What just happened? OpenAI is scaling its compute infrastructure beyond Microsoft with a $38 billion capacity purchase from Amazon Web Services, marking its first direct partnership with the largest player in global cloud infrastructure. The deal, announced Monday, establishes OpenAI as a new AWS customer and underscores its move toward a multi-cloud strategy as it readies for a potential public offering.
The agreement allows OpenAI to begin deploying workloads on AWS immediately, using hundreds of thousands of Nvidia GPUs across US data centers, with additional capacity expected to come online over several years.
Until recently, Microsoft held exclusive rights to host OpenAI's cloud services. The two companies signed their first investment deal in 2019, and Microsoft has since invested a cumulative $13 billion. That exclusivity ended in January, when Microsoft agreed to a non-exclusive framework granting the company the right of first refusal on new cloud workloads. That preferential status expired last week, giving OpenAI more flexibility to partner with other hyperscale providers.
The AWS deal follows a series of infrastructure commitments totaling roughly $1.4 trillion with companies including Nvidia, Broadcom, Oracle, and Google. Collectively, those contracts mark the most aggressive capacity expansion effort in the history of the AI sector, raising questions about how much energy and processing power can realistically be delivered in the next decade.
For Amazon, the OpenAI partnership arrives at a strong financial moment for its cloud division. The company reported over 20% year-over-year revenue growth at AWS in the company's recent earnings, beating Wall Street forecasts. That's slightly below growth rates at Microsoft Azure and Google Cloud, which reported 40% and 34% expansion, respectively, but still enough to maintain AWS's leadership in total cloud market share.
Dave Brown, Amazon's vice president of compute and machine learning services, said the OpenAI deal involves reserved capacity wholly separate from AWS's existing commitments. "It's completely separate capacity that we're putting down," Brown told CNBC. "Some of that capacity is already available, and OpenAI is making use of that."
The initial phase will use existing AWS data centers. Amazon plans to build out additional infrastructure dedicated to OpenAI's workloads, extending its commitment beyond current facilities.
The agreement centers on Nvidia's latest Blackwell GPUs. AWS will supply the compute power to train future frontier models and support real-time inference – including ChatGPT responses – through its Elastic Compute Cloud.
While the two companies have not yet disclosed plans involving Amazon's own Trainium chips, Brown noted that AWS has seen clients such as Anthropic adopt Trainium for advanced workloads. "We like Trainium because we're able to give customers better price performance and honestly give them choice," Brown said, while declining to confirm whether OpenAI would use the chip.
For OpenAI, the deal represents a shift away from total dependence on Microsoft Azure and a move toward operational independence. The partnership also strengthens OpenAI's presence on AWS's Bedrock platform, where its foundation and open-weight models are already accessible to enterprise users through managed services. Companies such as Peloton, Thomson Reuters, Comscore, and Triomics use those models for code generation, analytical workloads, and scientific applications.
OpenAI CEO Sam Altman said that the firm needs multiple partners to meet growing compute demands. "Scaling frontier AI requires massive, reliable compute," Altman said. "Our partnership with AWS strengthens the broad compute ecosystem that will power this next era and bring advanced AI to everyone."
OpenAI signs $38 billion Amazon Web Services deal for Nvidia-powered cloud compute
