trades at 30 or 40 times earnings, which he thinks is a lot for a used car company. Both agreed the thesis cannot be proven — "proving that is going to be impossible," and it has been impossible for eight years. "So you need a moment like that to really get paid" on a short, Daniel said: the moment the truth arrives.
12. Two Weird Longs
Eisman asked for two names from the long book. Seawolf owns about ten of what Collins called weird names, plus a lot of gold, and each of them took one.
Collins took Golar LNG. The company used to be a shipper and now runs four vessels that pull up to a country and liquefy its gas offshore. "And what these boats do is they pull up to a country like Argentina and convert the natural gas into liquid natural gas."
The alternative is an onshore plant of the kind Cheniere built in Louisiana; Collins said YPF in Argentina is going to spend 30 to 50 billion dollars building one
"But in the meantime, they're buying these boats and, you know, they're going to take a toll. But the capex is minimal." He put it at two and a half to three and a half billion
"And so it's basically a 20-year contract, a toll road."
The name is a way into an Argentina view. "And one of the things that we've been bullish on is Argentina." Collins said the country has balanced its budget under Milei, and that its gas has been trapped — findable but not transportable. "And the big thing for them is they're going to move these Golar ships in, two of them, and pull the gas out starting next year." He expects rates, already down from over 100% to about 30%, to fall a lot further. On the country's pull for investors: Peter Thiel has moved there and "he plays chess with the president."
Daniel took Glass House Brands, a California cannabis grower in the vegetable, tomato and blueberry belt. "This just happens to be cannabis that they produce rather than fruits and vegetable." The disclosure came with it: "Full disclosure, the last time that I smoked weed, May 1984." None of the three uses the product.
The operating case is cost. On a pro-forma basis Daniel said they can produce at roughly 90 to 100 dollars a pound, which he said is not by itself enough of a reason to own it
The catalyst is regulatory: the federal government rescheduled medical cannabis from schedule one — the heroin and fentanyl category — to schedule three, which removes a large part of the restriction
The trade is the price gap. Daniel said California sells at about $200 a pound while Germany is six or seven hundred, which on higher revenue would mean 50% to 65% EBITDA margins
On timing: "That's the $64,000 question." He called it a 2027 event for interstate or overseas progress, and said they have already sold hemp overseas this quarter
Seawolf's stated habit is to stay out of the mainstream names. It has owned Google for years and calls that the only household name it holds; the exceptions come when something is being thrown away, which is how it bought ServiceNow cheaply in the software sell-off and bought Google when the market decided search was being disrupted. It does not own Amazon.
Bonus Insights
Collins' closing word was the same as his opening one, and the position follows from it: "Again, I'll go back to what I started with in that I think that they're going to have to print a lot of money." He does not think that ends in catastrophe, and said so plainly, but he thinks it is the response to both states of the world. "And so I think that if something really goes bad, I know they're going to helicopter print money." Status quo produces the same answer. "And I think that's where the majority of our capital sits."
The Druckenmiller aside: Collins asked whether Eisman had read about the scandal involving Druckenmiller, and it turned out to be the admission that an op-ed had been written with AI. Collins: "He admitted to it. It's like that's why it's not a scandal." Eisman was unimpressed — "My response to him is you could have just written it yourself. It's not that complicated" — and when Collins relayed Druckenmiller's defense, "He said he's an econ major, not an English major," Eisman's answer was that a five-paragraph editorial is not that hard. Collins: "I think that doing it is actually the sign of weakness."
Eisman's history with the Carvana family is older than the short. He visited Ugly Duckling, Ernie Garcia Senior's used-car company, in the 1990s. "I got an Ugly Duckling towel which I loved and used all the time." He used it until it became a rag, and when he came to work with Daniel around 2001 his reaction to seeing the name on the position sheet was to ask what on earth it was.
On why the new media suits this material, Eisman said the conversations are a live debate on AI and private credit rather than a broadcast, and that it beats the old format for it. He also said he has learned more about technology in the past year than he ever had before. Collins said that five years ago neither of them would have believed they would start a Substack, and that they enjoy educating what he called "our own little pocket of crazy," taking questions from readers on a Discord all day.
