AI Pulse by Inblix

OpenAI bets $38B on AWS infrastructure in multi-year deal

OpenAI Blog · Jul 12, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: OpenAI bets $38B on AWS infrastructure in multi-year deal

For a company that spent years training its flagship models primarily on Microsoft Azure, OpenAI’s new multi-year deal with AWS marks a significant, if not entirely surprising, expansion of its compute portfolio. The $38 billion commitment gives Sam Altman’s team immediate access to hundreds of thousands of Nvidia GPUs—both GB200s and GB300s—clustered inside Amazon EC2 UltraServers, with the architecture designed to scale to tens of millions of CPUs for increasingly demanding agentic workloads.

Matt Garman, AWS’s CEO, framed the partnership as putting his company’s infrastructure at the center of OpenAI’s ambitions, calling it the “backbone for their AI ambitions.” The deployment timeline is aggressive: all capacity is targeted to be online before the end of 2026, with room to grow through 2027 and beyond. The clusters will handle everything from ChatGPT inference to training the next generation of frontier models, leaning on AWS’s experience running clusters that have already topped 500,000 chips.

This isn’t the first time the two companies have linked arms. OpenAI’s open-weight foundation models landed on Amazon Bedrock earlier this year, attracting thousands of customers—including Peloton, Thomson Reuters, and Comscore—who are using them for coding, scientific analysis, and agentic workflows. That existing relationship likely greased the wheels for a much deeper infrastructure tie-up. Still, it’s hard to ignore the optics: OpenAI, which has been synonymous with Microsoft’s cloud, is now writing enormous checks to its chief rival.

The broader signal here is about the sheer hunger for compute among frontier labs. When Altman says scaling AI requires “massive, reliable compute,” he’s not being poetic—he’s describing a market where no single cloud provider can satisfy the appetite of a leading model builder. The question now is whether this diversification becomes the norm for other labs, or if OpenAI’s deal is simply a pressure release valve for a uniquely strained relationship with its primary backer.

💡 Key Takeaways

  1. OpenAI will pay $38 billion over multiple years for AWS infrastructure, marking a major diversification away from its primary reliance on Microsoft Azure.
  2. The new AWS clusters, built on Nvidia GB200 and GB300 GPUs via EC2 UltraServers, are designed to scale from hundreds of thousands of chips to tens of millions of CPUs.
  3. AWS is targeting full deployment of OpenAI’s allotted capacity by the end of 2026, with options to expand further through 2027 and beyond.
  4. Thousands of enterprises already use OpenAI’s models on Amazon Bedrock, including Peloton, Thomson Reuters, and Comscore, for tasks ranging from coding to scientific analysis.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles