OpenAI signs contracts for 10GW US AI compute

OpenAI says it locked up 10GW of US AI capacity

OpenAI reported that it has signed contracts totaling 10GW of AI compute capacity in the United States, with 3GW+ added in roughly the past 90 days. The company framed this as progress toward a target it previously aimed to reach by 2029, underscoring how urgently model makers are securing power and hardware for training and inference.

The announcement matters because compute is increasingly the bottleneck behind the next wave of AI capabilities. Even when model performance improves, deploying larger systems at scale requires extensive investment in data centers, power delivery, and specialized infrastructure such as accelerators.

In the broader AI infrastructure picture reflected across other coverage, major cloud and AI spend continues to accelerate: multiple companies are building out data-center capacity to meet demand, and growth often comes with rising construction and energy pressures. In parallel, concerns about cost, grid constraints, and public opposition to new data-center development have intensified in several regions—creating potential friction between AI expansion plans and local infrastructure realities.

OpenAI’s 10GW figure also signals a shift from “just-in-time” experimentation to long-lead, capacity-secured operations. That approach can reduce uncertainty about availability, but it also increases exposure to infrastructure timelines—like permitting, power interconnection, and ramp-up of new facilities.

Overall, the compute contracts highlight why AI strategy is now inseparable from infrastructure strategy: winners are likely those that can both secure capacity early and deliver it reliably at scale.