OpenAI announced a strategic partnership with NVIDIA to deploy 10 gigawatts of NVIDIA systems, marking a significant expansion of its infrastructure capacity [source]. The agreement, disclosed on 22 September 2025, represents one of the largest single commitments to AI computing infrastructure by any provider to date.

The deployment scale—10 gigawatts—indicates substantial planned capacity increases for OpenAI's model training and inference operations. For context, a single gigawatt can power approximately 750,000 homes, suggesting the partnership will support infrastructure equivalent to powering 7.5 million households.

OpenAI did not specify a timeline for the deployment or which NVIDIA systems would be procured. The announcement did not detail whether the capacity would support new model development, expanded inference serving for existing models, or both. No information was provided about the geographic distribution of the systems or their intended operational start dates.

The partnership follows OpenAI's ongoing capacity constraints, which have periodically affected service availability for ChatGPT and API users. Previous outages and rate limiting incidents have been attributed to demand exceeding available compute resources.

NVIDIA has supplied GPU infrastructure to most major AI providers, including Anthropic, Google, and Microsoft. The 10-gigawatt commitment positions OpenAI among the largest single customers for NVIDIA's data center products.

The announcement did not address how the expanded capacity might affect model performance, availability, or pricing for OpenAI's services. Users have not yet reported observable changes to model behaviour or service reliability following the partnership disclosure.