OpenAI says Stargate has passed 10 GW of AI infrastructure

OpenAI says its Stargate programme has already passed 10 GW of secured AI infrastructure in the US, with more than 3 GW added in the last 90 days.

The facts: in a 29 April post, OpenAI describes Stargate as the long-term compute foundation behind its models. The company says the original January 2025 target — securing 10 GW by 2029 — has already been exceeded. OpenAI also says GPT-5.5 was trained at the Stargate site in Abilene, Texas, running on Oracle Cloud Infrastructure with NVIDIA GB200 systems.

This is not just a data-centre story. For executive teams, AI capacity is becoming a strategic input alongside energy, capital, cloud contracts and scarce technical skills. When model quality, latency and unit cost increasingly depend on physical capacity, supplier choice becomes more than an API decision.

CIOs should read this as a warning about deeper infrastructure lock-in. OpenAI points to an ecosystem of cloud providers, neoclouds, chipmakers, energy providers, financiers and local authorities. That can accelerate deployment, but it also makes the value chain more complex. The risk is not only in the model; it sits in power, permitting, water, GPU supply, regional availability and contract structure.

For CFOs and boards, the central question is cost and resilience. If advanced AI services concentrate around a small number of hyperscale and chip supply chains, companies need to understand how pricing, data location, contingency planning and bargaining power will be affected. This matters most as AI moves from pilots into customer operations, software development, analytics and case handling.

The judgement: companies should not try to copy Stargate. They should copy the discipline. Map which processes genuinely need frontier models, which can run on smaller or open models, and which supplier dependencies need limits before usage scales. An AI strategy without capacity planning and an exit plan can quickly become an expensive cloud strategy under a new name.