OpenAI has agreed to rent an additional 4.5 gigawatts of data-center capacity from Oracle, deepening the companies’ “Stargate” partnership and underscoring the surging power demands behind generative-AI models. The capacity—roughly enough electricity for four million US homes—will be built across multiple states and is intended to run more than two million AI processors. People familiar with the contract said the deal will bring Oracle about $30 billion in annual revenue, confirming that the unnamed customer behind a blockbuster cloud agreement the company disclosed last month was OpenAI. Including the new tranche, Stargate now has more than 5 GW of capacity under development, with OpenAI’s first site in Abilene, Texas, already handling early workloads on Nvidia GB200 chips. OpenAI and Oracle are evaluating sites in Texas, Michigan, Wisconsin and Wyoming, while expanding the Abilene campus from 1.2 GW to 2 GW. OpenAI estimates the construction and operation of the additional facilities will create more than 100,000 US jobs, although data centers typically employ relatively few full-time staff once they are running. The expansion moves the partners closer to—and potentially beyond—the $500 billion, 10 gigawatt US infrastructure build-out they pledged at the White House in January. SoftBank, part of that original commitment, is not financing the latest phase. The agreement highlights intensifying competition among US tech companies to secure power-hungry cloud infrastructure as Washington seeks to keep domestic AI development ahead of Chinese rivals.