Nvidia’s Jensen Huang wants lenders to finance Nvidia chips as neocloud bonds hit 16%

Jensen Huang has a new audience for his sales pitch, and it is not an AI lab. Nvidia‘s chief executive wants banks and credit funds to treat the company’s chips the way they treat aircraft or office towers: as an asset you can lend against. He made the case on the sidelines of GTC Berlin, in the very week the companies that would do the borrowing watched their own bonds trade like distressed paper.

His diagnosis is that the neoclouds, the firms that buy Nvidia GPUs and rent the capacity to AI developers, have no shortage of customers, only a shortage of cash. Most coverage ran that line as reassurance. Read from the lender’s side of the table, it says something less comfortable. The supplier is asking the credit market to underwrite its own sales.

What Jensen Huang asked lenders for at GTC Berlin

“One of the things that we’re trying to do is to establish Nvidia’s compute as a financeable asset,” Huang told CNBC. “The residual value of our compute, because it’s fungible, it’s so productive, it’s so durable, it is now becoming an asset class.” On demand he was blunter still: “If you look at the fundamentals, the demand for the neoclouds is off the charts. Their capabilities are extraordinary, and so what is missing in the middle is the working capital to fuel that growth, to fund that growth.”

Capital is falling short, in his telling, because so many corners of AI are competing for the same money, and neoclouds need to scale to build what he called “regional excellence”. The remarks follow the $500 billion AI financing plan Nvidia announced in August with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The market did not hear them as reassurance. Wedbush said his comments on neocloud financing helped spark Thursday’s selloff, the Nasdaq’s worst session since June.

Lenders are already pricing the collateral

The word carrying the pitch is “durable”. Collateral is worth what it fetches when the borrower fails, and the credit market is pricing that scenario now. CoreWeave’s 9.75% notes due 2031 fell to around 79 cents on the dollar, a yield near 16%, and its 9.625% issue dropped to 77.5, PitchBook reported.

Durability is also the part of the claim Nvidia is least placed to vouch for. New GPU generations can erode the value of older equipment, and Gartner has put potential impairments on AI infrastructure above $3.5 trillion by the end of the decade. The same week, Nvidia launched DSX POD, a containerized reference design for its Vera Rubin AI factories. The company whose business depends on making last year’s hardware look slow is not a neutral appraiser of how long that hardware holds its value.

Neocloud demand is real; the cash is the gap

Huang is not inventing the demand. CoreWeave chief executive Michael Intrator told staff that a contracted backlog of about $104 billion more than covers the company’s debt, and that markets can “misunderstand and act irrationally”. Nebius operating chief Ophir Nave pointed to roughly $40 billion in long-term customer contracts that could serve as collateral. But a backlog is future revenue subject to delivery, not cash in hand. CoreWeave’s second-quarter interest bill was $640 million against $128 million of adjusted operating income, and D.A. Davidson’s Gil Luria estimates it borrows at roughly 9% while earning returns near 1%.

The equity route is narrowing too. Elice Group, billed as South Korea’s CoreWeave, was nearly 400 times oversubscribed and priced at the top of its range, then fell more than 32% on its KOSDAQ debut. Australia’s Firmus Technologies shelved its listing within three days, and Microsoft‘s cancellation of a UK data centre deal clouded Nscale’s IPO plans. Nebius, which Invezz describes as Nvidia-backed, has just taken a $3.5 billion Citi facility that matures in 364 days. Short money is funding long assets.

Nvidia stands on both sides of the loan

This is the move beneath the news. Nvidia sells the chips, has backed some of the buyers, and now wants a say in how lenders value the collateral. Every GPU financed is a sale booked at a company that reported $96.2 billion in second-quarter revenue, up 106%. If creditors accept Huang’s residual-value story, the risk of the cycle shifts from Nvidia’s order book to their balance sheets. Gartner expects up to 80% of neoclouds to stop operating independently by 2030.

The next tests arrive quickly: CoreWeave’s quarterly results in November, the yields on its bonds, and whether any lender writes Huang’s “asset class” into actual terms. On the GTC stage he supplied the arithmetic himself: a one-gigawatt AI factory costs about $50 billion, could bring in around $40 billion a year, and a three-month delay therefore costs about $10 billion, which he said “will make anybody anxious”.

A lender reads that sum from the other end. A three-month delay is exactly the case a creditor plans for, and the collateral sitting in that idle building is the asset Huang is asking them to value on his word.

Tags: nvidia, Jensen Huang, AI chips, CoreWeave, Nebius