OpenAI clears 10GW early, turning the AI race into a power race

OpenAI's April 29 Stargate update matters because it turns AI scale from an abstract ambition into a physical buildout race. In the source post, the company said it has already moved past the 10GW U.S. infrastructure target that Stargate had set for 2029. More than 3GW, OpenAI says, came online in the last 90 days alone. That is not a model benchmark. That is a reminder that power, land, permitting, construction, and supply chains are now central AI variables.

The post frames compute as the core input that determines what advanced systems can do, how quickly they can be served, and how cheaply they can be delivered over time. OpenAI also says it is now evaluating data center locations beyond the initial 10GW goal, which suggests the ceiling it set last year has already become the floor. In other words, the industry did not stabilize when the first giant target was reached. The target simply moved.

Abilene, Texas is the concrete example OpenAI wants policymakers and investors to look at. The company says its flagship Stargate site there uses closed-loop cooling rather than traditional evaporative towers, and that annual water use for the full cooling system at buildout should be comparable to a medium-sized office building, or about four average households. OpenAI also disclosed that GPT-5.5 was trained at that Abilene site on Oracle Cloud Infrastructure using NVIDIA GB200 systems. The infrastructure story is therefore inseparable from the product story.

This is why the update lands harder than a routine expansion note. The next moat is not only model quality. It is who can secure power, bring capacity online on schedule, and do it without triggering local political blowback on water, labor, or land use. OpenAI is arguing that the Intelligence Age will be constrained by construction speed as much as research speed. After this update, that argument looks difficult to dismiss.