Jensen Huang joined six Wall Street banks to backstop loans buying his chips. The $500 billion AI-financing plan announced this month is, on the published description in The Wall Street Journal, a consortium of the chip maker and six of Wall Street’s biggest firms, with the stated purpose of backstopping lending to chip customers who otherwise could not afford the silicon. The distinction between a backstopper and a co-signer is real in some worlds. In this one it is less real than it looks, because the structure puts Nvidia’s balance sheet within the loss path of its own revenue. If the loans go bad, the consortium absorbs the loss. If the consortium absorbs the loss, Nvidia’s contribution to the backstop is the part of Nvidia that was supposed to be paid for the chips.
To be fair — the phrase is doing real work here, not Letterkenny work — Nvidia is not, on the published description, writing the cheques itself. The consortium is the formal vehicle. The underlying relationship is what it has been for two years: a chip maker whose biggest customers can no longer afford its chips without help arranging the financing. The difference is that, this month, the help has a name and a number.
The customers are not in the condition the 2024 capex forecasts assumed they would be in by the summer of 2026. OpenAI told investors its revenue rose 18 percent in the second quarter while losses deepened. The hyperscalers — the data-center operators running the model-training fleets, among them some of Nvidia’s largest customers — were once the cash-printing machines of the consumer-internet era. They are now, per the Journal, “relying more on debt.” Microsoft’s quarter was good enough to deliver, per the Journal, the largest one-day market-cap gain by any company. SpaceX went public and saw a trillion dollars of paper value evaporate. An $890 billion wipeout followed Alphabet’s and Tesla’s July results. The hedge fund Situational Awareness unwound in the same window.
It is, in other words, a bezzle of the kind J.K. Galbraith named for the 1929 crash and that Cory Doctorow, borrowing Galbraith’s term for his 2024 novel, has spent the last several years describing in the AI trade: the interval during which the embezzler has the gain and the victim does not yet feel the loss. The customers have the chips. The lenders have the interest payments. The chip maker has the order book. The investors have the chart. The actual revenue that justifies any of this has not yet arrived. Shaia Hosseinzadeh, who runs OnyxPoint Global Management, told the Journal: “The macro data is really quite robust. Of course, there’s a level at which everything breaks.” The level at which everything breaks is, in a backstopped-loan structure, the level at which the backstopper has to decide whether to honour the backstop.
The market’s own tell is more interesting than the consensus is letting on. Wall Street is hoping Nvidia can beat expectations again — analysts project $92 billion in second-quarter sales, up from a forecast of $78 billion at the start of 2026, with the company needing to outrun 95 percent annual earnings growth to clear a target of more than $51.5 billion. Nvidia has beaten estimates for fourteen consecutive quarters since ChatGPT kicked off the AI race in late 2022, posting 210 percent annual net-income growth in its last reported quarter. By the metrics that drive the chart, the trade is a triumph. The options market is pricing a 5.3 percent swing in either direction after Wednesday’s report — higher than the 4.8 percent average post-earnings move over the last twelve months. The most actively traded contracts in recent days have been puts tied to the stock falling from Friday’s $214.75 close to $205 and $210 apiece. Per the Journal, investors have tended to sell Nvidia stock immediately after blockbuster earnings — shares falling each trading session after the past four quarterly reports. Smart money is not buying the chip rally. Smart money is paying for the right to be short the chip rally.
David Lefkowitz, who heads U.S. equities at UBS Global Wealth Management, told the Journal: “We joke internally that we’re all Nvidia analysts now.” The joke is the warning. When the analyst class of every major bank on the Street becomes functionally a single-company research department, the price discovery the Street is supposed to provide is the price discovery of one name. Brian Mulberry at Zacks Investment Management called the upcoming report “kind of becoming more and more like the World Cup final than the Super Bowl.” He meant it as a compliment to the spectacle. He was also describing the condition in which a single company’s earnings matter more to global equity indices than any plausible theory of diversification should permit. Frank Lee at HSBC Global Investment Research raised his price target for Nvidia to $360 from $325.
The chip maker is also, while the structure runs, doing the things the structure requires it to do. Nvidia took a stake in Cloverleaf Infrastructure, which arranges power for data centers. Nvidia struck a $6 billion deal with Poolside, a startup, aimed at developing a powerful open-weight AI model. The chip maker is becoming, by increments, the financing arm, the power procurer, and the model developer for the customers to whom it is simultaneously selling the chips. The vertical integration of the AI build-out is no longer metaphorical. The vertical integration is now financial, electrical, and computational in the same company.
There is no public consultation that this trade requires, because it is not, formally, a public trade. Nvidia is not a regulated utility. The $500 billion consortium is not, on the published description, a Federal Reserve facility. The customers are not FDIC-insured. The backstop is a private contract among sophisticated counterparties — and sophisticated counterparties are the constituency that has spent the last fifteen years arguing that they should be exempt from the kind of prudential regulation that would require them to mark this structure to its actual risk and capitalise it accordingly. They are correct, in the narrow sense, that no one forced them to take the other side of these loans. They are less correct that the rest of the system is unaffected when the structure goes. The 2008 analogue here is not the subprime trade — that one was a public-catalogue trade dressed as private credit. The 2026 analogue is closer to the 1929 case Galbraith actually named: a leveraged speculation in which the speculators have arranged to be paid in full during the rising market and to be paid in full, again, out of one another’s pockets, during the falling one. The deadline that matters is the level Hosseinzadeh named — the level at which everything breaks. The deadline is the level at which Huang’s bank has to decide whether to be a bank.