Commitment
Capital and project ambition are announced.
← Back to Compute College
Compute College
What is Stargate? OpenAI’s AI infrastructure buildout
How to separate Stargate sites that are operating, under construction, and still planned.
Plain-English definition
Stargate is OpenAI’s multi-site AI infrastructure program. OpenAI says its Abilene flagship operates NVIDIA GB200 systems and trained GPT-5.5, while the 1-gigawatt Michigan campus broke ground in June 2026 and other sites remain at mixed delivery stages. Secured and planned capacity is not capacity already online.
Memory trick: A planned power plant does not charge a battery today; a planned compute campus does not run a model today.
Why it matters
Stargate is a broad AI infrastructure platform built through partnerships across data centers, cloud, chips, energy, construction, and operations. OpenAI says its Abilene flagship operates NVIDIA GB200 systems and trained GPT-5.5, while the 1-gigawatt Michigan campus broke ground in June 2026.
- It is a multi-site infrastructure effort rather than one isolated campus.
- Abilene provides evidence of operating capacity and completed frontier-model training, not only an announcement.
- Michigan provides evidence of construction, not yet operating supply.
- It links compute demand to data centers, energy, chips, construction, and financing.
- OpenAI says it has secured more than the original 10-gigawatt target, but secured and planned capacity is not the same as 10 gigawatts operating today.
- It shows that AI demand is large enough to require national-scale infrastructure planning.
- It can affect expectations for future compute availability long before all capacity is delivered.
- It ties together chips, power, sites, and cloud partners into one supply story.
- It gives readers a framework for separating future commitments from current market supply.
Simple example
Large infrastructure projects move through stages. Each stage matters, but only the final stages become usable market capacity.
Site
Locations, land, and partners are identified.
Infrastructure
Power, construction, cooling, and networking are built.
Deployment
Racks, chips, and systems are installed.
Usable capacity
Real workloads begin running on the infrastructure.
The further a project moves down that path, the more it should influence how readers think about future compute supply. Any figures shown are illustrative calculations, not current quoted market prices.
Sources
Primary sources
Site status was rechecked on Sep 16, 2026. Secured and planned capacity is not presented as fully operating capacity.
Stargate infrastructure and Abilene status
OpenAI says Abilene operates NVIDIA GB200 systems and trained GPT-5.5, while distinguishing secured infrastructure from capacity online.
Stargate Michigan groundbreaking
OpenAI records the June 1, 2026 groundbreaking and the Michigan campus’s planned one-gigawatt scale.
Common mistake
A large buildout can be real and important before all of its capacity is usable. But readers should still distinguish between committed, under-construction, and already-operating infrastructure.
- Committed: What the project intends to build.
- Under development: What is moving through sites, power, construction, and deployment.
- Operating: What is already serving real workloads.
Practical takeaway
What you can do with this
Read Stargate through delivery milestones: financing, site selection, power access, construction, equipment installation, commissioning, and the capacity eventually available for workloads.
- Analysts: separate committed, under-construction, energized, and operating capacity.
- Procurement teams: do not assume a mega-project changes near-term availability until access is documented.
Decision check: attach a date and delivery status to every capacity number used in a market interpretation.
Compute College learning path
Specialty lessons
Step 5 of 6: What is Stargate? OpenAI’s AI infrastructure buildout