OpenAI and Oracle have abandoned plans to grow their flagship Stargate data center campus in Abilene, Texas, beyond its current footprint, citing at least a one-year wait for additional power and a company-wide shift toward Nvidia’s next-generation hardware. The cancellation closes the door on what was originally envisioned as a 2-gigawatt AI computing complex and marks a significant recalibration of one of the most high-profile infrastructure bets in recent memory.
The site, developed by Crusoe on Lancium’s Clean Campus in Abilene about 180 miles west of Dallas, launched its first two buildings in September 2025. Oracle had pushed to fill six additional buildings that would have brought total capacity to roughly 1.2 gigawatts, with a longer-term target of 2 gigawatts. Both companies passed on the expansion when it became clear power delivery could not be guaranteed for another year.
“Our flagship Stargate site is one of the largest AI data center campuses in the United States,” said Sachin Katti, who oversees compute scaling at OpenAI. “We considered expanding it further, but ultimately chose to put that additional capacity in other locations. Today we have more than half a dozen sites under development across multiple states, including the site we’re building with Oracle in Wisconsin, where the first steel beams went up just this week.”
Power Outage, Shifting Demand, and a Key Departure
Power was not the only complication. A multi-day outage caused by winter weather damaged liquid cooling equipment at the Abilene site earlier this year, souring the relationship between OpenAI and Crusoe. The episode followed a period of shifting demand forecasts at OpenAI and, according to reporting by Bloomberg and Data Center Dynamics, contributed to the departure of Keith Heyde, the company’s director of physical infrastructure, who had championed the in-house data center push. OpenAI has since leaned more heavily into cloud contracts.
Perhaps the clearest signal of where OpenAI is heading came earlier this week, when OpenAI and Nvidia announced a strategic partnership to deploy 10 gigawatts of Nvidia systems, with the first gigawatt scheduled for the second half of 2026 on the Vera Rubin platform. Rather than filling Abilene with Blackwell GPUs already headed there, OpenAI is effectively skipping a generation at the Texas site and planning to deploy Vera Rubin hardware elsewhere.
That leaves Crusoe holding empty buildings and Nvidia working to fill them. The chip company paid a $150 million deposit to Crusoe to secure the site and is in early-stage talks with Meta to take over the capacity. No deal has been signed. Nvidia’s involvement is partly driven by a desire to keep the facility from landing in the hands of a chip designer that competes with Nvidia’s own products.
OpenAI’s revised compute spending commitments reflect the broader recalibration. The company had projected $1.4 trillion in compute investment by 2033; that figure has since been revised down to $600 billion by 2030, with up to $300 billion expected to flow to Oracle. Oracle, meanwhile, is carrying more debt and has signaled it plans to cut thousands of jobs.