Nvidia is, in crucial ways, nothing like Enron – the Houston energy giant that imploded through multibillion-dollar accounting fraud in 2001. Nor is it similar to companies such as Lucent or Worldcom that folded during the dotcom bubble.
But the fact that it needs to reiterate this to its investors is less than ideal.
Now worth more than $4tn (£3tn), Nvidia makes the specialised technology that powers the world’s AI surge: silicon chips and software packages that train and host systems such as ChatGPT. Its products fill datacentres from Norway to New Jersey.
This year has been an exceptional one for the company: it has struck at least $125bn in deals, ranging from a $5bn investment into Intel – to facilitate its access to the PC market – to $100bn invested in OpenAI, the startup behind ChatGPT.
But even as those deals have fuelled surging stock prices and paved the way for chief executive Jensen Huang’s energetic world tour, doubts have emerged about how Nvidia does business, especially as it has become increasingly central to the health of the global economy.
The start of these concerns has been the circular nature of many of its deals. These arrangements resemble vendor financing: Nvidia lending money to customers so they can buy its products.
The largest of these is its deal with OpenAI, which involves Nvidia investing $10bn into the company each year for the next 10 years – most of which will go to buying Nvidia’s chips. Another is its arrangement with CoreWeave, a company that provides on-demand computing capacity to big AI firms, essentially leasing out Nvidia’s chips.
The circularity of these deals has drawn comparisons with Lucent Technologies, a telecoms company that also aggressively lent money to its customers, only to overextend itself and unravel in the early 2000s. Nvidia has aggressively rebutted suggestions of any similarity, saying in a leaked recent memo that it “does not rely on vendor financing arrangements to grow revenue”.
James Anderson, a renowned tech investor, describes himself as a “huge admirer” of Nvidia, but said this year that the OpenAI deal presented “more reason to be concerned there than before”.
He added: “I have to say the words ‘vendor financing’ do not carry nice reflections to somebody of my age. It’s not quite like what many of the telecom suppliers were up to in 1999-2000, but it has certain rhymes to it. I don’t think it makes me feel entirely comfortable from that point of view.”
Other high-profile recent deals include the tech firm Oracle spending $300bn on datacentres for OpenAI in the US – with the ChatGPT developer then paying back roughly the same amount to use those datacentres. In October, OpenAI and the chipmaker AMD signed a multibillion-dollar chip deal that also gave OpenAI the option to buy a stake in the Nvidia rival.
There is also a deal with CoreWeave where, along with a commitment to buying $22bn of data centre capacity from the cloud provider, OpenAI is receiving $350m in CoreWeave stock. Asked this month about circularity in the AI industry, the chief executive of CoreWeave, Michael Intrator, said: “Companies are trying to address a violent change in supply and demand. You do that by working together.”
All these moves form part of OpenAI’s $1.4tn bet on computing capacity to build and operate models that, it argues, will transform economies – and make back that expenditure. OpenAI argues that, while the Nvidia and AMD deals have an investment component, it only kicks in once the chips have been bought and deployed, while the investments themselves create aligned incentives to build out AI infrastructure at huge scale.
Nvidia has also used structures called special-purpose vehicles (SPVs) in financing deals. The best-known example is the SPV linked to Elon Musk’s xAI: an entity into which Nvidia invested $2bn, money that will be used to buy Nvidia’s chips.
This drew comparisons with Enron, which used SPVs to keep debts and toxic assets off its balance sheets, convincing investors and creditors that it was stable while concealing ballooning liabilities.
Nvidia has also strongly denied that it is like Enron: in the same leaked memo where it discussed Lucent, it said its reporting was “complete and transparent” and “unlike Enron” it “does not use special-purpose entities to hide debt and inflate revenue”.
The journalist Ed Zitron, a noted sceptic of the AI boom, agrees that Nvidia is not like either company. Unlike Lucent, it does not appear to be taking on a great deal of debt to finance its circular deals, he says, and most of the customers it is supporting are not as obviously risky as Lucent’s dotcom bubble partners. And it isn’t like Enron, Zitron argues, because it’s being fairly transparent about its own complex, off-balance sheet deals.
So what could warrant a comparison? Nvidia “is not hiding debt, but it is leaning heavily on vendor-financed demand, which creates exposure if AI growth slows,” says Charlie Dai, an analyst at the research firm Forrester. “The concern is about sustainability, not legality.”
Essentially, whether Nvidia is able to stick the landing depends on whether AI really takes off, generating billions for its corporate users and putting companies such as OpenAI, Anthropic and CoreWeave – Nvidia’s customers – firmly in the black, and able to keep buying its systems. That possibility alone is debatable. If this does not happen, says Dai, Nvidia “could face write-downs on equity stakes and unpaid receivables”: meaning, it could lose a lot of money and its stock price could then tank.
Approached for comment, an Nvidia spokesperson referred the Guardian to remarks its chief financial officer, Colette Kress, made to investors in early December. Kress said they were not seeing an AI bubble, instead gesturing at trillions of dollars of business that lie ahead for Nvidia in the next decade.
In particular, Kress argued that Nvidia’s recent – massive – deals are just the start for the company, and the real money will be made in the coming years, largely through replacing almost all the chips in existing datacentres with its products.
There is another complexity, which is that Nvidia’s health – and therefore the health of the entire global economy – also depends on whether AI takes off in time for Nvidia and its customers to service the debt from their huge datacentre buildouts and significant capital expenditures.
Add to this a final category of concern: recent, big-ticket deals with countries such as South Korea and Saudi Arabia, worth multiple billions of dollars, whose terms are opaque. In October, Nvidia said that it would supply 260,000 of its Blackwell chips to South Korea’s government and South Korean companies. The value of this deal was not disclosed, but is estimated to be in the billions.
Likewise with Saudi Arabia. Its government-owned AI startup, Humain, has committed to deploying up to 600,000 Nvidia chips: when that deployment will involve actual purchases, and at what price, is again undisclosed. Nvidia has a number of other strategic partnerships like this – with Italy, with the French AI champion Mistral and with Deutsche Telekom, for example – all involving thousands of chips and unknown sums.
Governments are likely to pay. There’s nothing circular about a sovereign partnership with Germany. But the deals mean more – quite large – uncertainties nested within a straining web of commitments that require massive capital outlay, and rely on ambitious assumptions about the economy undergoing a revolution in the next years.
“They concentrate risk in a few big customers,” says Dai. “If execution delays occur, Nvidia’s revenue recognition and cashflow could be affected.”