OpenAI has walked away from expanding its flagship Stargate data center in Abilene, Texas, capping the site at 1.2 gigawatts instead of the planned 2 GW buildout. The Information reports that the breakdown stems from a mix of power grid delays, reliability issues with site operator Crusoe, and OpenAI’s own shifting priorities toward next-generation Nvidia chips at new locations.
The original plan called for Oracle to expand OpenAI’s presence at the Abilene campus into six additional buildings, nearly doubling capacity. But neither side was willing to wait more than a year for the additional power supply to come online, according to The Information. That timeline gap killed the deal.
What Went Wrong
Three factors converged:
- Power grid delays. The expansion to 2 GW required grid upgrades that would take over a year to deliver. Neither OpenAI nor Oracle wanted to wait.
- Crusoe reliability problems. A multi-day outage earlier this year, caused by winter weather damaging liquid cooling equipment, strained the relationship between OpenAI and site operator Crusoe.
- Next-gen chip strategy. OpenAI wants newer Nvidia hardware at fresh sites rather than packing more capacity into an existing location. The company’s demand forecasts have been shifting frequently, making long-term commitments harder to nail down.
Nvidia Steps In With $150M
Here’s where it gets interesting. Nvidia reportedly paid a $150 million deposit to Crusoe to secure the now-vacant expansion capacity at Abilene. The move appears designed to keep a rival chip designer’s GPUs out of that space. Nvidia is actively working to bring Meta in as a replacement tenant for the excess capacity.
That’s a remarkable play: Nvidia spending its own money to control where its chips end up and, more importantly, where competitor chips don’t.
Oracle Pushes Back
Oracle isn’t taking the reports quietly. The company hit back at what it called “incorrect reporting,” insisting that the broader 4.5 GW Oracle-OpenAI partnership, announced in July 2025, remains fully on track. Oracle emphasized that the Abilene campus is already operational on Oracle Cloud Infrastructure, with NVIDIA GB200 racks delivered and OpenAI running early training and inference workloads.
The distinction Oracle is drawing: the Abilene expansion may be off the table, but the multi-site agreement spanning several U.S. locations hasn’t changed.
Why This Matters
This is a signal that the AI infrastructure buildout isn’t a straight line. Even inside a $500 billion initiative like Stargate, individual site deals can fall apart when power timelines slip and operational partners underperform.
What stands out here is OpenAI’s willingness to walk away from its own flagship campus expansion. It suggests the company is prioritizing flexibility and access to the latest silicon over raw megawatt commitments at any single location. For the broader data center industry, it’s a reminder that power availability remains the real bottleneck, not financing or demand.
Meta potentially stepping into the Abilene capacity, brokered by Nvidia, would add yet another twist to the already tangled web of AI infrastructure partnerships. More details are available in the original report from The Information.