Intro
Better Offline host Ed Zitron uses his weekly monologue to take apart Nvidia's second-quarter results and the financing structure around them, from customer concentration and payment terms to the company's investments in Poolside, Perplexity, Hugging Face and a string of data center projects. He also covers Jensen Huang's claim that artificial general intelligence has already arrived, the CFO's 70% growth guidance, and the debt every buyer in the chain is relying on.
Host: Ed Zitron, host of Better Offline
Published: 27 August 2026 on Better Offline
Listen on Omny | 9 min
Key Takeaways
Nvidia's revenue is concentrated in a handful of buyers
"$96 billion worth of revenue and 16% of that coming from a single unnamed customer"
44% of first-half fiscal 2027 revenue came from three customers
Nvidia is now financing the customers who buy from it
Payment terms of 90 days to a year for certain investment-grade customers, and 70% of receivables sitting with five of them
Morgan Stanley analysts call it "a balance sheet as a service company"
Huang told an analyst call that AGI has already been achieved
Zitron: "This has, of course, not actually happened, will not actually happen"
The 70% growth guidance requires the biggest buyers to roughly double their orders
It also requires neoclouds like CoreWeave to raise more debt, on top of the over 9% interest it already pays
The announced $500 billion fund is not a fund
Per Bloomberg, the number "had no obvious provenance"
Nothing has broken yet, and that is the only reason nobody is panicking
"when those resources slow or stop, so too will the music"
Record Earnings, and the Customers Behind Them
Zitron opens on the results themselves: "The Groundhog once again saw its shadow with Nvidia reporting record earnings, continuing to inflate an ever more dangerous AI bubble with $96 billion worth of revenue and 16% of that coming from a single unnamed customer."
The concentration runs deeper than one quarter. For the first half of fiscal year 2027, he says, 44% of Nvidia's $177.8 billion in revenue came from three customers
He stops to explain the calendar rather than assume the listener knows it: Nvidia's fiscal years run from February 1st of the year before to January 31st of the year of, so the company is currently in fiscal year 2027, which began on February 1st, 2026, and these were second-quarter fiscal 2027 results
The explainer ends in an aside to someone off-mic: "Annoying, but I didn't do it. Don't blame me. Don't be mad at me, Farah."
70% of Receivables Sit With Five Customers
Nvidia is shipping on credit at scale. Zitron says 70% of the company's current accounts receivable is owed by five customers, which Nvidia disclosed while adding that for certain investment-grade customers it was offering payment terms ranging from 90 days to an entire year
On what investment grade means here: both CoreWeave and Nebius have had their debt rated investment grade, and he describes them as "two unprofitable neoclouds that exist only to raise debt and buy GPUs, mostly to rent them to the same three or four companies"
The ratings rest on the contracts behind them — the debt is collateralized by contracts with companies that have good credit
His view of the agencies is a joke rather than an argument: "What is it with ratings agencies? It's just like every time you put your hands in front of your face and they'll be like, where'd they go?"
The structural point he draws from all of it: the vast majority of Nvidia's revenue comes from a handful of companies, including hyperscalers that have made up more than 50% of its revenue over the last few years, and those companies are now reliant on debt because of what he calls skyrocketing GPU costs
Nvidia is unquestionably profitable and thriving on revenue, he says, but "it's only doing so because of a few companies who are willing to be fin-dommed by Jensen Huang"
The Run-Up to Earnings Was a Spending Spree
Poolside: Nvidia sunk $6 billion into what Zitron calls "a flailing open-source LLM company that failed to raise billions of dollars earlier in the year", hiring away most of its staff in what its founders said was not an acqui-hire — "Other than the fact that everyone's leaving, and other than the founders."
Perplexity: Nvidia is rumored to be sinking billions into the search company at "a $30 billion valuation, which is insane and it's not worth that", which he says is likely because Perplexity is one of the few companies in the world that buys any significant amount of AI compute. He puts no weight on his own estimate of how much — "Likely a few hundred million dollars a year, but I'm just guessing."
