Microsoft and OpenAI drop cloud exclusivity in revised agreement
- Microsoft and OpenAI said on April 27 they amended their partnership, replacing Azure exclusivity with a right of first refusal on new OpenAI cloud capacity. - The deal still runs through 2030, keeps Microsoft’s IP rights and revenue sharing, and leaves OpenAI free to build extra capacity elsewhere if Microsoft declines. - That matters because OpenAI’s compute appetite now exceeds any one cloud’s footprint, pushing AI infrastructure toward a more multi-cloud, more competitive shape.
Cloud infrastructure is the real story here — not just corporate relationship drama. OpenAI has gotten so compute-hungry that the old setup, where Azure was effectively the mandatory home for everything, stopped fitting the scale of the job. So on April 27, Microsoft and OpenAI rewired the partnership: Azure is no longer the exclusive cloud, and Microsoft now gets a right of first refusal on new OpenAI capacity instead. (blogs.microsoft.com) ### What actually changed? The cleanest way to say it is this: exclusivity became priority access. Back in January 2023, OpenAI said Azure would remain the exclusive cloud provider for all OpenAI workloads across research, API, and products. In Microsoft’s January 21, 2025 update, exclusivity on OpenAI’s APIs was still explicitly in place through 2030. The April 27, 2026 amendment cha(blogs.microsoft.com)n a blanket exclusive lock. (openai.com) ### Why does “right of first refusal” matter? Because it sounds small, but it is a big legal and commercial shift. A right of first refusal means Microsoft gets the first chance to supply the infrastructure OpenAI wants. But if Microsoft can’t or won’t meet the need, OpenAI can go elsewhere. Basically, Microsoft keeps pole position, but loses the power to be the only lane on the track. That is much more flexible than full exclusivity. (blogs.microsoft.com) ### Why did the old model stop working? Scale. OpenAI has been building out Stargate as a giant compute platform with multiple partners, and that buildout has kept expanding. OpenAI’s January 2025 Stargate announcement named Oracle, NVIDIA, SoftBank, Arm, and Microsoft as key initial partners. By July 2025, OpenAI and Oracle alone said they were developing 4.5 gigawatts of additional (blogs.microsoft.com) planned capacity and over $400 billion in investment over three years. That is not “one cloud vendor plus some overflow” territory anymore. (openai.com) ### Does Microsoft still get the valuable parts? Yes — and this is the catch people miss. Microsoft says the amended agreement preserves key economics and product rights. The April 27 posts say the contract still includes Microsoft’s access to OpenAI IP, revenue-sharing arrangements, and a strong joint roadmap through 2030. Earlier joint statements in February 2026 also said Microsoft’s exclusive license and access to OpenAI intel(openai.com)ower surrender of exclusivity in one layer of the stack — cloud capacity. (blogs.microsoft.com) ### Why does Oracle keep showing up? Because Oracle moved from side character to core infrastructure partner. In June 2024, Oracle, Microsoft, and OpenAI said OCI would extend the Azure AI platform to provide extra capacity for OpenAI. That was still Azure-centered. But Stargate pushed further. OpenAI’s later announcements describe Oracle as a direct build-and-operate partner for huge(blogs.microsoft.com)one multi-cloud before the contract fully caught up. (oracle.com) ### Does this open the door to AWS too? In principle, yes. The amended structure no longer says Azure must host everything. If Microsoft passes on capacity or cannot meet the requirement fast enough, OpenAI can source elsewhere. The official statements do not name AWS in this amendment, so the safe read is narrower: the agreement now permits broader cloud optionality. That makes future hosting or distribution deals with additional providers much easier to imagine. (blogs.microsoft.com) ### What changes for customers and developers? Probably less in the short term than the headlines suggest. Microsoft still remains deeply embedded in OpenAI’s product and IP stack, and Azure still matters a lot. But over time, more OpenAI infrastructure spread across multiple clouds could mean faster capacity expansion, more regional deployment options, and more competition on price, (blogs.microsoft.com)ay need to operate across clouds instead of assuming one default platform. (blogs.microsoft.com) ### So what is the bottom line? The old Microsoft-OpenAI deal was built for an era when one hyperscaler could plausibly be the center of gravity. This revised deal is built for an era when OpenAI wants so much compute that even Microsoft has to share the load. Microsoft kept the crown jewels. OpenAI got room to expand. And the AI cloud market just became more competitive. (blogs.microsoft.com)
