Nvidia adds $100 bln to AI’s load-bearing startup

NEW YORK, Sept 22 (Reuters Breakingviews) - The ouroboros – a dragon eating its own tail, signifying eternal recurrence – is as apt a symbol for the modern mysticism of artificial intelligence as it was for ancient Egypt. Consider Nvidia’s NVDA.O agreement, announced Monday, to invest up to $100 billion in OpenAI to help the ChatGPT maker build data centers that will, of course, use its chips. The logic of such circular deals eventually becomes self-sustaining: OpenAI is so intertwined with a series of suppliers that their fates are becoming inextricable.

Building out cutting-edge server farms to juggle the computational demands of AI isn’t cheap. A facility consuming a gigawatt of power might cost up to $50 billion. OpenAI will need more than that to train new models and run existing offerings as it chases ten-fold growth to $150 billion of revenue by 2029.

Yet the company run by Sam Altman expects to incinerate some $115 billion of cash by that point, according to The Information. In order to fuel its wildly unprofitable growth, OpenAI has signed a bewildering array of deals – like its partnership with Microsoft MSFT.O, in part for cloud-computing capacity – to keep it going. Sure, most tech giants have similarly hitched their fortunes to AI. But consider OpenAI’s relationships, and how they might affect Nvidia.

Nvidia already owns a chunk of OpenAI. This agreement will swap chip orders for equity, increasing its economic ownership. In March, data center operator CoreWeave CRWV.O, a big Nvidia customer, announced a contract to provide OpenAI with up to $11.9 billion of cloud capacity. As part of that deal, CoreWeave issued $350 million of stock to the ChatGPT developer. Nvidia is also an investor in CoreWeave, and has agreed to buy cloud capacity that CoreWeave might not be able to sell.

Put simply, if OpenAI’s ambitions are cut short, Nvidia would see a dramatic reduction in demand from more than one source, as well as a potential hit to the value of intertwined investments throughout the AI ecosystem.

Vendors aided the purchase of equipment back during the telecommunications bubble around the turn of the millennium, too. Granted, fiber-optic cable layers and wireless firms racked up $2 trillion in leverage globally, a wild free-for-all. And Nvidia generates vast sums of cash on its balance sheet, expected to swell to half a trillion dollars by 2030, according to Visible Alpha data. The promise of “artificial general intelligence” – an AI so powerful it can effectively replace human effort – heralds vast, if hazy, further value, if only model-makers can spend enough cash to develop it. As such, investors cheered the party’s continuation, sending Nvidia’s market value up by $150 billion. That’s yet more incentive to keep feeding, and overstuffing, AI’s ouroboros.

CONTEXT NEWS

Nvidia said on September 22 that it had agreed to invest up to $100 billion in OpenAI. Under a partnership between the two companies, the money will be invested in stages as OpenAI builds out data centers based on Nvidia systems, with at least 10 gigawatts of capacity envisioned under the agreement.

The partnership will complement other agreements between the companies and a network of partners including Microsoft, Oracle and SoftBank, Nvidia said.

The deal will involve two separate but intertwined transactions, Reuters reported citing a source familiar with the matter. OpenAI will pay Nvidia in cash for chips. Nvidia will then invest in OpenAI and receive non-controlling shares.

On September 15, CoreWeave said that it had signed a $6.3 billion order with Nvidia, under which the chipmaker will purchase cloud capacity the data center operator cannot sell to customers.

In March, CoreWeave announced a contract to provide cloud computing capacity to OpenAI valued at up to $11.9 billion. As part of that deal, CoreWeave issued $350 million of stock to the ChatGPT developer.