OpenAI unhappy with Nvidia chips and looking for alternatives - Reuters

OpenAI is looking for alternatives to Nvidia's latest chips from 2025, expressing dissatisfaction with their performance in the task of inferencing - generating responses to user queries, Reuters reported, citing eight sources. The ChatGPT developer's change of strategy creates tension between the two leaders of the artificial intelligence boom: while Nvidia chips dominate the AI model training segment, inferencing is becoming a new front of technological competition, forcing OpenAI to diversify suppliers, the agency said.

OpenAI is not satisfied with the speed at which Nvidia's hardware can deliver responses to ChatGPT users for certain types of tasks, such as software development and AI interaction with other software, seven sources told Reuters. OpenAI has been in talks with processor startups Cerebras and Groq to supply chips that provide faster responses, two sources told the agency.

Nvidia processors remain the standard for AI training, but for inference and reasoning tasks, OpenAI is looking for alternatives with more memory built directly into the chip (SRAM). This architecture provides the high speed needed for mass query processing - unlike Nvidia's solutions with external memory (DRAM), which slows down response. This problem has become particularly noticeable for the Codex code generation tool that OpenAI is actively marketing: the startup attributes the product's weaknesses to the hardware limitations of Nvidia's architecture, according to one of Reuters' sources.

Upon learning of OpenAI's intentions to switch suppliers, Nvidia attempted to acquire Cerebras and Groq. Cerebras rejected the offer for the sake of a direct contract with OpenAI, but the latter's negotiations with Groq were thwarted: Nvidia took over the asset, signing a $20 billion license agreement in December 2025. As a result of the deal, Nvidia hired the startup's chip development team, while Groq itself refocused on selling cloud software, according to a Reuters article.

After the Reuters piece was published, OpenAI head Sam Altman wrote on social network X that Nvidia makes "the best AI chips in the world" and that OpenAI hopes to remain "a giant customer [of the company] for a very long time."

Participants of over-the-counter trading in the U.S. on February 3 ignore the negative on Nvidia: the company's securities on the Blue Ocean ATS are still growing by 0.5%, correcting after yesterday's fall. The day before, Nvidia shares on the Nasdaq exchange fell by 2.9%. Investors reacted with a sell-off to the publication in The Wall Street Journal about the suspension of the chipmaker's plans to invest up to $100 billion in OpenAI, as well as to the clarification of Nvidia CEO Jensen Huang that this amount "was never a commitment".

Wedbush Securities analyst Dan Ives noted in a note to clients on Feb. 2 that investors are nervous about Nvidia's round-robin financing deals and concerns that OpenAI is becoming a "too big to fail" company. Nevertheless, Ives maintained an Outperform rating ("above market," buy recommendation) for Nvidia shares with a target price of $230 - 24% above the Feb. 2 closing price of $185.6, Investor's Business Daily reported.

This article was AI-translated and verified by a human editor