Nvidia reportedly financed AI startups to buy its chips as Huang warned of an 'avalanche'
A Wall Street Journal investigation reports Nvidia offered billions in financial guarantees to AI startups and small cloud providers that could not otherwise afford its chips — and that its former chief executive privately feared the flood of prepaid silicon heading its way.
The News
A Wall Street Journal investigation published on November 13 reports that Nvidia offered billions of dollars in financial guarantees to AI startups and small cloud and datacenter providers that could not otherwise afford its graphics chips. A financial guarantee is a promise to cover the bill if the buyer cannot pay — in this case, Nvidia effectively stood behind customers so they would buy up chips it had already committed to purchase from TSMC, the Taiwanese company that manufactures its chips. The report was relayed by VideoCardz, and by TechPowerUp and Wccftech, from the Journal's paywalled original — all three draw on the same WSJ investigation, so they are relays, not independent corroboration.
The report says former chief executive Jensen Huang privately described the incoming prepaid wafers and memory as an 'avalanche' — a flood of silicon Nvidia had already paid for, or promised to pay for, that was heading its way whether or not buyers were ready. At a meeting in August with chief financial officer Colette Kress, who is now interim chief executive, Huang resisted reining the deals in. Kress has since cut the proposed OpenAI compute backstop roughly in half, from about 8 gigawatts and up to $250 billion to about 4.25 gigawatts and up to $125 billion. A gigawatt is a measure of electrical power — one gigawatt is roughly enough to power 750,000 homes — and datacenter deals are often sized this way because the chips need enormous amounts of electricity to run.
Wccftech, relaying the report, adds that Nvidia has contacted TSMC about cutting upcoming chip orders — a claim Wccftech itself reports without independent verification. And the largest cloud companies — the hyperscalers that buy the most chips — were quietly reducing their reliance on Nvidia chips, the Journal reported. Nvidia shares closed Friday at $162.85, down 2.01%. Microsoft closed at $523.88, up 0.17%; Alphabet at $295.77, down 0.98%; Amazon at $266.66, down 0.29%; and Meta at $641.69, up 2.90%. Nvidia reports third-quarter results for fiscal 2027 on Tuesday, November 17, at about 1:20 p.m. Pacific time — its first results under Kress.
Nvidia was lending its customers the money to buy its own chips — the sales looked real, but the demand underneath them is the question Tuesday has to answer.
Why It Matters
This matters because it changes the question at the heart of the AI boom: how much of the demand for Nvidia's chips is real, and how much was helped along by Nvidia's own money? A guarantee lets a sale happen that otherwise would not. That is not fraud — the chips are real, the customers are real, and Nvidia's filings list about $530 billion in commitments and guarantees, per SemiAnalysis — but it does mean the order book looked stronger than the buyers' own balance sheets could support.
The 'avalanche' image matters too. Wafers are the silicon discs chips are cut from, and Nvidia pre-pays TSMC for them months ahead. If demand cools, those prepaid wafers become inventory Nvidia has to sell at a discount or write down — an accounting term for admitting something is worth less than you paid. Contacting TSMC to wind down orders is the clearest sign yet that Nvidia itself expects to need fewer chips than it planned for.
The leadership angle is the third piece. The OpenAI backstop was halved on Kress's watch, and Tuesday's earnings are her first as interim chief executive. Investors will be listening for whether she keeps tightening — smaller guarantees, fewer prepaid orders — or whether she argues the deals were the right call at the time. Either way, the era in which Huang's word alone settled the question is over.
Analyst's Take
Our call is Mixed. On the one hand, the guarantees worked: chips that might have sat in a warehouse found buyers, revenue kept flowing, and the AI buildout kept its pace. A supplier helping its customers buy is an old industrial trick, and in a boom it can look like genius. On the other hand, a company lending its customers the money to buy its own products is flattering its own sales — and if those customers cannot pay on their own, the demand underneath is softer than the headline numbers suggest. Both things are true at once, which is why this is a Mixed call, not a Bearish one.
The read is different for the hyperscalers. Microsoft, Alphabet, Amazon and Meta are the ones building their own chips, and every chip they design themselves is one they do not have to buy from Nvidia at Nvidia's prices. That is why the per-ticker view is bullish for the hyperscalers even though the story is Mixed for Nvidia. The catch is time: custom chips take years and billions of dollars to design and build, so in the near term the hyperscalers are still Nvidia customers — just customers with an exit plan.
Reader poll: will Nvidia's Tuesday earnings calm the worry, or add to it? Vote: Buy the dip, Hold, Sell, or Wait for earnings.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is a financial guarantee?
It is a promise to cover the bill if the buyer cannot pay. In this case, Nvidia effectively stood behind AI startups and small cloud providers so they would buy chips it had already committed to purchase from TSMC — the chips were real and the customers were real, but Nvidia's promise is what made some of the sales possible.
What did Jensen Huang mean by an 'avalanche'?
According to the Journal, Huang privately used the word for the incoming prepaid wafers and memory — silicon discs that chips are cut from, and memory chips, that Nvidia had already paid for or promised to pay for. The image is of a flood of silicon arriving whether or not buyers were ready for it.
What changed under Colette Kress?
Kress, the chief financial officer who is now interim chief executive, cut the proposed OpenAI compute backstop roughly in half — from about 8 gigawatts and up to $250 billion to about 4.25 gigawatts and up to $125 billion. Nvidia has also contacted TSMC about winding down upcoming chip orders, according to the report.
When does Nvidia report earnings?
Nvidia reports third-quarter results for fiscal 2027 on Tuesday, November 17, at about 1:20 p.m. Pacific time — its first results under interim chief executive Colette Kress.
👁
Source: VideoCardz