OpenAI is expanding its cloud footprint through a new strategic alliance with Amazon Web Services (AWS). The deal provides OpenAI with direct access to AWS’s large-scale GPU infrastructure, comprising hundreds of thousands of Nvidia units distributed across multiple U.S. data centers. AWS and OpenAI intend to scale this capacity further as demand for advanced AI workloads continues to rise.
This move officially ends OpenAI’s exclusive run with Microsoft. Since Microsoft’s initial $1 billion investment back in 2019, now totaling around $13 billion, it’s been the only company hosting OpenAI’s cloud operations. The setup changed earlier this year to a looser deal where Microsoft got first dibs on new projects. Now that even that agreement has lapsed, OpenAI can finally branch out and work with other cloud providers.
The partnership with AWS is just one part of a much larger industry push. OpenAI has lined up infrastructure deals with major players, including Nvidia, Broadcom, Oracle, and Google, commitments valued at around $1.4 trillion in total. The scale of this expansion is unprecedented in the AI world, raising new questions about how far energy and computing capacity can realistically be stretched in the decade ahead.
This deal comes at a good time for Amazon. AWS just announced revenue growth topping 20% compared to last year, beating analysts’ expectations. The pace is a bit slower than Microsoft Azure and Google Cloud, but AWS still holds the top spot in the global cloud market.
Amazon says the servers set aside for OpenAI are brand new. Dave Brown, who leads Compute and Machine Learning Services at AWS, told CNBC that the capacity is “completely separate” from what’s already running on the platform. Some of it’s already online and being used by OpenAI, and Amazon plans to build more infrastructure just for the company’s workloads.
At the center of the deal are Nvidia’s newest Blackwell GPUs, which will power OpenAI’s next wave of “frontier” models. AWS will provide the compute resources needed to train these systems and handle real-time tasks, like powering ChatGPT responses, through its Elastic Compute Cloud. While neither company has said whether OpenAI will use Amazon’s own Trainium chips, Dave Brown noted that other AI firms, including Anthropic, have adopted them for advanced workloads, citing strong price performance and flexibility.
For OpenAI, this deal is about gaining more freedom. It reduces the company’s dependence on Microsoft’s Azure cloud and builds a stronger connection with AWS, especially through the Bedrock platform, where businesses already tap OpenAI’s models for coding help, data analysis, and research.
For Sam Altman, the equation is simple: advanced AI can’t move forward without immense, dependable computing power. “Scaling frontier AI requires massive, reliable compute,” he said. Partnering with AWS, he added, is part of building the infrastructure that will shape the next chapter of AI and open it up to more people around the world.
Maybe you would like other interesting articles?