OpenAI’s Computing Capacity Binge

The hardest-working people in tech right now must be the OpenAI team responsible for writing press releases. On Monday—Columbus Day or, if you prefer, Indigenous People’s Day—when much of American business and government was closed, OpenAI put out its sixth big news announcement of the past three weeks. The company unveiled another chip-supply deal, this time with Broadcom, which is working with OpenAI to develop a custom-made artificial intelligence chip for the ChatGPT creator.

For those who’ve given up counting, that makes three such chip-supply deals OpenAI has unveiled in less than a month. These three arrangements collectively aim to create a total of 26 gigawatts of chip capacity over several years (for reference, CEO Sam Altman on Monday described even a 1-gigawatt data center as a “tiny city”). Each of the semiconductor companies involved in the deals—Nvidia, AMD and now Broadcom—is due to start delivering chips to OpenAI by the second half of 2026, which is shaping up to be a monumentally important time for the AI firm. The way Altman and OpenAI President Greg Brockman put it in a video about the Broadcom deal released today, they need this extra capacity to meet demand for their services (and even internal demand among OpenAI staffers).