Nvidia bets its own money to unlock $500B in AI infrastructure deals

Nvidia just made a move that’s either brilliantly pragmatic or a sign the AI infrastructure boom needs a crutch. The company has signed letters of intent with six financial heavyweights — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR — to mobilize over $500 billion in third-party capital for data centers, chip factories, and power plants. The hook that makes the math work: Nvidia will guarantee up to 25 percent of the residual value of its own chips installed in those projects.

The guarantee is a form of depreciation insurance. If the resale or reuse value of Nvidia hardware tanks below expectations by the end of a financing term, the chipmaker covers part of the gap — but only up to a quarter of a given transaction, assessed case by case. CEO Jensen Huang insists this exposure is “significantly lower” than typical compute financing arrangements, and that credit assessment stays with the lenders. Still, the market flinched. Nvidia’s stock dropped about 1.4 percent on the news, vaporizing more than $70 billion in market cap. Investors appear to be asking: if the chips are such a sure thing, why does Nvidia need to backstop their value?

This lands squarely in the middle of a heated debate about depreciation. Investor Michael Burry has called hyperscalers’ accounting practices “one of the more common frauds of the modern era,” arguing that GPUs become obsolete too fast for five-to-seven-year depreciation schedules — especially with Nvidia’s two-to-three-year upgrade cycle. Huang is now effectively arguing the opposite, claiming the A100 from 2020 is still commercially viable six years later and that CUDA improvements stretch economic lifespans toward a decade. As evidence, he points to rising rental prices: H100 annual contracts jumped from $1.70 per GPU-hour last October to $2.35 in March, while B200 capacity runs between $5.30 and $7.05.

Whether those rental prices hold up over years is the trillion-dollar question. Morgan Stanley expects hyperscaler spending of $3.5 trillion between 2026 and 2028, and Apollo president Jim Zelter puts the total investment need above $8 trillion. But the Bank of England’s July Financial Stability Report warned the pace is historically unprecedented — a shock to highly leveraged AI companies could trigger a credit crunch, and banks have limited visibility into their indirect exposure. Nvidia is now tying a piece of its own balance sheet to the bet that won’t happen.