OpenAI has been in talks with chip designers, including Broadcom, to discuss the possibility of developing a new AI server chip, and according to The Information, OpenAI CEO Sam Altman has been leading the discussions. Broadcom was instrumental in the development of Google's TPU AI chip.
A source added that the server chip is expected to compete with Nvidia's hardware, but admitted that such a product is still far from fruition, while its development could be detrimental to Huang's, currently OpenAI's largest chip supplier.
The Information further reports that Altman is looking into plans to build data centers to house chips. The data centers would be backed by one or more companies founded by Altman. The investors' money would be used to buy the necessary real estate, power supplies and specialized AI chip servers, which OpenAI would commit to renting.
Last year, DCD exclusively reported that OpenAI had hired Richard Ho, former director of chip development at Lightmatter and head of TPU at Google, as director of hardware. Also in 2023, Altman's people asked TSMC for talks to get a chip design project off the ground.
In March 2024, job postings from OpenAI revealed that the company was looking for a team to "co-develop future hardware from multiple vendors to achieve programmability and performance"
It had previously been revealed that Altman was pushing for OpenAI to develop its own AI chips. Before he was abruptly fired and rehired in November 2023, Bloomberg reported that the CEO was looking for investment to build an AI chip company, codenamed "Tigris", which may focus more on chip design.
In January this year, the same media outlet claimed that Altman had been in contact with Middle Eastern figures to discuss investment opportunities and that he was trying to raise between $5 billion and $7 billion for his AI chip company.