OpenAI and Oracle expand Stargate with 4.5 additional gigawatts

OpenAI and Oracle agreed to develop 4.5 additional gigawatts of data-centre capacity for the Stargate project, bringing the pair’s combined contracted capacity, once added to the original Abilene, Texas site, to more than 5 gigawatts able to run over 2 million chips. OpenAI said the expansion put the project ahead of the pace needed to meet the commitment it made at the project’s January launch — $500 billion over four years toward 10 gigawatts of US AI infrastructure — and estimated the added capacity alone would create more than 100,000 construction and operations jobs.

The announcement also confirmed that parts of the original Abilene site were already operational, with Oracle delivering Nvidia GB200 racks and running early training and inference workloads — a concrete marker after months in which the huge headline figures from January’s launch had drawn scepticism about whether comparable capital was actually being deployed, and not just pledged.

The deal was one instalment in a buildout that continued expanding through the rest of the year, with OpenAI, Oracle and SoftBank adding five more Stargate sites that September and the project’s committed capacity growing further into the autumn. Each expansion reinforced the same underlying wager: that AI progress would continue to be gated by data-centre and power capacity rather than by algorithms, and that locking in gigawatts years in advance was worth the financial exposure that implied.