Nvidia Bet Billions on Financing the AI Boom. Now the Bill Is Coming Due.

Nvidia doesn't just sell the chips powering the AI boom anymore. It increasingly bankrolls the companies buying them. That arrangement is now getting its first serious stress test, as the neocloud firms Nvidia has backed with billions in guarantees and investment take a beating in the market.

Shares of CoreWeave, Nebius and IREN — all cash-burning cloud providers in Nvidia's portfolio — are down at least 30% this month. CoreWeave, the largest of the three, has seen the cost of insuring against its default soar, with credit default swap spreads topping 11%. Nvidia itself just closed its third straight down week on Wall Street and is off 7.6% in October, its worst month since last November.

The selloff is a test of a strategy CEO Jensen Huang has pursued for the past year or so: using Nvidia's investment-grade credit rating and enormous cash pile to make sure its customers can actually afford to keep buying its hardware. As of Nvidia's last quarterly filing in July, the company had $36 billion in off-balance-sheet AI cloud agreements — essentially backstops for neoclouds — on top of a guarantee covering up to $105 billion in chips, land and power for a proposed 10-gigawatt OpenAI data center in Ohio. It had also signed on as a tenant for $20 billion in future data center leases it plans to hand off to third parties later.

In most of these deals, Nvidia only pays if something goes wrong — if a borrower defaults, or a tenant can't make rent.

"If the demand for compute doesn't meet these really super lofty expectations, we're going to have a problem," David Trainer, founder and CEO of financial research firm New Constructs, told CNBC.

Why Nvidia became a lender

The logic behind Huang's approach is straightforward: Nvidia's newest hardware is staggeringly expensive, and somebody has to pay for it. The company's flagship Vera Rubin system is a seven-foot rack packing 72 GPUs, weighs about as much as a pickup truck, and runs more than $7 million per unit, according to a Morgan Stanley estimate cited by CNBC. Getting that gear into data centers at scale means helping customers line up financing — so Nvidia has been offering loan guarantees, equity stakes and lease backstops to grease the wheels.

It's working, at least by the revenue numbers. Nvidia pulled in $216 billion in its latest fiscal year, roughly eight times what it made before ChatGPT launched in late 2022. The company generated $70 billion in free cash flow in just the first half of this year — dwarfing rivals like AMD, which posted a little over $4 billion in the same period, and Alphabet, which actually went cash-flow negative in the second quarter funding its own AI buildout. That cash advantage, combined with a $99 billion equity portfolio spanning dozens of companies including OpenAI and Anthropic, gives Nvidia leverage competitors like Google's TPU business and AMD's new Helios AI racks don't have.

Nvidia's August filing disclosed more than $500 billion in off-balance-sheet commitments stretching out through fiscal 2029. That same month, Huang unveiled a $500 billion financing partnership with six large asset managers aimed at treating AI compute infrastructure as its own investable asset class, backed by up to $125 billion in residual value guarantees on GPU financing. Nvidia has also seeded capital into private equity vehicles aimed at AI infrastructure, including $2 billion into Brookfield's debut AI Infrastructure Partners Fund, which is targeting $10 billion total, plus undisclosed amounts in KKR's $11 billion Helix Digital Infrastructure fund and Valor Equity Partners' $5.4 billion compute fund.

"We used to build chips that we sell, and these are technology components that people buy and use," Huang said when he announced the asset managers' deal in August. "But now, Nvidia's AI factory platform is really an investable asset."

The Ohio deal and what happens if it goes wrong

Nvidia's single largest guarantee backs OpenAI's planned data center in Pike County, Ohio — potentially the largest such facility in the world — and locks in use of Nvidia hardware for at least 20 years. If OpenAI misses payments or runs into insolvency, Nvidia has several options: take over the lease itself, find a replacement tenant, or have landlord SB Energy sell the facility. In the latter two scenarios, Nvidia guarantees to cover the gap, up to that $105 billion ceiling, if the lease or sale price comes in low. Nvidia could also just write a lump-sum check. If OpenAI posts its own collateral or reaches a high enough credit rating on its own, Nvidia's guarantee goes away.

CoreWeave carries a different kind of Nvidia exposure: an equity stake plus a commitment to buy up to $6.3 billion of its compute capacity if no other buyer materializes. With other neoclouds, like Sharon AI, Nvidia has structured revenue-sharing arrangements instead.

Huang insists CoreWeave doesn't actually need the backstop and said its cost of capital is falling. That's a notable claim given the numbers: CoreWeave carries roughly $51.6 billion in total debt, including about $35 billion in notes, bonds and loans, and yields on bonds maturing in 2031 and 2032 have topped 15%.

"They will sign contracts into the future, and that all looks fine and dandy in their earnings reports, but is that contract going to be stable?" said Brian Stutland, portfolio manager at Equity Armor Investments. "Is it going to break at some point?"

Nvidia isn't only exposed through the big names. The company is an investor in Australian AI cloud provider Firmus, which pulled its IPO plans this month. It's also backed Nscale out of the U.K. and SoftBank-affiliated SB Energy, both of which filed to go public in the U.S. in September and have gone quiet on their plans since.

Huang's pitch, and the skeptics

Huang isn't backing down from the strategy. He told CNBC this week that demand for neoclouds is "off the charts," and that the real bottleneck isn't appetite for AI compute but working capital. "There's just so much demand across so many sectors of AI, there's a scarcity of capital," he said at Nvidia's GTC conference in Berlin. On a recent podcast with investor Brad Gerstner, Huang described Nvidia's equipment in three words: "We're the most rentable. We're fungible, and we're durable."

Deepwater Asset Management founder Gene Munster has made a similar case, telling CNBC last month that Nvidia's projected $700 billion in revenue next year "affords them the ability to finance some of their customers."

Not everyone is convinced the exposure is manageable. Research from New Constructs puts the net present value of Nvidia's off-balance-sheet liabilities in fiscal 2026 at more than $153 billion — up over $100 billion from the prior year. Ratings agency Moody's has flagged guarantees as the fastest-growing category of off-balance-sheet financing arrangements across the AI sector this year. Moody's doesn't currently treat those guarantees as debt when rating corporate buyers, but the agency is watching closely.

"These things are advanced structures that are unique, and so there is no blanket real approach you can take," said David Gonzales, a Moody's vice president. He noted chips will likely hold value above their contractual residual amounts for now, which cushions some of the risk — but how much exposure investment-grade AI players can absorb remains an open question. "It's the center of all of our discussions right now," Gonzales said.

Nvidia is also shopping for insurers willing to take on some of the risk tied to potential GPU value declines. "Insurers are not yet a complete panacea to this problem," said Tom Hoad, executive director of product innovation at insurance group Howden, "but they are probably going to turn up and be helpful in the right situations."

Warren Hosseinion, co-founder of AI infrastructure startup Compute Labs, framed Nvidia's position bluntly: the company doesn't want to be an insurer of last resort, "but they're willing to play that role while the market and the industry catches up." That gives Nvidia outsized influence over the pace of the entire buildout. "They have to be very careful about how they release their technology at what pace," Hosseinion said, "because a lot of that will dictate how the market moves around it."

Notably, this is all a recent development. Nvidia's annual filing from February — just eight months before the current turmoil — didn't mention AI cloud agreements at all. Most of its credit backstops have materialized in the past six to twelve months, right as the companies they support have started wobbling on Wall Street.