Microsoft and OpenAI just rewrote their deal. Exclusivity is dead, and so is the AGI clause.

Microsoft and OpenAI announced a rewrite of their partnership on April 27, ending the exclusivity that has defined the deal since 2019. The pair confirmed in a joint blog post that Microsoft’s license to OpenAI’s intellectual property is now non-exclusive, and that the revenue-sharing flow between the two companies has been reshaped from the ground up.

What actually changed

Three things flipped at once. The first: Microsoft’s license to OpenAI’s models, which previously locked OpenAI to Azure for distribution, is now non-exclusive. As Microsoft put it, “OpenAI can now serve all its products to customers across any cloud provider.” Azure remains the primary cloud partner, and new OpenAI products will still ship first on Azure unless Microsoft can’t or won’t support them. But the rest of the OpenAI catalog can now run on AWS and Google Cloud the day it launches.

The second: revenue share now flows in one direction. Microsoft has stopped paying OpenAI a revenue share on OpenAI products it resells, per Bloomberg. OpenAI keeps paying Microsoft, but at the same percentage as before, with one big change.

That change is the third one. The OpenAI-to-Microsoft revenue share is now capped at a fixed total dollar amount, will expire in 2030, and is no longer contingent on OpenAI declaring artificial general intelligence. The old deal had a clause that ended Microsoft’s revenue cut the moment OpenAI’s board judged it had reached AGI. That clause is gone. The license to OpenAI’s IP runs through 2032 either way.

The Microsoft side of the math

Look at this from Redmond’s seat for a second. Microsoft gives up exclusivity on a model lineup it spent the last seven years co-building. In exchange, it stops paying OpenAI on resold products and gets a guaranteed revenue stream from OpenAI through 2030 that no longer evaporates if a research milestone gets reached.

Wall Street wasn’t sure how to read it. Microsoft shares slipped on the news, Sherwood reported, while Amazon initially popped before giving back its gains. The Motley Fool framed the change as Microsoft trading some moat for predictability, and that reading lines up with what got removed: the AGI tripwire was an open-ended risk that has now been priced into a cap.

The other thing Redmond keeps is its position as primary cloud partner and the right of first refusal on new launches. If OpenAI ships a successor to GPT-5.5 next quarter, Azure still gets it on day one. The shift only kicks in once a product is in the catalog. We covered the GPT-5.5 launch on April 23 when API tokens doubled in price; under the new terms, GPT-5.5 ships on Azure first and then anywhere else OpenAI wants to run it.

What it unlocks for OpenAI

The other side is more interesting. OpenAI now has a clean path to multi-cloud distribution in a market where its biggest customers want exactly that. The company has been pursuing a $50 billion compute deal with Amazon Web Services, and the old exclusivity terms reportedly created legal complications around shipping OpenAI models on AWS. That blocker is now gone.

It also resets the competitive frame. Anthropic just locked in up to $40B from Google on April 24, plus five gigawatts of TPU capacity, five days after Amazon committed $5B at a $100B-cloud-spend tier. For most of the past year, Anthropic has been the AI company with optionality across hyperscalers and OpenAI has been the one with the deepest single partner. As of today, that asymmetry is gone.

The independence cuts both ways though. OpenAI still pays Microsoft a revenue share on its own resold products through 2030, and Microsoft holds the IP license through 2032 regardless. So while OpenAI gets distribution freedom, Microsoft gets capped, deterministic revenue from a partner whose exit ramp is now a calendar date instead of a research declaration.

What this means for you

If you build on the OpenAI API today, the immediate practical change is that Azure is no longer the only first-party path. Once OpenAI’s catalog finishes migrating to other clouds, you’ll be able to pick a hyperscaler for non-technical reasons. Compliance, region, existing spend commits, the way your auth stack is wired up, all of it.

For builders weighing Anthropic versus OpenAI, the multi-cloud parity question is now off the table. Both are deployable across all three major clouds. The decision goes back to model behavior, pricing, and tooling. If you’re tracking AI vendor concentration risk on your architecture review, the Microsoft-OpenAI rewrite is the line you should record as the moment exclusivity stopped being a planning constraint. The next time someone tells you “we’re locked into Azure for OpenAI,” they’re describing 2025, not 2026.