Nvidia without Huang: who guarantees the guarantor?

Posted on: 13 November 2026

On Wednesday 11 November Nvidia's board accepted, with immediate effect, the resignation of Jensen Huang, who co-founded the company in 1993 and had run it as chief executive ever since. Until a successor is found his job has passed to Colette Kress, chief financial officer since 2013 after a long career at Microsoft and Cisco, who will keep her old role as well. Harvey Jones, a director since 1993 and a former chief executive of Synopsys, the company whose software is used to design chips, has become chair of the board. The announcement points out that he is the first chair in Nvidia's history, which means that for thirty-three years the job of overseeing the founder never had a holder separate from him. The resignation came a few days after the board opened an internal inquiry, when Huang admitted a relationship with Alice Huang, a former executive of Megaspeed, a Singapore data centre company accused of moving Nvidia chips into China. The filing with the Securities and Exchange Commission, the American markets regulator, gives no reason for his departure and there is no need to invent one, because the part of this story that matters has nothing to do with Huang's private life.

On Thursday the Financial Times reported that Nscale has told prospective investors it needs further funding for its 2027 commitments, after ordering the Nvidia chips required for its contract with Anthropic. I wrote about the company last week in connection with Microsoft, so a brief reminder will do. Nscale is a British neocloud, a company that builds data centres and rents out their computing capacity to artificial intelligence businesses. It hopes to list in New York at a valuation of up to $35bn and its largest contract, signed in August with Anthropic for the first 460-megawatt building on a campus in West Virginia, is worth up to $44.6bn. Its own listing prospectus warns, however, that the contract depends on Nscale finding the money to build that building, so much so that Anthropic can walk away if the financing does not arrive by a deadline.

The two stories look as if they belong on different pages of the paper, yet Nvidia sits on almost every side of that contract. It supplies the Vera Rubin chips that will go into the building and it is a shareholder in Nscale, which in September secured Nvidia's commitment to put in $1bn through convertible notes, loans that can turn into shares, due to arrive in mid-November. Since late October Nvidia has also been negotiating to put its own balance sheet behind billions of dollars of debt for the construction and the electricity supply of the campus. Huang had called Nscale a national champion for the UK, and when asked about the delays on its building sites he replied, according to the Financial Times, that they should work around the clock and try less sleep, as he does. That is the language of someone who treats a customer as his own creation.

Nscale is not an isolated case. In the first four months of 2026 Nvidia had already committed more than $40bn to stakes in the sector, $30bn of it in OpenAI, while according to the Financial Times its investments in unlisted companies had risen by the end of April to $42bn, from $3bn a year earlier. Over the summer the Wall Street Journal reported talks on guaranteeing up to $250bn so that OpenAI could lease a ten-gigawatt campus in Ohio, and at the end of July Nvidia's credit default swaps, the contracts investors buy to insure themselves against a borrower's default, hit a record. Huang has always rejected the charge that Nvidia finances its own demand, calling it ridiculous and arguing that the sums invested are small compared with the business they generate. On the equity stakes he is probably right. On the guarantees the argument changes, and this is where a story from almost a century ago becomes useful.

In 1932, in the depths of the Great Depression, General Electric set up the General Electric Contracts Corporation so that American families could buy its refrigerators in instalments. It was a sales tool and nothing more, since the banks were not lending and without credit the warehouses stayed full. Over the following decades that credit office became GE Capital and stopped financing only the group's own customers, moving into lending to companies across half the world and into commercial property. The secret lay in the triple A, the highest grade a rating agency can give a borrower, which GE had held since 1956 and which allowed it to raise money through commercial paper, IOUs lasting from a few days to a few months, at almost the same cost as a government, and then lend it on at much higher rates. The machine grew above all under Jack Welch, GE's chief executive from 1981 to 2001. By October 2008 GE Capital managed $650bn of assets, owed $550bn and produced around half of the group's profits.

That autumn GE's turbines and jet engines were still selling, yet the market stopped seeing GE as an industrial company and began pricing it as a bank. A rumour that it could not roll over its commercial paper was enough for the shares to drop 10 per cent all of a sudden while its credit default swaps jumped. GE had to ask Warren Buffett, the investor who runs Berkshire Hathaway, for $3bn in exchange for preference shares paying 10 per cent a year, then turned to the scheme under which the FDIC, the federal agency that insures American bank deposits, guaranteed the debt of financial institutions. In February 2009 it cut its dividend for the first time since 1938 and in March it lost its triple A. The product had not changed. What had changed was the way the market judged the signature of the company selling it.

The mechanism can be stated in a few words. As long as financing customers remains an accessory to selling, it is a commercial tool. Once demand starts to depend on that financing, the creditworthiness of the seller becomes part of what it sells. At that point the market stops valuing the company by its best part and values it by its riskiest, because in a crisis that is the part which decides who survives.

Nvidia is not GE and the difference needs saying plainly. GE Capital borrowed for days to lend for years, so it lived on the daily confidence of those who rolled over its debt, whereas Nvidia finances its customers mainly by buying their shares with its own cash, which the Financial Times put at $80bn of liquid assets. An equity stake that goes wrong gets written down and forgotten. A guarantee works the other way round. It does not leave the bank account on the day it is signed and it does not weigh on the accounts until someone calls it, but that day always arrives together with all the others, when the weakest customers stop paying and the credit markets close. According to the Wall Street Journal, neoclouds such as CoreWeave and Nscale have in recent weeks been effectively shut out of the bond market, which means Nvidia's guarantee is no longer extra support but the very condition of the loan. The banks would lend to Nscale because Nvidia stands behind it, just as the markets lent to GE Capital because GE stood behind it.

In a chain like this, though, Nvidia's balance sheet was not the only guarantee, because Huang's word counted for at least as much. When he was asked in February about the $100bn that had been talked about for OpenAI, he replied that it had never been a commitment and that Nvidia would invest one step at a time, which means the figure lived in his intentions rather than in a contract. While the founder stayed in place this elasticity was an advantage, since customers and lenders knew that behind every deal there was a person who wanted it and had the power to see it honoured. Now the company is run, if only for the time being, by the person who has to put those risks on the balance sheet. A founder reads a guarantee as market share defended, while a finance director reads it as a contingent liability with her own signature underneath, and neither reading is wrong. By leaving Kress with both roles the board has put the balance sheet at the top of the organisation chart. At GE, too, the machine Welch built was tested under his successor, Jeffrey Immelt, who had inherited along with the company a triple A that no longer depended on the turbines.

I may be wrong, and there is a simple way to find out. If in the coming weeks Nvidia signs the Nscale guarantee on the terms discussed at the end of October, the strategy of financing customers belongs to the company and not to its founder, and my hypothesis falls. If instead the guarantee shrinks or arrives loaded with new conditions, we will have evidence that part of the demand for Nvidia's chips depended on the will of a single man. Meanwhile, according to the Daily Telegraph, Nscale is considering postponing its listing, with some investors convinced it is now worth less than half the $35bn it was seeking.

Investors should not count a neocloud's order book as security when its main contract can be cancelled in precisely the scenario in which it would be needed, that is, when the financing fails to arrive. Board members should not let commitments on which part of the revenue depends live on the word of a single person, because on the day that person leaves the commitments stay while the confidence that made them credible has to be negotiated all over again.

In the meantime Anthropic is listing this very week with demand for its shares six times greater than the offer, while the company that is supposed to build its data centre is still looking for someone willing to lend it the money.