Every market has its favorite story. Right now, the most beloved one is Nvidia’s: the indispensable supplier of the AI revolution, a company whose chips are so vital that Wall Street has decided to treat them not as depreciating hardware, but as long-lived, investable assets. On August 10, the company partnered with six of the biggest names in finance—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR—to raise over $500 billion for AI infrastructure. The logic is seductive: customers can borrow against their GPUs, turning a capital expense into a financed asset, much like an airline financing its fleet 2.
But a forensic reader should pause at the accounting premise. The entire structure depends on treating a GPU as a durable asset with a stable residual value. That is an assumption, not a fact. It is also an assumption that runs directly against the historical evidence of the semiconductor industry, where the useful life of a chip is measured in years, not decades, and where the value of a cutting-edge processor often collapses the moment a faster one ships. The deal does not require fraud to be dangerous; it only requires a modest miscalculation of depreciation to ripple through a half-trillion dollars of leverage.
The contradiction deepens when you look at the surrounding numbers. SK Hynix, Nvidia’s memory supplier, just approved a $38.3 billion investment in two new fabs, with the first cleanrooms opening in 2029 5. That is a bet on sustained demand, but it is also a bet that the current boom is not a bubble. Meanwhile, the macroeconomic backdrop is cooling: global inflation eased in July, with US consumer prices rising 3.4% year-on-year, down from 3.5% 3. That is good news for rate-sensitive assets, but it is not the kind of environment that usually justifies a $500 billion financing spree. The Nikkei jumped 2.1% on a weaker jobs report 4, and the Mexican peso strengthened on moderating US inflation and progress in Iran negotiations 7. Markets are celebrating the end of tightening. Nvidia is borrowing as if the party will never end.
The incentives here are worth spelling out. For the Wall Street partners, the deal generates fees and a new asset class to manage. For Nvidia, it locks in future demand for its GPUs, guaranteeing revenue even if end customers struggle to pay. That is the classic structure of a vendor financing scheme: the manufacturer lends its customers the money to buy its own products, which inflates current sales at the cost of future risk. The defense, which Nvidia would surely offer, is that the counterparties are sophisticated institutional investors who understand the technology cycle. That is true, but sophistication does not immunize a market from collective mispricing. It never has.
The broader context is no less concerning. Anthropic, an AI company, is reportedly targeting a $2 trillion valuation for an October IPO, backed by a forecast of explosive revenue growth 11. The company says second-quarter revenue exceeded $11.5 billion, more than 14 times the year-earlier figure 11. That is an extraordinary claim, and it deserves extraordinary scrutiny. Meanwhile, a Ferrari EV sold for $40 million at auction, far above its $1.1 million estimate 8. It is a reminder that when prices detach from fundamentals, the most beautiful stories can be the most dangerous.
The unresolved risk is not whether Nvidia’s chips are good. They clearly are. The risk is whether the financial engineering built on top of them can withstand a normal correction in the technology cycle. If GPU prices fall faster than the depreciation schedules assume, the collateral backing that $500 billion will be worth less than the debt it supports. That is not a prediction of fraud; it is a statement of arithmetic. The question for the reader is whether the market is pricing in that arithmetic, or only the story.