OpenAI sells models across any cloud

Cloud distribution is the real story here — not just model branding. For years, OpenAI’s commercial setup was basically tied to Microsoft Azure, which gave Microsoft a powerful choke point over how OpenAI’s models reached big customers. On April 27, 2026, that changed. OpenAI and Microsoft rewrote the partnership so OpenAI can now serve products across any cloud provider, while Microsoft stays the primary cloud partner and keeps important rights through 2032. ### What actually changed? The big shift is simple. OpenAI can now sell and run its products for customers on AWS, Google Cloud, or Azure instead of being commercially boxed into Azure. Microsoft still gets first shot in an important way — OpenAI products are supposed to ship first on Azure unless Microsoft can’t support the needed capability and chooses not to. But exclusivity is gone. (openai.com) ### Why is that a bigger deal than it sounds? Because cloud placement decides who owns the customer relationship, the security boundary, the compliance story, and a lot of the margin. If a bank, retailer, or government agency wants OpenAI models but has standardized on AWS or Google Cloud, OpenAI can now meet that customer where it already operates instead of forcing a migration toward Azure. That makes OpenAI easier to buy — and harder for Microsoft to fully gatekeep. (blogs.microsoft.com) ### So is Microsoft losing the partnership? No — but the partnership is clearly looser. Microsoft remains OpenAI’s primary cloud partner, keeps a license to OpenAI IP for models and products through 2032, and still participates as a major shareholder. The important nuance is that the IP license is now non-exclusive. Microsoft still has privileged access. It just no longer has the whole field to itself. (blogs.microsoft.com) ### What happened to the money flows? That changed too. Microsoft will no longer pay a revenue share to OpenAI under the new phase of the deal. But OpenAI will keep paying Microsoft a revenue share through 2030 at the same percentage as before, subject to a total cap. In plain English, Microsoft still gets paid when OpenAI grows, but the upside is now bounded rather than open-ended. (openai.com) ### Why mention AWS specifically? Because this isn’t just a theoretical contract rewrite. One day after the partnership reset, OpenAI announced that GPT models, Codex, and Managed Agents were coming to AWS for enterprise customers. That makes the “any cloud” language feel operational, not symbolic. OpenAI had already announced a broader AWS infrastructure partnership months earlier, so the distribution opening and the AWS product launch line up cleanly. (openai.com) ### What about Google Cloud? The April 27 language says OpenAI can serve products across any cloud provider. That clearly covers Google Cloud in principle. But the public rollout has been much more concrete on AWS so far. So the safest way to read this is: Google Cloud is now allowed by the structure, even if AWS is where the first visible expansion landed. (openai.com) ### Why did the old model break? Basically, OpenAI got too big to be commercially constrained by one hyperscaler. It needs more compute, more distribution paths, and more leverage in negotiations. Microsoft still matters enormously — especially because Azure remains the first destination for launches — but OpenAI now has room to pursue a multi-cloud strategy without pretending the old exclusivity still fits reality. (blogs.microsoft.com) ### Bottom line This is OpenAI becoming a real multi-cloud platform company. Microsoft is still inside the tent — deeply. But the old Azure-only era is over, and that changes who can sell, host, and profit from OpenAI’s models over the next few years. (openai.com) (blogs.microsoft.com)