Analysis
I do not think Nvidia is running a fraud. I think Nvidia is running a strategy that works beautifully in an up-cycle and becomes unreadable in a down-cycle, and the industry has stopped distinguishing between those two statements.
Here is the mechanism, stripped of the noise. Nvidia put $1.5 billion into SB Energy, which is funding a data center that OpenAI will fill with Nvidia GPUs, Axios reported. Nvidia books revenue on the chips and holds an equity claim on the infrastructure those chips sit in. It has done versions of this with OpenAI, with xAI, with Mistral, and through NVentures across roughly $1 billion of startup investments in 2024 alone. Jensen Huang calls it a virtuous cycle. He is not wrong that it accelerates buildout. He is wrong that the label is the interesting question.
The interesting question is what fraction of Nvidia's revenue would exist if Nvidia stopped writing checks. Nobody outside Santa Clara can compute that number, and Nvidia's disclosures do not make it computable. What we do know, from this quarter's filings, is that supply and capacity commitments jumped from $119 billion to $279 billion, equity investments sit at $25 billion, and free cash flow fell from $48.5 billion to $21.3 billion quarter over quarter. Those are the numbers of a company converting cash into claims on its own ecosystem at speed.
“It has done versions of this with OpenAI, with xAI, with Mistral, and through NVentures across roughly $1 billion of startup investments in 2024 alone.”
The reason this matters to founders and GPs rather than just to public-market analysts: Nvidia's capital is now a material input to startup valuations in AI infrastructure. When Nvidia participates, the round prices higher, closes faster and signals supply access. That is real value. It is also a correlated exposure sitting inside a lot of otherwise diversified portfolios, and almost nobody is modeling it as one. If Nvidia's investment pace slows -- not stops, slows -- a set of infrastructure companies discover simultaneously that their last mark was partly a function of their supplier's balance sheet.
Room for disagreement: The strongest counter-argument is that this is what every dominant platform company has done at the start of a compute cycle, and it has usually been right. Intel funded the PC ecosystem through Intel Capital and made the market bigger for everyone, including its competitors. Nvidia is not lending money to shell companies to buy chips; it is funding power, land and real data centers that will be occupied regardless of who supplies the silicon. And OpenAI, Google, Amazon and Microsoft are all shipping custom accelerators -- OpenAI claims its Jalapeno chip beats Nvidia hardware on some workloads -- which is hard to square with a story where Nvidia is manufacturing fake demand. If the buildout is genuinely supply-constrained, financing supply is just competent capital allocation.
Where I land: the strategy is defensible and the disclosure is not. Nvidia should break out revenue from entities in which it holds an equity position or a financing commitment. Until it does, every bull and bear argument about AI demand is being made with the most important variable missing, and the $279 billion commitment number gives that variable a lot of room to move.