Oracle and OpenAI Scrap Texas Stargate Data Center Expansion
Oracle and OpenAI abandoned plans to expand their flagship Abilene data center from 1.2 GW to 2 GW after financing talks collapsed and a winter cooling outage strained relations with operator Crusoe.
4 min read
Oracle and OpenAI have scrapped plans to expand their flagship AI data center campus in Abilene, Texas, after negotiations broke down over financing and OpenAI's shifting infrastructure needs, Bloomberg reported. Meta is now in talks to pick up the abandoned capacity - with Nvidia playing matchmaker to keep AMD out.
TL;DR
- Oracle and OpenAI dropped plans to expand Abilene from 1.2 GW to 2 GW after financing talks dragged and OpenAI's demand projections kept changing
- A multi-day winter outage at the Crusoe-operated site damaged liquid cooling systems and strained the OpenAI-Crusoe relationship
- Nvidia paid Crusoe a $150 million deposit to hold the site and is brokering Meta as the new tenant - keeping AMD GPUs out
- Oracle insists the broader 4.5 GW Stargate agreement with OpenAI remains on track, including a new site near Detroit
- The cancellation comes days after Oracle announced up to 30,000 layoffs to free cash for AI data center construction
What Was Planned
Since mid-2025, Oracle, OpenAI, and site operator Crusoe had been negotiating to nearly double the Abilene facility's power capacity from 1.2 gigawatts to roughly 2 gigawatts. The expansion was part of the broader Stargate initiative - the $500 billion AI infrastructure program President Trump announced in January, spanning five US data center sites.
The Abilene campus is Stargate's flagship. Two buildings are already operational, running Nvidia Blackwell processors. The expansion would have added approximately 600 megawatts of additional capacity.
Why It Fell Apart
Three factors killed the deal:
Financing. The terms never came together. Extended negotiations over who would pay for what dragged on without resolution - a pattern that echoes Oracle's broader balance sheet pressures. The company is cutting up to 30,000 jobs to free $8 to $10 billion in cash flow for data center construction.
OpenAI's shifting demands. OpenAI's capacity forecasts changed repeatedly during negotiations. The company is now prioritizing next-generation Nvidia chips at new sites rather than expanding an existing facility that would use current-generation Blackwell processors. The additional Abilene power wasn't projected to come online for over a year - by which point OpenAI wants to be running newer hardware in bigger clusters elsewhere.
The Crusoe outage. A multi-day outage earlier this year, caused by winter weather damaging liquid cooling equipment at the Crusoe-operated facility, reportedly strained the relationship between OpenAI and the site operator. Texas power grid reliability - or lack thereof - is a recurring concern for hyperscale facilities in the state.
Meta Steps In, Nvidia Plays Gatekeeper
With the OpenAI expansion dead, the scramble for the capacity began immediately. Nvidia paid Crusoe a $150 million deposit to secure the site and began facilitating talks between Crusoe and Meta as the replacement tenant.
The deposit is strategic, not altruistic. Nvidia wants to ensure its own GPUs - rather than AMD's - fill the expanded facility. Meta, which has committed $600 billion to AI infrastructure over the next three years, is a natural buyer for gigawatt-scale capacity.
Negotiations between Meta and Crusoe remain active.
Oracle Pushes Back
Oracle disputed the framing. The company said it and Crusoe are "operating in lockstep" at the Abilene campus, with two buildings already operational and the remainder of the site progressing on schedule.
On the broader partnership, Oracle insisted the separate 4.5 GW agreement with OpenAI - announced in July 2025 - remains fully on track, with additional projects including a new site near Detroit. Bloomberg's own reporting confirmed this distinction: the Abilene expansion lease is dead, but the larger multi-site Stargate buildout continues.
What It Means
The Abilene cancellation is a data point, not a death blow, for Stargate. But it reveals real friction in the AI infrastructure buildout:
- Financing is harder than announcements. A $500 billion headline doesn't mean the capital is committed. When individual site expansions require actual term sheets, the deals can fall apart.
- Demand forecasting is a moving target. OpenAI's requirements changed enough times during negotiations to collapse a deal with a partner that was already building for them.
- Reliability matters. A single winter outage was enough to damage a critical operator relationship and factor into a billion-dollar expansion decision.
Oracle stock shed its gains and turned negative after the Bloomberg report. Nvidia and AMD shares also hit session lows - any sign of weakening demand from major AI spenders rattles the entire upstream supply chain.
Sources: Oracle and OpenAI End Plans to Expand Flagship Data Center - Bloomberg | OpenAI, Oracle abandon 2 GW AI data center expansion - Interesting Engineering | Oracle drops after report - Sherwood News | Oracle hits back at Stargate data center cancellation reports - Tom's Hardware | OpenAI and Oracle Cap Texas AI Data Center at 1.2 GW - WinBuzzer
