Nvidia — correction, Nvidia — unveiled a financing structure this week under which Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR will commit up to $500 billion to build AI data centers, with Nvidia itself guaranteeing the resale value of the GPUs used as collateral. The company has agreed to cover up to 25% of any shortfall if those chips fail to hold their book value at liquidation. That commitment, not the headline dollar figure, is the actual news.
The mechanism is straightforward. A data center operator borrows against a stack of Nvidia GPUs. If the operator defaults and lenders have to sell the hardware, and if the market price comes in below what the loan assumed, Nvidia writes a check for a quarter of the gap. The financiers get downside protection. Nvidia gets $500 billion in third-party capital flowing into buyers of its chips without putting the principal on its own balance sheet.
Bond markets reacted sharply enough that Nvidia CEO Jensen Huang went on X and business television to explain the exposure. The concern is what financiers call wrong-way risk: Nvidia's obligation grows precisely when GPU demand weakens, which is also when Nvidia's own revenue would be under pressure. It is the kind of correlation that turns a manageable guarantee into a compounding problem in a downturn.
Key facts
- 01Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR committed up to $500 billion to build AI data centers backed by Nvidia GPUs.
- 02Nvidia pledged to cover up to 25% of the shortfall if GPUs pledged as collateral fail to hold their expected resale value.
- 03Bloomberg tallied roughly $750 billion in other circular Nvidia deals struck this summer, including tie-ups with OpenAI, Anthropic, CoreWeave, Nebius, Firmus and Lambda.
- 04CEO Jensen Huang defended the structure on X after bond markets reacted to the guarantee.
- 05Microsoft CEO Satya Nadella recommended '1873,' a book on railroad-era financial engineering, on the company's most recent earnings call.
Huang's counterargument is that the structure is the opposite of circular financing, not an example of it.
He has a point on the mechanics. Unlike the Lucent Technologies playbook of the dotcom era, where the vendor lent customers the money to buy its own equipment, Nvidia is not funding the buildout directly here. The six financial firms are. Nvidia is only backstopping a portion of the residual value. That said, the Lucent comparison is not baseless. Nvidia has committed billions to customers buying its chips, including frontier labs OpenAI and Anthropic, and neoclouds CoreWeave, Nebius, Firmus and Lambda. Bloomberg calculated roughly $750 billion in circular Nvidia-linked deals struck this summer alone.
The timing matters. Traditional funding channels for AI infrastructure are showing strain. Oracle has issued substantial debt. Google has raised equity. Meta has burned cash aggressively, with free cash flow collapsing last quarter as capex climbed. Microsoft CEO Satya Nadella recommended '1873,' a book about railroad-era financial engineering that ended in a crash, on the company's most recent earnings call — a notable choice of reading list from the executive with the largest AI infrastructure commitments in the industry.
Underneath the headline number sits the more interesting bet. By guaranteeing residual value, Huang is trying to will a secondary market for used Nvidia GPUs into existence.
“When needs change, the factory can be used by another customer, another cloud or another operator. This broad ecosystem gives NVIDIA compute a deep market of potential users and offtakers, helping protect residual value.”— Jensen Huang, Nvidia CEO
If it works, the implications extend well beyond the six named financiers. A liquid market for aging Nvidia hardware would let startups, enterprises and researchers buy previous-generation H100s and Blackwell chips at meaningful discounts, the way developers now reach for open-weight models alongside frontier ones. It would also let Nvidia sell new chips into hyperscalers without cannibalizing itself when those hyperscalers rotate inventory. The company would earn a spread on every generation twice — once on the sale, and again by keeping the used-hardware ecosystem tied to its software stack.
The risk is that the entire premise depends on AI demand continuing to outstrip capacity for years. If enterprises pull back on AI usage, or if a new architecture makes today's GPU fleets obsolete faster than depreciation schedules assume, Nvidia's 25% obligation becomes real money against declining revenue. Huang's framing of AI servers as 'AI factories' comparable to railroads or airlines is a bet on a specific timeline. Buggy whips also had a long-term story until they didn't.
The structural read for the AI market is that Nvidia is now underwriting the financing plumbing of its own customer base, not just selling hardware. That is a meaningful shift in what it means to be a chip company, and it locks Nvidia deeper into the fortunes of every hyperscaler, neocloud and frontier lab building on its silicon. If the AI capex cycle keeps running, this move looks like the smartest financial engineering in the industry. If it stalls, Nvidia will discover exactly how correlated its guarantees are with its revenue — and the $500 billion number will read very differently in retrospect.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
