Nvidia’s plan to invest up to $100bn in OpenAI to train and run its latest AI models has stalled due to doubts inside the chip giant, people familiar with the matter said.
The pair unveiled the agreement in September at Nvidia’s Santa Clara, California headquarters, pitching a memorandum of understanding rather than a finished contract.
Nvidia agreed to build at least 10 gigawatts of computing power for OpenAI and invest up to $100bn to help pay for it, with OpenAI leasing the chips.
At the time, the ChatGPT maker expected negotiations to wrap up within weeks, but talks have not moved beyond early stages, the people said.
Even with the September pact on ice, Nvidia is still lining up a separate big investment in OpenAI as the AI outfit tries to raise $100bn in new equity.
Amazon.com and SoftBank are among those discussing tens of billions of dollars for the round, which could total $100bn. In Taipei on Saturday, Nvidia’s chief executive, Jensen Huang, told reporters: “Nvidia would absolutely be involved in OpenAI’s latest funding round… We will invest a great deal of money, probably the largest investment we’ve ever made,” he said.
Asked if that meant more than $100bn, Huang replied: “No, no, nothing like that.”
Away from the microphones, Huang has been telling industry associates for months that the original $100bn agreement was nonbinding and not finalised, people familiar with the matter said.
He has privately criticised what he calls a lack of discipline in OpenAI’s business approach and has expressed concern about competition from Google and Anthropic.
OpenAI spokesman said: “Our teams are actively working through details of our partnership. Nvidia technology has underpinned our breakthroughs from the start, powers our systems today, and will remain central as we scale what comes next,” an OpenAI spokesman said.
An Nvidia spokeswoman said the company has been OpenAI’s preferred partner for the past 10 years and looks forward to continuing to work with OpenAI.
OpenAI is laying the groundwork to go public by the end of 2026 and has spent much of the past year scrambling to lock down computing capacity.
