OpenAI and Oracle scrap Texas Stargate AI hub expansion as Meta, Nvidia circle

The collapse of the Abilene Stargate expansion is less about one site and more about the strain that multi‑gigawatt AI campuses are putting on capital markets, power grids, and planning timelines. OpenAI and Oracle are effectively saying that adding another 800 MW at a single location is too lumpy and too risky right now, even with unprecedented demand for frontier‑model compute. That’s a notable data point in a year when AI infrastructure had looked like an unstoppable, money‑no‑object arms race.

But the vacuum doesn’t last long: Meta is already in talks to take over the unused capacity, and Nvidia has paid a hefty deposit to ensure its GPUs remain the de facto standard at the site. That underscores a deeper shift: control over high‑end compute is concentrating in a handful of hyperscalers and chip vendors, even as individual projects wobble. For the race to AGI, the signal is that aggregate compute growth is still on track, but Turbocapitalism is forcing painful triage on where, and with whom, that capacity gets built.