Oracle Exits Stargate Expansion, Meta Steps In
Key insights
- Oracle and OpenAI have dropped plans to expand the Abilene data center from 1.2 to 2 gigawatts, citing financing disputes and changing needs
- NVIDIA paid Crusoe $150 million to ensure its chips power the site, blocking AMD from the expansion
- The broader Stargate commitment of 4.5 gigawatts across multiple sites reportedly remains on track
This article is a summary of Oracle, OpenAI End Plans to Expand Flagship Data Center. Watch the video →
Read this article in norsk
In Brief
Oracle and OpenAI have scrapped plans to expand a flagship AI data center in Abilene, Texas, after negotiations stalled over financing and OpenAI's changing needs. The 1,000-acre site is part of the Stargate project. With the expansion shelved, NVIDIA stepped in as a broker, connecting Meta with the site's developer, Crusoe, to take over the unused capacity.
What happened
Oracle had an agreement with data center developer Crusoe and OpenAI to expand the Abilene site from 1.2 gigawatts (GW) of capacity to 2 GW (0:30). That expansion has now been abandoned. The existing 1.2 GW facility is still running, and Oracle continues to use it (1:17).
According to Bloomberg's reporting, NVIDIA paid Crusoe about $150 million as a deposit (1:36). The purpose: to guarantee that whoever leases the expansion capacity uses NVIDIA GPUs (graphics processing units, the chips that power AI workloads), not chips from rival AMD (1:58). NVIDIA then introduced Meta as a potential tenant for the site (1:06).
Early talks between Meta and Crusoe are now underway, though negotiations could still change direction.
Why this matters
The collapse highlights how difficult AI data centers are to build. These projects cost tens of billions of dollars, require massive power supplies, and depend on coordination between developers, chip suppliers, cloud operators, and tenants (0:09).
Market reactions
The news rattled several stocks. Both NVIDIA and AMD shares fell after the report (2:02). CoreWeave, a cloud computing company that serves AI workloads, dropped 3.3% (2:20). The concern: if a high-profile site like Stargate runs into trouble, the broader AI infrastructure buildout may not have the momentum investors assumed (3:23).
The bigger picture remains intact
Oracle and OpenAI still have a commitment for 4.5 GW of capacity across multiple sites (4:03). Bloomberg's reporting suggests this broader effort continues, with other locations outside Texas seen as more practical: easier to build, faster timelines, better conditions, and more available labor (4:12).
Trouble at the Abilene site
Part of the reason for the pullback appears to be operational issues. In January, weather conditions caused outages at the existing Abilene data center (4:25). Bloomberg reports tension between Oracle and Crusoe, though both companies issued statements calling the relationship strong and ongoing (4:42).
What we are tracking next
- Meta-Crusoe negotiations. Whether Meta finalizes a lease for the expansion capacity, and on what terms.
- Oracle's credit profile. Investors are watching Oracle's debt levels closely as it borrows heavily to fund massive data center projects (5:03).
- Paper vs. reality. Many announced data center projects have not broken ground yet. As Bloomberg's Ed Ludlow notes, much of the projected energy demand exists only on paper (5:46).
Glossary
| Term | Definition |
|---|---|
| Gigawatt (GW) | A unit of electrical power equal to one billion watts. For context, 1 GW can power roughly 750,000 homes. |
| Data center | A facility housing thousands of servers that store data and run computing tasks, including AI training and inference. |
| GPU (Graphics Processing Unit) | A specialized chip originally designed for graphics, now widely used to power AI workloads because it can run many calculations simultaneously. |
| Project Stargate | A large-scale AI infrastructure initiative involving OpenAI, Oracle, and other partners, announced at the White House in early 2025. |
| Neo-cloud | A newer generation of cloud computing companies, like CoreWeave, that specialize in GPU-heavy workloads for AI rather than general-purpose computing. |
| Hyperscaler | The largest cloud providers (Amazon Web Services, Microsoft Azure, Google Cloud) that operate at massive scale worldwide. |
Sources and resources
Want to go deeper? Watch the full video on YouTube →