Oracle to Supply OpenAI with 2 Million Chips for Next-Gen ChatGPT Training

OpenAI has reached an agreement with Oracle to construct additional Stargate data centers with a total capacity of 4.5 GW. With the ongoing development of the Stargate I site in Abilene, Texas, the total confirmed projects for Stargate now exceed 5 GW, aimed at powering over 2 million chips dedicated to AI tasks. The project is expected to be completed by the end of 2027.

OpenAI emphasizes that the pace of construction is surpassing the U.S. government’s promise to invest $500 billion in 10 GW of AI infrastructure over four years. This initiative is projected to create 100,000 new jobs across various sectors, ranging from electricians and equipment operators to service specialists.

Currently, the first NVIDIA GB200 chips have already been deployed at the Stargate I site, serving to train new versions of ChatGPT and other company models. On July 20, Sam Altman mentioned on X that by the end of the year, OpenAI is expected to have a total of one million chips available:

OpenAI’s nearest competitors are also ramping up their efforts. Meta (designated as extremist in Russia) plans to deploy 1.3 million chips by the year’s end. Meanwhile, Elon Musk’s xAI is set to launch a second supercomputer, Colossus, in Memphis by early 2026, equipped with one million chips, complementing the 200,000 GPUs currently in use.

P.S. You can support me by subscribing to the channel «escaped neural network«, where I explore the creative aspects of AI.