OpenAI claims 10GW of AI infrastructure capacity secured. The company added 3GW in the last 90 days alone, according to a report by Data Center Dynamics.
Key facts
- OpenAI secured 10GW of AI infrastructure capacity.
- 3GW added in the last 90 days.
- Target year for capacity is 2029.
- Hyperscalers spend $90B+ quarterly on data centers.
- Google signed 5GW deal with Anthropic in May 2026.
OpenAI claims to have secured 10GW of AI infrastructure capacity ahead of its 2029 target, according to a report by Data Center Dynamics. The company added 3GW in the last 90 days alone, signaling an accelerated buildout to support training and deployment of large-scale AI models.
This capacity is intended to support training and deployment of large-scale AI models, including future GPT-series systems. The announcement comes as hyperscalers like Google, Microsoft, and Amazon collectively spend over $90 billion quarterly on data center infrastructure, per a recent report [According to Data Center Dynamics].
The unique take: OpenAI's claim is less about raw capacity and more about timing. The 3GW added in 90 days suggests the company is front-loading infrastructure to avoid bottlenecks that plagued earlier model launches. The 2029 target aligns with public forecasts OpenAI made in December 2028, projecting $121 billion in AI research hardware costs for that year [According to prior reporting].
OpenAI's capacity claim directly competes with Google's recent 5GW compute capacity deal with Anthropic, signed in May 2026 [According to prior reporting]. That deal gave Anthropic a dedicated pipeline, while OpenAI's 10GW figure suggests a broader, potentially multi-tenant infrastructure strategy.
How the numbers stack up
The 10GW figure is massive but not unprecedented. Google's 5GW deal with Anthropic and the hyperscalers' $90B+ quarterly capex show that AI infrastructure demand outstrips supply [According to prior reporting]. OpenAI's 3GW in 90 days implies a buildout rate of roughly 1GW per month, which would outpace most current data center construction timelines.
What OpenAI hasn't disclosed
The company did not disclose the geographic distribution of this capacity, the partners involved (e.g., Microsoft Azure, Oracle, or others), or the financial terms. It also did not specify whether this capacity is dedicated to training, inference, or both. These details matter for assessing whether the claim is achievable or aspirational.
What to watch
Watch for OpenAI's next quarterly update or a potential announcement about specific data center partners (e.g., Microsoft, Oracle). The key metric is whether the 3GW/quarter build rate persists through 2027, which would validate the 10GW claim as achievable rather than aspirational.
