Oracle and OpenAI have dropped plans to expand their flagship Stargate data centre campus in Abilene, Texas, after negotiations broke down over financing and OpenAI’s changing capacity forecasts, Bloomberg reported on 6 March, according to Tom’s Hardware. The existing campus is not affected, and Oracle says its wider data centre programme for OpenAI is on track.
What was dropped was a planned expansion of about 600 megawatts. OpenAI had first described it in September 2025 as “an additional potential expansion of 600 megawatts near the flagship Stargate site in Abilene”, in a post announcing five new Stargate sites. Since mid-2025, Oracle, the developer Crusoe and OpenAI had discussed taking the campus from about 1.2 gigawatts to roughly 2 gigawatts, Tom’s Hardware said.
What stays in place
The main Abilene campus covers about 1,000 acres and eight buildings, with two already running training and inference workloads on Nvidia GB200 racks and the rest due by mid-2026, according to a Tom’s Hardware explainer. OpenAI’s compute chief at the time, Sachin Katti, said the company “considered expanding it further” but chose to put that capacity elsewhere. Katti has since left OpenAI, as we reported when it named a new head of compute and Stargate.
Oracle pushed back hard on the coverage. In a statement on X on 9 March, it called reports about the site “false and incorrect”, said it and Crusoe were “operating in lockstep”, and said two buildings were “completely operational”, Tom’s Hardware reported. Bloomberg’s Ed Ludlow then clarified that while the two companies “are not moving ahead with the planned expansion lease”, their separate agreement for 4.5 gigawatts of capacity “remains on track”.
That 4.5-gigawatt agreement is the larger commitment. OpenAI announced it in July 2025 as part of its $500bn, 10-gigawatt Stargate plan, and later put the Oracle partnership at more than $300bn over five years.
Why the expansion fell through
Bloomberg pointed to financing terms and OpenAI’s shifting forecasts. Tom’s Hardware added that winter weather earlier in the year had disrupted parts of the liquid-cooling system and taken several buildings offline for days, straining relations between Oracle and Crusoe. The Information reported that power at the expansion site would not be ready for about a year, which it called the “real reason” OpenAI walked away, Tom’s Hardware said.
Where Meta comes in
With the space unclaimed, Crusoe looked for another tenant. Nvidia gave Crusoe a $150m deposit to secure the space and approached Meta about taking the lease, Bloomberg reported, with the aim of keeping AMD hardware out of a site already running Nvidia chips. No deal between Meta and Crusoe had been confirmed, Tom’s Hardware said. Meta was also buying capacity elsewhere: a week later it agreed a deal worth up to about $27bn with Nebius, which we covered in our Nebius story.
What the Abilene episode shows about AI build-outs
The first headlines described Stargate collapsing. The facts were narrower: one expansion lease, at one campus, did not go ahead, while the main site kept running and the larger Oracle agreement stood. For readers, the lesson is to separate three things that are easy to blur: an announced commitment, a signed lease and a building that is switched on. Abilene’s 600 megawatts was only ever a “potential expansion”.
The reasons given are also worth noting because none of them is about demand for AI. Financing terms, a cooling outage and the timing of grid power were the problems reported. Power arriving a year late matters because the hardware moves on in the meantime; capacity built around today’s chips can look less attractive by the time it is live.
Nvidia’s reported role is the other telling detail. A chip supplier putting down a deposit to keep its own hardware in a building shows how much the suppliers now care about where capacity lands, not just how much is built. Oracle’s financing has stayed under scrutiny since, as our report on record prices for credit insurance on Oracle debt and on Broadcom’s reported financing for OpenAI’s chip set out.
Correction, 22 October 2026: an earlier version of this article said the OpenAI-Oracle Stargate partnership had collapsed and described it as a $100bn deal. In fact, a planned 600MW expansion at the Abilene campus was dropped; the existing campus and the separate 4.5GW Oracle-OpenAI agreement continued. Unsourced figures on GPU prices, supply and demand have been removed.
Sources
- Tom’s Hardware: OpenAI’s massive Stargate data center canceled as firm can’t reach terms with Oracle, 8 March 2026
- Tom’s Hardware: Oracle hits back at Stargate data center cancellation reports, 9 March 2026
- Tom’s Hardware: Oracle and OpenAI’s Abilene expansion saga detailed
- OpenAI: OpenAI, Oracle, and SoftBank expand Stargate with five new AI data center sites, 23 September 2025
- OpenAI: Stargate advances with 4.5 GW partnership with Oracle, 22 July 2025
Related stories
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