Data centers and power: it is also investing in Stargate, Abilene and Lancium at a multi-billion dollar valuation, considering putting money into AI data center power company Cloverleaf, and backstopping over $100 billion of a SoftBank-backed data center for OpenAI to rent out in Ohio — a backstop that only kicks in if the thing actually gets built
Hugging Face: he interrupts himself to add that Nvidia is buying the AI model hosting platform for a little under $13 billion, against $150 million in annualized run rate, "which works out to a pathetic $12.5 million a month, which is insanely small considering its notoriety"
His conclusion from the price: "It's almost as if every AI company just kind of stinks."
Balance Sheet as a Service
The portfolio, as he reads it out: Nvidia has invested in Anthropic, OpenAI, CoreWeave, Nebius, IREN, Nscale, Intel and SpaceX
It has also backstopped CoreWeave's leases and signed over $30 billion in multi-year agreements to rent back its own GPUs
On the revenue-share structures: Nvidia has "created bizarre revenue share deals where it agrees to backstop data centers and receive a cut of the revenue above a certain threshold, which is obviously dependent on the bloody thing being built and also being paid for"
The company denies the framing. On the latest earnings call, CFO Colette Kress said Nvidia did not see this as circular financing — a denial Zitron rejects "despite there being so many different deals where the money moves in a perfect circle"
The sell-side label he borrows: Morgan Stanley analysts describe Nvidia as "a balance sheet as a service company", with over $366 billion in commitments including $25 billion of data center leases yet to commence
Which prompts the question he puts to the CEO directly: "Why does Nvidia need to rent back the GPUs? I thought the demand for AI was so high, Jensen."
Huang Says AGI Has Already Arrived
On an analyst call, Zitron says, Huang claimed "the world has already achieved artificial general intelligence, which he defines as the ultimate form of the technology where the machines can think and act for themselves"
Zitron's response is flat denial: "This has, of course, not actually happened, will not actually happen, and should genuinely get it lambasted in the press, but because number has gone up, everybody is happy."
He acts out the reaction he says the claim got instead of scrutiny, cheering Huang and the arrival of AGI before cutting himself off
The 70% Guidance and What It Would Take
The line he calls the worst of the call: "somehow the dumbest thing that was said on the earnings call came from Nvidia's CFO, who said that Nvidia will grow its revenue by 70% in fiscal year 2028", a year that begins on February 1st, 2027
The arithmetic he runs on it: "And based on consensus estimates for the current fiscal year being at $396 billion, this means that Nvidia is expecting to make over $674 billion next year."
He does not call it impossible. He says it would require Nvidia's current customers — predominantly Amazon, Google, Microsoft, Meta, SpaceX and Oracle — to likely double their current orders for GPUs, and neoclouds like CoreWeave to raise even more debt than they are already raising
On what that debt already costs: "CoreWeave is already having to pay over 9% interest on its debt just to get investors through the door"
His estimate of the additional borrowing: "I mean, this must be over a trillion at this point in the next year. It's completely fucking insane."
The same treadmill applies to the model companies, which he says must keep growing revenue, customers and funding rounds to meet over $1.1 trillion in compute agreements, while Nvidia "must, through science or magic, find ways for its customers to be able to buy more and more and more GPUs every single quarter from here into eternity"
Everything in the Chain Runs on Debt
The thesis he says he keeps returning to: "the AI boom is only possible as long as debt can sustain it"
SoftBank has taken out tens of billions of dollars of debt to fund OpenAI and is now trying to sell $20 billion of bonds to refinance that debt
CoreWeave can only keep building data centers as long as the banks give it money
Oracle, Google, Amazon, Meta and SpaceX cannot afford to buy GPUs out of cash flow, and so depend on debt to keep buying them
Nvidia has borrowed too, and he corrects himself on the size live: "Even Nvidia itself has had to take out over $25 billion in bonds. Maybe it's $20 billion. Regardless, still a lot of money."
