A look at OpenAI's search for the sites of its Stargate data centers in the US; OpenAI has received 800+ applications since January and has ~20 finalist sites
It wasn't how Keith Heyde envisioned celebrating the holidays. Rather than hanging out with his wife back home in Oregon …
Context & Ripple Effects
Stargate had already moved from a first build in Abilene—where eight near-identical data-center structures were under construction—to a broader U.S. footprint. More than 800 submissions and roughly 20 finalists show that site selection, rather than simply announcing capacity, had become a major execution step.
The breadth of interest gives OpenAI leverage in choosing among local partners and infrastructure conditions. But later reporting that Stargate stalled amid disputes over its buildout makes the conversion of finalists into operating capacity the central question.
First-order effects
- OpenAI can concentrate diligence and negotiations on about 20 U.S. locations, while most applicant communities and proposed partners fall out of the active process.
- Finalist locations gain a nearer-term opportunity to compete for a Stargate facility, putting local infrastructure readiness and partner commitments under closer scrutiny.
Second-order effects
- The shortlist raises the stakes for local developers, utilities, and public officials to present credible power, construction, and permitting pathways; OpenAI has more alternatives if any one site cannot deliver.
- A larger menu of potential sites may improve OpenAI's negotiating position, but it does not remove the delivery risk later reflected in a shift toward renting more AI servers from cloud providers.
Third-order effects
- If hyperscale AI projects continue to solicit sites at this scale, data-center location decisions will increasingly resemble utility-infrastructure planning, with power availability and local consent shaping AI capacity deployment.
- The gap between attracting applications and delivering usable compute could favor AI labs that combine long-horizon construction plans with flexible cloud sourcing rather than relying on a single buildout path.
The trend: AI labs are turning compute expansion into a distributed infrastructure and site-selection challenge, balancing owned facilities against rented capacity when construction execution is uncertain.