OpenAI, Oracle and Softbank have announced 5 new data centers under the Stargate project good for 7 gigawatts of capacity. The data centers will be largely powered by Nvidia-based infrastructure.
The companies said the flagship site in Abilene, Texas is operational and can deliver 5.5 gigawatts of capacity built on Oracle Cloud Infrastructure (OCI) and Nvidia's stack. Combined with projects with CoreWeave, Stargate now has 7 gigawatts of planned capacity and more than $400 billion in investment over the next three years.
According to OpenAI, Stargate, announced in January, is well on its way toward securing $500 million and 10-gigawatts committed by the end of 2025.
The AI infrastructure boom is all about a chase for consumer AI with a dash of enterprise, high-performance and scientific workloads. In this week's saga, OpenAI entered a deal with Nvidia for $100 billion in funding where the LLM player builds data centers (on Nvidia GPUs) and then gets paid for every gigawatt deployed. Sam Altman hinted at big plans.
In a nutshell, consumer AI is more like covering a sports story. OpenAI pledges $300 billion to buy AI infrastructure from Oracle. Oracle buys GPUs. Nvidia backstops CoreWeave with purchase guarantees if it has extra capacity. OpenAI also appears to be a big future buyer of Broadcom XPUs (which probably led to the Nvidia deal). Meanwhile, Amazon, Google and AWS all have to buy Nvidia but building their own custom chips for AI workloads.
The Stargate project is the big reason why Oracle's remaining performance obligation surged in its latest quarter. OpenAI and Oracle partnered on up to 4.5 gigawatts of additional Stargate capacity.
Over the next five years, Oracle and OpenAI will develop three sites in Shackelford County, Texas; Doña Ana County, New Mexico; and a site in the Midwest.
In addition to the flagship Stargate site in Abilene, two separate sites will be developed over the next 18 months. Stargate sites in Lordstown, Ohio and another in Milam County, Texas can scale to 1.5 gigawatts of capacity. "The AI race is on and the unit of measure is gigawatts going live. At launch Stargate was questioned as a viable partnership, but it's clear now that they are at the forefront of getting AI capacity online," said Constellation Research analyst Holger Mueller.
According to a report by Bain & Co., $2 trillion in annual global revenue is needed to fund the computing power needed to meet anticipated AI demand by 2030. Bain estimated that even with anticipated AI savings, the world is $800 billion in short to keep pace with demand.
In an earlier blog post, OpenAI CEO Sam Altman said the time to build is now. He said:
"Our vision is simple: we want to create a factory that can produce a gigawatt of new AI infrastructure every week. The execution of this will be extremely difficult; it will take us years to get to this milestone and it will require innovation at every level of the stack, from chips to power to building to robotics. But we have been hard at work on this and believe it is possible."