The running joke about clothing got two stories. A viewer once asked whether Collins was wearing a jacket with a hoodie and another replied, "you must be new here." Daniel was stopped in the middle of Penn Station by people soliciting him who called him GQ and a chief executive, and he paid them twenty dollars for it: "Like whether it's true or not, it just felt good." Collins remembered Daniel buying a bright pink button-down, socks and a blue sweater at Paul Stuart on the way to a meeting he was not dressed for.
Collins and Daniel are positioned for a government that keeps printing rather than cutting — gold as the core holding, a handful of small, unfashionable longs around it, and a short book kept deliberately small because the leveraged multi-manager funds now set the price of being early.
Products, Companies & Tools Mentioned
Nvidia (Eisman's note-7 finding: the top five direct customers were 70% of accounts receivable in a quarter when revenue was up more than 100%)
OpenAI and Anthropic (The two customers Collins says Wall Street research puts at 70% of hyperscaler AI revenue; Eisman's quarterly numbers had Anthropic at "11 and a half billion" and growing over 100%, OpenAI at "6.5 billion" and growing 18% with costs of "12 billion")
Google (Issued "$85 billion in equity" on Collins' account, and the only household name Seawolf says it owns, bought when the market decided search was being disrupted)
SpaceX (Collins' template for the supply problem — a private giant coming to market; Daniel and Collins both noted asteroid mining appears in the S-1 as a future activity)
Citadel, Millennium and Point72 (The multi-manager "pods" Daniel says dominate the short side: levered roughly five to one, beta and factor neutral, happy with a 3% to 5% spread and covering a short that has not worked in three weeks)
Carvana (The short: 75% to 100% of pre-tax income from gain on sale, with a two-year average gain of "109 110" that implies a buyer paying north of 110)
Ally (One of Carvana's biggest whole-loan buyers, paying an estimated 102 to 104)
Delaware Life (The insurance company Daniel says owns "a heck of a lot of Carvana paper", with related-party transactions he said rose from about 3% to 30% or 40%)
FICO (The other short: Daniel says it raised prices 1,600% instead of inflation plus 2% — "It was pigish" — and Eisman says mortgage pulls cost three to five times a credit-card pull)
Visa and Mastercard (The comparison Daniel used for pricing restraint nobody would have objected to)
Circle ("Circle doesn't make any sense to me. Zero," Daniel said, while noting the stock has rallied enormously)
Tesla (On and off short for the better part of five or six years, with earnings falling and the stock rising; Collins' conclusion was "don't short a cult")
Golar LNG (A long: four vessels that liquefy gas offshore on 20-year toll-road contracts, with two headed to Argentina next year)
Cheniere and YPF (The onshore alternative — Collins said YPF is going to spend 30 to 50 billion dollars on a plant in Argentina)
Glass House Brands (The other long: a California greenhouse cannabis grower, with the trade resting on rescheduling and the gap between roughly $200 a pound in California and six or seven hundred in Germany)
ServiceNow (Bought cheaply during the software sell-off, as an example of Seawolf only entering large-cap names when something is being thrown away)
Wells Fargo (Collins' precedent for buying a failed business cheaply: Wachovia "was purchased for nothing," and the same with Bear Stearns)
Paul Stuart (Where Daniel bought a pink button-down and a blue sweater on the way to a meeting he was not dressed for)
Books & Resources Mentioned
What Are We Doing? (Daniel and Collins' newsletter, and the source of the shorting article Eisman asked about)
Where's Your Ed At – Ed Zitron (The weekly AI research Eisman described as about 20 pages long, and where the Nvidia receivables point came from)
The Untouchables (Collins built his shorting essay on Sean Connery's character — knowing better than to go in, and going in anyway)
Finding Forrester (Eisman's own Connery pick)
For All Mankind (The Apple science-fiction series Collins noted is built partly on asteroid mining, the same activity in SpaceX's S-1)
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: