When OpenAI first rolled out ChatGPT in late 2022, few could’ve guessed the astronomical compute demands that would follow. Fast‑forward to July 2025, and the company behind the viral chatbot has teamed up with Oracle to tack on an eye‑watering 4.5 gigawatts (GW) of new data‑center capacity in the U.S. — all under the banner of what it calls “Stargate.”

Bloomberg reporters Shirin Ghaffary and Brody Ford broke the news that OpenAI and Oracle will co‑develop 4.5GW of additional U.S. data‑center power, bringing the total Stargate buildout to more than 5GW once you count the 1GW Abilene, Texas, campus already underway. Together, these sites will eventually host over two million AI chips running training and inference workloads at industrial scale.

“Stargate is an ambitious undertaking designed to meet the historic opportunity in front of us. That opportunity is now coming to life through strong support from partners, governments, and investors worldwide—including important leadership from the White House,” OpenAI said in its press release, referencing its January pledge at the White House to invest $500 billion into 10GW of AI infrastructure over four years.

While OpenAI hasn’t publicly named the new sites, Bloomberg confirmed that Texas, Michigan, Wisconsin and Wyoming are all in contention. Site surveys are already underway, but if history is any guide, at least one of these states will roll out the welcome mat in the form of tax breaks:

• Texas offers a sales‑tax exemption on equipment and electricity for data centers that invest at least $200 million and span 100,000 sq ft.
• Midwestern states such as Michigan and Wisconsin often waive personal‑property taxes on servers, while Wyoming is known for generous sales‑tax holidays on high‑value manufacturing projects.

These incentives, combined with abundant land and affordable power, have turned the U.S. heartland into a hotbed for AI infrastructure.

Construction is already humming at Stargate I in Abilene. Oracle began delivering NVIDIA GB200 racks last month, and OpenAI reports it’s running early experiments on next‑generation models there. Those GB200s pack incredible density for both training and inference, making them the workhorses of any large‑language‑model pipeline.

And yes, the meme that “data centers create few jobs” only tells half the story. OpenAI estimates that building, developing and operating this extra 4.5GW will generate over 100,000 jobs across construction, specialized electricians, technicians and indirect roles such as equipment manufacturing and local services.

“We estimate that building, developing and operating the additional 4.5 GW of data center capacity we’re announcing today will create over 100,000 jobs across construction and operations roles in the U.S.,” the company noted, pointing out that many of these positions arise early in a site’s life cycle.

This announcement lands as Big Tech scrambles for power deals of unprecedented scale. Just last week, Google inked a $3 billion, 20‑year pact with Brookfield to secure up to 3GW of U.S. hydropower — the largest corporate hydroelectric deal ever — to feed its AI‑hungry servers. Meanwhile, Microsoft is helping revive the Three Mile Island nuclear plant, aiming to restart Unit 1 (835MW) by 2027 under a two‑decade power purchase agreement that offsets its data‑center footprint in the PJM grid region.

These moves underscore a simple fact: AI at scale is nothing without industrial‑scale power. It’s a new energy arms race, and U.S. states are salivating at the prospect of clean‑tech jobs, construction booms and tax revenues.