There have been a billion pixels sacrificed to make the circular finance argument, including here on the blog (though we are still the only ones to invoke the ourobouros)
Here is a look at things from a different angle.
From Barron's, September 26:
Artificial-intelligence leader Nvidia generated $112 billion in free cash flow in the past six quarters—and more is coming, at an increasing rate. For companies that have reached this rarefied air, the inevitable question is what to do with all that money. Apple, for one, decided to buy back billions worth of stock and reinvest in its business. For Nvidia, the decision is to backstop the AI investment boom, where it remains the primary winner.
Nvidia, to be sure, has also done buybacks and reinvestments. In the past six quarters, the company has spent $57 billion on share repurchases. In August, the board authorized another $60 billion. In its recently reported second quarter, research and development expenses were up 39% from the year before.
Even with that spending, Nvidia’s cash and short-term investments have risen 11 quarters in a row, now up to $57 billion. In the past, companies would also use their cash for mergers and acquisitions, but the regulatory environment doesn’t favor large deals anymore, as seen by Nvidia’s failed effort to buy Arm Holdings from 2020 through 2022. Today, Nvidia faces the same dilemma as Apple did in 2012, when it had over $100 billion in cash and securities. The companies have more than they know what to do with, and it’s piling up every day.
Nvidia’s AI backstop strategy began in 2023, when CEO Jensen Huang identified a small AI cloud provider called CoreWeave, which at the time had $16 million in annual revenue. Since then, he has helped it grow into a company with a $80 billion market value. Huang didn’t want to be dependent on a few “hyperscaler” cloud companies like Amazon Web Services, Microsoft Azure, and Google Cloud, so he created a new customer by seeing that CoreWeave got early allotments of Nvidia’s AI servers.
CoreWeave says it doesn’t get any special treatment from Nvidia. “What makes our relationship strong is execution,” a company spokesperson says. “CoreWeave has built the scale, infrastructure, and technical expertise to be a reliable partner, which is why customers and suppliers choose to work with us.”
But there’s no denying that Nvidia is CoreWeave’s most important supplier and a key customer. It also owns about 5% of the company’s stock. Huang’s latest move came two weeks ago, when CoreWeave signed a $6.3 billion deal for Nvidia to buy any excess AI data center capacity, should demand from CoreWeave’s multiyear contracts falter. CoreWeave’s business model requires rapid growth and massive debt to fund it. Nvidia is promising to backstop the effort. In so doing, it’s giving others more confidence to lend CoreWeave money for its ambitious growth strategy.
That same week, during a trip to the United Kingdom, Huang seemed to christen a new British CoreWeave-style company called Nscale. Like the CoreWeave of yesteryear, Nscale has little in the way of revenue. According to FactSet, the company raised one round of funding worth $155 million in December.
But now Nscale has the full faith and credit of Nvidia, and a $1.1 billion round of financing from a group that includes Nvidia. It also sports new contracts with Microsoft and OpenAI for U.K. data center deployments, and it has plans to expand internationally.
In a video posted to LinkedIn, Huang “signed” the Nscale partnership on a bottle of Johnny Walker Blue scotch, writing “zero to 50 billion,” in reference to Nscale’s growth. “I’ve never seen a start-up take off like that before,” Huang said to Nscale CEO Josh Payne. “I think you’re on the Starship.”
“We couldn’t do so without your support,” Payne replied.
But the CoreWeave and Nscale deals pale in comparison to what came this past week, when Nvidia committed up to $100 billion to OpenAI, as part of OpenAI’s plan to build 10 gigawatts of data centers for itself.
Data center developers have taken to sizing their projects by the power consumed, since the massive energy requirements are a limiting factor. Ten gigawatts would consume all of the power output from the most recently built U.S. nuclear reactor, Georgia’s Vogtle 4, which opened in 2024, plus another eight just like it....
....MORE
