George TsagkarakisNvidia just told the market it will eat part of the loss if its own GPUs turn out to be worth less than promised. Not a slogan. A contractual guarantee, capped at 25 percent of a given transaction, reviewed project by project.
That’s the mechanism behind the headline number: letters of intent with six financial firms to mobilize more than $500 billion in third-party capital for data centers, chip factories and power plants. The partners are Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR.
Investors didn’t cheer. Nvidia’s stock fell about 1.4 percent after the deal surfaced, erasing more than $70 billion in market cap.
What the guarantee actually covers
If the resale or reuse value of installed hardware comes in below expectations at the end of a financing term, Nvidia covers part of the gap. Up to 25 percent of the transaction. The chipmaker is taking on some of the depreciation risk on its own products.
Nvidia CEO Jensen Huang said that share is “significantly lower” than in other compute financing arrangements. The actual credit assessment, meaning the evaluation of the customer, demand, utilization, cash flow and residual value, stays with the capital providers.
Read that carefully. Nvidia absorbs the tail risk on the asset. Somebody else underwrites the borrower.
The number is a target, not revenue
Huang described the move on X as a shift from one-off projects to repeatable financing platforms, with “AI factories” funded the way power grids or transportation networks are. Plenty of AI companies have demand for compute and no way to raise capital at the scale they need.
He was explicit about what the $500 billion isn’t. It’s an aggregate target spread over years. Not Nvidia revenue. Not a single fund. Not a commitment to any one customer.
Nvidia didn’t share terms, individual commitments or a timeline. That’s a lot of unspecified structure sitting under a very specific number.
The circularity question Huang couldn’t dodge
Nvidia regularly supports its partners in taking on debt, and that debt turns into Nvidia revenue. Huang addressed the accusation directly, and the residual-value guarantee is his answer to it: real risk transferred back onto Nvidia’s balance sheet rather than pure vendor financing dressed up as demand.
The company is also negotiating a guarantee for a 10-gigawatt data center in Ohio leased to OpenAI.
Michael Burry called this the fraud, and Huang is arguing the other side
Investor Michael Burry called the hyperscalers’ depreciation practices “one of the more common frauds of the modern era.” His argument: GPUs go obsolete too fast to justify five-to-seven-year useful lives, because Nvidia itself ships a new generation every two to three years. He put the understatement at roughly $176 billion between 2026 and 2028 alone.
Huang claims the opposite. He says the A100, launched in 2020, is still in commercial use six years later, with an economic lifespan stretching toward a decade, and that CUDA keeps improving installed hardware over time.
His market evidence is rental pricing. H100 annual contracts went from $1.70 per GPU-hour in October 2025 to $2.35 in March 2026. B200 capacity runs between $5.30 and $7.05.
Rising rents on a five-year-old part is a genuine data point. It’s also exactly the kind of price signal that holds until supply catches up.
The numbers everyone else is working with
Morgan Stanley expects hyperscaler spending of $3.5 trillion between 2026 and 2028. Apollo president Jim Zelter puts the total investment need at over $8 trillion. Against $8 trillion, a $500 billion financing target is a slice, not the plan.
The Bank of England warned in its July Financial Stability Report that the pace is historically unprecedented, and that a shock hitting highly leveraged AI companies could ripple through global financing conditions and trigger a credit crunch. Banks and private credit firms, the report noted, have limited visibility into their indirect exposure.
That last line is the one worth keeping. The residual-value guarantee makes a lender’s downside legible on paper, at 25 percent, project by project. What nobody has published is who holds the other 75 percent, or how much of it traces back through the same six firms.
