Nvidia could guarantee a number of the loans OpenAI plans to take out to build its own data centers.

The potential deal, part of a broader $100 billion investment pledge, was reported by the Wall Street Journal.

Such an arrangement could leave Nvidia with billions of dollars in debt obligations if OpenAI cannot pay up.

This September, OpenAI signed a letter of intent with Nvidia to deploy at least 10GW of the company’s systems in its data centers. This could include Nvidia leasing up to five million chips to OpenAI, which could cost $350bn by today’s prices.

At the same time, Nvidia will invest up to $100bn in OpenAI as the new systems are deployed, helping fund the order.

The WSJ reports that the investment was first mooted soon after OpenAI and SoftBank announced the $500bn Stargate data center joint venture in January, but that talks between OpenAI and Nvidia fizzled soon after.

It took a June report from The Information that OpenAI had started using Google’s TPU chips to get Nvidia back on board, with the deal announced just months later.

Despite the commitment, OpenAI has continued to look to broaden its compute base, with the company signing a 6GW deal with AMD in a deal that could give it a ten percent stake in the chip designer just a few weeks later.

A week after that, OpenAI officially unveiled its long-known custom AI hardware project with Broadcom, and said that it would deploy up to 10GW by 2029.

Nvidia, meanwhile, has made several other investments to shore up demand for its chips.

The company is an investor in neoclouds together.ai, Nscale, Lambda, Nebius, CoreWeave, Crusoe, and Firmus. In most cases, Nvidia also rents back a portion of the GPUs it sells to the companies, boosting demand and revenues.

Most notably, with CoreWeave, Nvidia announced that it would buy any of the company’s unsold GPU capacity until 2032.

https://www.datacenterdynamics.com/en/news/nvidia-in-talks-to-guarantee-openais-data-center-loans-report/