Oracle and OpenAI scrap 600MW Texas Stargate expansion plans

The collapse of the Abilene expansion is a rare visible crack in the otherwise relentless AI compute arms race. Oracle and OpenAI walking away from an additional 600MW of capacity at Stargate signals that even the most aggressive players are hitting financial and logistical constraints in building multi‑gigawatt AI campuses. For the race to AGI, it’s a reminder that frontier model roadmaps are now constrained as much by power, real estate, and balance sheets as by algorithms.([communicationstoday.co.in](https://www.communicationstoday.co.in/oracle-and-openai-drop-texas-data-center-expansion-plan/))

Strategically, the most interesting angle is the vacuum this creates. Crusoe still has land and power to monetize, Nvidia reportedly put down a hefty deposit to keep its GPUs in the mix, and Meta is circling as a potential tenant. If Meta secures a large chunk of that abandoned capacity, it strengthens its position as an independent AI stack builder, less reliant on Microsoft- or Oracle-mediated infrastructure. At the same time, Oracle and OpenAI will need to redistribute planned capacity across other Stargate sites, potentially slowing near-term scale-up at a location that had strong political backing and existing grid build-out.([thestar.com.my](https://www.thestar.com.my/tech/tech-news/2026/03/07/oracle-and-openai-end-plans-to-expand-texas-data-center-site-bloomberg-news-reports))

More broadly, this is an early test of whether the current $500B-plus AI infrastructure super‑cycle is sustainable on private balance sheets and vendor financing alone. If marquee projects start getting resized or re‑scoped, expect more creative financing structures, more regional diversification of data centers, and fiercer competition among hyperscalers and sovereigns for where the next 5GW of AGI‑class compute actually gets built.