The $500 Billion Fund Bloomberg Says Isn't One
He puts air quotes around the supposed $500 billion fund, noting for an audio audience that the gesture cannot be seen, and says that per Bloomberg it is nothing of the sort
What the announcement actually was: "that announcement was literally just Nvidia saying that a group of asset managers would invest half a trillion dollars in AI data centers, and that number, and I quote, had no obvious provenance"
He says he will link to the Bloomberg article in the episode notes
Why the Hyperscalers Keep Writing Checks
The demand underneath the capital spending is two customers. Hyperscalers are only spending this money, he says, because Anthropic and OpenAI have committed to spend over $400 billion renting GPUs in the next three and a half years, according to analysts
"And without that spend, 30% or more of their cloud revenues will evaporate."
The single-cloud version of that number comes from UBS: Anthropic and OpenAI will make up over 48% of Google Cloud's revenues in 2027, which he says will only be possible if they can both afford to spend the money and the data center capacity is actually available
Capacity is his caveat: "And data centers are taking forever. Another problem."
Why Nvidia Invests in Everyone, and Why Nobody Is Panicking Yet
The reason for the investment spree, as he reads it: "Nvidia is investing in every schmuck with a penchant, that's definitely how you say it, for AI models" because any AI company becoming insolvent or doing a down round would break the illusion that AI is the next industrial revolution rather than "a series of different companies handing money to each other in the hopes that a business model arrives"
"The reason nobody is freaking out is that nothing bad has happened yet."
Nvidia's big customers can still raise money to pay it, and its "horrible neocloud progeny can continue to loot the debt markets from gullible investors who have been lied to by analysts and the media about AI's promise and stability"
The condition the whole structure rests on: "Everything comes down to whether near-infinite resources are available for AI in perpetuity, and when those resources slow or stop, so too will the music."
He signs off promising to return next week with two co-hosts "to talk about horrible AI slop and whatever else crosses our wretched little minds"
Zitron's bottom line is that Nvidia's record quarter is being paid for with borrowed money by a handful of customers Nvidia is itself financing, and that the whole arrangement holds only for as long as the debt markets keep lending.
Products, Companies & Tools Mentioned
Nvidia (The subject of the monologue: record revenue, extreme customer concentration, receivables owed by five customers, and an investment and backstop program across most of its own buyers)
CoreWeave and Nebius (Unprofitable neoclouds rated investment grade on collateralized contracts; CoreWeave's leases are backstopped by Nvidia and it pays over 9% interest on its debt)
OpenAI and Anthropic (Committed to over $400 billion of GPU rental in three and a half years per analysts, and to over $1.1 trillion in compute agreements they must keep raising money to meet)
Poolside (Nvidia sunk $6 billion into it and hired away most of its staff in what its founders said was not an acqui-hire)
Perplexity (Rumored Nvidia investment at a valuation Zitron says the company is not worth, explained by its being one of the few real buyers of AI compute)
Hugging Face (Being bought by Nvidia for a little under $13 billion against $150 million of annualized run rate)
SoftBank (Took on tens of billions of debt to fund OpenAI and is selling $20 billion of bonds to refinance it; also behind the Ohio data center Nvidia is backstopping)
Amazon, Google, Microsoft, Meta, SpaceX and Oracle (The customers who would have to roughly double GPU orders for the 70% guidance to land, and who cannot fund those purchases from cash flow)
Stargate, Abilene, Lancium and Cloverleaf (Data center and data center power projects Nvidia is investing in or considering)
IREN, Nscale and Intel (Further names on the list of Nvidia's investments)
Books & Resources Mentioned
Bloomberg's reporting on the $500 billion fund (Zitron's source that the announced number "had no obvious provenance"; he says he will link it in the episode notes)
Morgan Stanley research on Nvidia (Source of the "balance sheet as a service" description and the commitments figure)
UBS estimate on Google Cloud (Source for Anthropic and OpenAI making up over 48% of Google Cloud revenues in 2027)
Get the latest market chatter and takes as they happen:
X | Threads | Instagram | YouTube | TikTok | Facebook