As the U.S. heads into Memorial Day weekend, with pools opening, millions set to travel, and backyard barbecues just ahead, a late-session report from the Financial Times revealed that one of President Trump's first Project Stargate AI infrastructure projects is underway in Texas.
According to FT's sources, the under-construction AI data center in Abilene, Texas, is the first U.S. Stargate project, and Oracle is purchasing a staggering 400,000 Nvidia GB200 "Superchips" that will be housed in eight buildings. When the facility comes online next year, the AI chips will require a massive 1.2 gigawatts of electricity.
Site owners Crusoe and Blue Owl Capital have raised $15 billion in debt and equity to finance the Abilene project. JPMorgan provided a sizeable chunk of the debt financing, totaling $9.6 billion across two loans, with $7.1 billion announced this week, according to people familiar with the matter.
Crusoe and Blue Owl have separately invested about $5 billion in cash. Oracle has signed a 15-year lease for the mega project that will be fully operational by mid-2026.
Your hard hat tour of OpenAI’s Stargate megafactory in Abilene, TX with @sama @ChaseLochmiller and lots of red dirt
— Emily Chang (@emilychangtv) May 20, 2025
pic.twitter.com/mMkeKpLVOU
The Abilene facility is part of the broader Stargate project, a $500 billion effort to scale U.S. AI computing backed by OpenAI, SoftBank, Oracle, and Abu Dhabi's MGX. The new Abilene data center will rival Elon Musk's plans to expand his "Colossus" data center in Memphis, Tennessee.
"The Abilene data center is a crucial step in OpenAI's move to reduce its dependence on Microsoft," FT noted, adding, "OpenAI and Microsoft agreed to terminate their exclusivity agreement earlier this year after the startup became frustrated that its demand for power far exceeded the US tech giant's supply."
Sources said Oracle will be leasing the computer power of the 400,000 Nvidia GB200 chips to chatbot startup OpenAI.
Meanwhile, massive U.S. data centers continue to come online—even as China's DeepSeek AI has demonstrated high efficiency, delivering more useful computation with less power. This raises a critical question: Have U.S. AI projects achieved similar levels of efficiency, or are they falling behind in the AI race? Remember what Goldman said about the data center peak capacity in April? Well...
