OpenAI has signed a deal to buy $38 billion worth of capacity from Amazon Web Services, its first contract with the leader in cloud infrastructure and the latest sign that the $500 billion artificial intelligence startup is no longer reliant on Microsoft.
Under the agreement announced on Monday, OpenAI will immediately begin running workloads on AWS infrastructure, tapping hundreds of thousands of Nvidia’s graphics processing units (GPUs) in the U.S., with plans to expand capacity in the coming years.
Amazon stock climbed about 5% following the news.
The first phase of the deal will use existing AWS data centers, and Amazon will eventually build out additional infrastructure for OpenAI.
“It’s completely separate capacity that we’re putting down,” said Dave Brown, vice president of compute and machine learning services at AWS, in an interview. “Some of that capacity is already available, and OpenAI is making use of that.”
OpenAI has been on a dealmaking spree of late, announcing roughly $1.4 trillion worth of buildout agreements with companies including Nvidia, Broadcom, Oracle and Google — prompting skeptics to warn of an AI bubble and question whether the country has the power and resources needed to turn the ambitious promises into reality.
Until this year, OpenAI had an exclusive cloud agreement with Microsoft, which first backed the company in 2019 and has invested a total of $13 billion. In January, Microsoft said it would no longer be the exclusive cloud provider for OpenAI, and was moving to an arrangement where it would have right of first refusal for new requests.
Last week, Microsoft’s preferential status expired under its newly negotiated commercial terms with OpenAI, freeing the ChatGPT creator to partner more widely with the other hyperscalers. Even before that, OpenAI forged cloud deals with Oracle and Google, but AWS is by far the market leader.
“Scaling frontier AI requires massive, reliable compute,” OpenAI CEO Sam Altman said in Monday’s release. “Our partnership with AWS strengthens the broad compute ecosystem that will power this next era and bring advanced AI to everyone.”
OpenAI will still be spending heavily with Microsoft, reaffirming that commitment by saying last week that it will purchase an incremental $250 billion of Azure services.
For Amazon, the pact is significant both in the size and scale of the deal itself and because the cloud giant has close ties to OpenAI rival Anthropic. Amazon has invested billions of dollars in Anthropic, and is currently constructing an $11 billion data center campus in New Carlisle, Indiana, that’s designed exclusively for Anthropic workloads.
“The breadth and immediate availability of optimized compute demonstrates why AWS is uniquely positioned to support OpenAI’s vast AI workloads,” AWS CEO Matt Garman said in the release.
In its earnings report last week, Amazon reported more than 20% year-over-year revenue growth at AWS, beating analyst estimates. But growth was faster at Microsoft and Google, which reported cloud expansion of 40% and 34%, respectively.
Starting on Nvidia
The current agreement with OpenAI is explicitly for use of Nvidia chips, including two popular Blackwell models, but there’s potential to incorporate additional silicon down the road. Amazon’s custom-built Trainium chip is being used by Anthropic in the new facility.
“We like Trainium because we’re able to give customers something that gives them better price performance and honestly gives them choice,” Brown said, adding that he can’t provide any details on “anything we’ve done with OpenAI on Trainium at this point.”
The infrastructure will support both inference — such as powering ChatGPT’s real-time responses — and training of next-generation frontier models. OpenAI can expand with AWS as needed over the next seven years, but no plans beyond 2026 have been finalized.
OpenAI CEO Sam Altman (L) shakes hands with Microsoft Chief Technology Officer and Executive VP of Artificial Intelligence Kevin Scott during the Microsoft Build conference at the Seattle Convention Center Summit Building in Seattle, Washington, U.S., on May 21, 2024.
Jason Redmond | Afp | Getty Images
OpenAI’s foundation models, including so-called open-weight options, are already available on Bedrock, AWS’s managed service for accessing leading AI systems.
Companies including Peloton, Thomson Reuters, Comscore, and Triomics use OpenAI models on AWS for tasks ranging from coding and mathematical problem solving to scientific analysis and agentic workflows.
Monday’s announcement establishes a more direct relationship.
“As part of this deal, OpenAI is a customer of AWS,” Brown said. “They’ve committed to buying compute capacity from us, and we’re charging OpenAI for that capacity. It’s very, very straightforward.”
For OpenAI, the most highly valued private AI company, the AWS agreement is another step in getting ready to eventually go public. By diversifying its cloud partners and locking in long-term capacity across providers, OpenAI is signaling both independence and operational maturity.
Altman acknowledged in a recent livestream that an IPO is “the most likely path” given OpenAI’s capital needs. CFO Sarah Friar has echoed that sentiment, framing the recent corporate restructuring as a necessary step toward going public.
WATCH: AWS CEO Matt Garman on Amazon’s massive new AI data center for Anthropic
