Nvidia's $153 Billion Bet as AI Financier Meets Its First Stress Test

Nvidia is sitting on more than $153 billion in projected off-balance-sheet liabilities for fiscal 2026 — up over $100 billion from last year — and the neoclouds it finances are getting hammered in the market. Shares of CoreWeave, Nebius and IREN, all cash-burning companies in Nvidia's portfolio, are down at least 30% this month. CoreWeave, the biggest of the group, has watched the spread on its credit default swaps soar past 11% as investors pay up for insurance against default.

The numbers land at an awkward moment for CEO Jensen Huang. Nvidia closed its third straight down week on Wall Street on Friday. The stock is off 7.6% in October, on pace for its worst month since November of last year.

Nvidia declined to comment for this story. But Huang, speaking to CNBC's Arjun Kharpal on the sidelines of the company's GTC conference in Berlin, made clear he isn't flinching.

"What is missing in the middle is the working capital to fuel that growth," Huang said, after telling CNBC that demand for neoclouds is "off the charts." "Because there's just so much demand across so many sectors of AI, there's a scarcity of capital."

How did Nvidia become a bank?

The company built the chips that powered the early generative AI era. Then it started funding the buyers. Huang's strategy has been to pair the most advanced technology with investment and loan assistance, so his systems can reach many more hands.

The mechanics matter here. Nvidia has an investment-grade credit rating, and it has used that rating to provide backstop guarantees on mega-projects for OpenAI so the buildouts can get funded at lower interest rates. For the vast majority of these deals, Nvidia doesn't have to pay up unless project participants default.

The scale is extraordinary:

• $36 billion in off-balance-sheet AI cloud agreements, or backstops, for neoclouds, as of its July quarterly filing
• A guarantee of up to $105 billion for chips, land and power at a proposed 10-gigawatt OpenAI data center in Ohio
• $20 billion in future data center leases signed as a tenant, to be assigned later to third parties
• A portfolio of equity stakes worth $99 billion as of the second quarter, including big positions in OpenAI and Anthropic
• Over $500 billion in total off-balance-sheet commitments over the coming years, a majority maturing through fiscal 2029, per its August filing

The credit backstops have mostly kicked in over the past six to 12 months. Nvidia's last annual filing, published in February, makes no mention of AI cloud agreements at all.

"If the demand for compute doesn't meet these really super lofty expectations, we're going to have a problem," said David Trainer, founder and CEO of financial research firm New Constructs. It was his firm's research that put the net present value of those fiscal 2026 liabilities at over $153 billion.

What's spooking the neocloud market?

Nvidia has backed other neoclouds that have run into trouble beyond the stock slide. The company is an investor in Australian AI cloud provider Firmus, which abruptly withdrew its IPO plans this month. It's also behind Nscale out of the U.K. and SoftBank-backed SB Energy — both filed to go public in the U.S. in September but haven't provided updates on their plans in weeks.

CoreWeave carries roughly $51.6 billion in total debt, including about $35 billion in notes, bonds and loans. Yields on its bonds maturing in 2031 and 2032 have topped 15%. Huang said this week that CoreWeave doesn't need Nvidia's backstop and that the neocloud's cost of capital is decreasing.

Brian Stutland, portfolio manager at Equity Armor Investments, said the market jitters reflect concerns about the stability of the underlying business.

CoreWeave and other neoclouds "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?" Stutland said. "Is it going to break at some point?"

At CoreWeave, Nvidia's investment includes equity and a commitment to buy up to $6.3 billion of its compute if it can't find another taker. Nvidia's backstops can take many forms: guaranteeing a minimum resale price or minimum hourly offtake price for the chips if a lender has to recoup the assets. With Sharon AI and other neoclouds, the deals come with revenue-sharing agreements.

"We believe these AI cloud partners have strong customer demand and robust sales pipelines but are constrained by the large-scale infrastructure that is required to meet that demand," Nvidia said in its quarterly filing.

Why does the Ohio deal matter so much?

Nvidia's biggest single guarantee covers OpenAI's forthcoming facility in Pike County, Ohio, which could be the largest data center in the world. That deal requires the use of Nvidia hardware for at least 20 years.

The risk structure is spelled out plainly. If OpenAI misses payments or faces insolvency, Nvidia can take over the lease, find a new tenant, or ask the landlord, SB Energy, to sell the facility — guaranteeing to pay the difference for the latter two arrangements up to $105 billion if the release or sale falls below a minimum price. Nvidia could also pay SB Energy a lump sum. On the other side, if OpenAI posts its own collateral or achieves a high enough credit rating, Nvidia's guarantee ends.

