Building the Compute Infrastructure for the Age of Intelligence

Stargate is OpenAI's long-term plan to expand compute infrastructure in order to reliably deliver AGI to the world. In January 2025, OpenAI committed to securing 10GW of AI infrastructure in the US, and in just over a year it has surpassed that goal, adding more than 3GW in just the last 90 days alone, the company said. As demand grows rapidly, OpenAI stated plainly that the answer is to secure and bring online more compute, faster.

Compute is the core input that makes advanced AI possible, and it sits at the center of the AI flywheel that underpins training better models, serving them reliably, improving performance, cutting costs, and enabling greater reinvestment. While financing structures and partnership forms may change, what matters, OpenAI explained, is securing large-scale capacity on time while retaining flexibility to adapt to shifts in technology and demand.

This expansion is defined not as the work of a single company but of a partner ecosystem. OpenAI believes that communities, utilities, energy providers, chip makers, cloud operators, construction firms, investors, skilled workers, and the public sector all need to work together, and it said it is reviewing potential data center sites across the US with plans to significantly increase capacity in the coming years.

The company also emphasized the impact it hopes to leave on local communities. This week, it kicked off a community engagement program with a donation to the Port Washington-Saukville Education Foundation in Wisconsin, joined by Vantage Data Centers and Oracle. The donation will support expanded opportunities for students, resources for teachers, and career-readiness programs.

The local effects OpenAI expects include:

• Good jobs
• Improved educational environments
• New local tax revenue
• Responsible energy planning
• Careful water management
• Early collaboration with state and local governments

Added to this is collaboration with NABTU and its affiliated skilled trade unions. OpenAI stressed that building AI infrastructure isn't just the work of engineers—it requires a trained, skilled workforce capable of building safely and quickly.

The Abilene, Texas site was presented as an example of this approach. Instead of traditional evaporative cooling towers, the site uses closed-loop cooling, filling once and then continuously recirculating the water. The initial fill is roughly the equivalent of 2 Olympic swimming pools per building, and once complete, the entire cooling system's annual water use is expected to be comparable to that of a mid-sized office building or about 4 average households.

The latest model, GPT-5.5, was trained at this site, and operations run on Oracle Cloud Infrastructure and NVIDIA GB200 systems. Through this, OpenAI aims to narrow the capability overhang—the productivity gap between advanced AI power users and everyday users—so that more people can use AI to create, learn, discover, and solve problems. The core message: the age of intelligence won't arrive on its own—it must be built together with partners and communities.