The shift from chips to capital is what makes this moment unusual. Before the AI boom, Nvidia sold graphics cards that fit into a standard server rack or PC. Its newest systems are far larger and far more expensive. The flagship Vera Rubin product is a seven-foot rack containing 72 GPUs, weighing as much as a pickup truck, and costs more than $7 million according to a Morgan Stanley estimate. Customers of that scale don't pay with a credit card.

Revenue hit $216 billion in the latest fiscal year, up roughly eightfold since OpenAI launched ChatGPT in late 2022.

Can anyone else play this game?

Not really, and that's the point. Nvidia generated $70 billion in free cash flow in the first half of this year. AMD totaled just over $4 billion. Google parent Alphabet went cash flow negative in the second quarter to fund its own buildout.

Cash is one edge. Competition is another question entirely. Nvidia still dominates the AI chip market, but Google rents out tensor processing units inside its massive cloud offering, and AMD launched its first AI rack, called Helios, in July.

Nvidia's answer is a full-service platform: software, services, integrated hardware and, for many customers, financing. In August, Huang announced a partnership with six large asset managers on a $500 billion financing push designed to treat compute infrastructure as an asset class. That agreement promised financial firms up to $125 billion in residual value guarantees on GPU financings.

"We used to build chips that we sell, and these are technology components that people buy and use," Huang told CNBC in announcing the agreement. "But now, Nvidia's AI factory platform is really an investable asset."

The company has also committed capital as a limited partner. It seeded $2 billion in Brookfield's inaugural AI Infrastructure Partners Fund, which aims to raise $10 billion in total, plus unconfirmed sums in KKR's $11 billion Helix Digital Infrastructure vehicle and Valor Equity Partners' $5.4 billion dedicated compute fund.

Huang has a marketing mantra to reinforce the pitch. "We're the most rentable. We're fungible, and we're durable," he said on a podcast with investor Brad Gerstner published last week.

What do the rating agencies say?

Off-balance-sheet liabilities have been growing across the entire AI trade, mainly in the form of lease obligations. Moody's has been sounding the alarm: in 2026, guarantees are the fastest-growing type of off-balance-sheet financing arrangement, the agency said.

Moody's doesn't count guarantees as debt when rating corporate buyers, but it's "something we're constantly monitoring," said David Gonzales, a vice president at the agency. "These things are advanced structures that are unique, and so there is no blanket real approach you can take," he said.

Gonzales said it's likely chips will retain "values in excess of contractual residual value amounts" in the near future, which would mitigate risk in many of Nvidia's financing structures. Still, how much exposure investment-grade AI players can handle is "the question we're evaluating at all times every day," he said. "It's the center of all of our discussions right now."

Nvidia is also actively looking for insurers to bear some of the risk tied to potential GPU value loss.

"Insurers are not yet a complete panacea to this problem," said Tom Hoad, executive director of product innovation at global insurance group Howden. "But they are probably going to turn up and be helpful in the right situations, and therefore open up some commercial transactions."

Is Nvidia stuck playing insurer?

Warren Hosseinion, co-founder of AI infrastructure startup Compute Labs, which tokenizes GPUs to open up capital pools for neocloud operators, doesn't think Nvidia wants the job. He thinks Nvidia will hold it anyway.

Nvidia doesn't want to be in the position of insurer, "but they're willing to play that role while the market and the industry catches up," Hosseinion said. "They kind of control how things move. They have to be very careful about how they release their technology at what pace, because a lot of that will dictate how the market moves around it."

Bulls point to the sheer scale of the revenue engine. "They'll do $700 billion in revenue next year," Deepwater Asset Management founder Gene Munster told CNBC last month. "That affords them the ability to finance some of their customers."

Investors told CNBC the big question is how Nvidia uses its cash and $5 trillion market cap to get more chips into customers' hands — and how much risk it assumes along the way. Nvidia's own framing is that financial support helps customers build the infrastructure that strong chip demand requires, and that more computing capacity helps AI labs develop new models and services, creating further demand for Nvidia products.

For now, the loop depends on one thing staying true: that compute demand keeps meeting Huang's "off the charts" expectations. If it does, the guarantees never get called. If it doesn't, Nvidia discovers the price of becoming the AI economy's lender of last resort.