OpenAI can now serve every product on every cloud. Microsoft's IP license is non-exclusive through 2032. The AGI clause is gone. Microsoft no longer pays revenue share to OpenAI. OpenAI continues paying Microsoft through 2030 at the same percentage but with a total cap.

That is the new agreement, announced jointly on April 27. The headlines have called it a divorce. For the agentic commerce stack, it is closer to a distribution unlock.

The agent layer just stopped being a cloud-specific product. That is the part of this deal that matters to merchants and payment infrastructure, and it is the part the headlines have buried.

What The Deal Says

The amendment swaps exclusivity for portability. Under the original arrangement, Azure held first call on OpenAI workloads, an arrangement that turned the model into a Microsoft enterprise distribution asset. Under the amendment, OpenAI's products ship across any cloud, with Azure remaining the primary partner only when Microsoft can support the relevant capability.

Microsoft retains its license to OpenAI IP through 2032, but the license is non-exclusive. The AGI clause that would have triggered a license review at some future declaration of artificial general intelligence has been removed. The agreement now runs the full term regardless.

On revenue, the flow is one direction. OpenAI continues to pay Microsoft a share of revenue through 2030, at the same headline percentage as before, but capped at a total amount. Microsoft's reciprocal payment to OpenAI ends.

Equity is unchanged in shape. Microsoft remains a major shareholder in the restructured OpenAI entity. The revenue mechanics shift in Microsoft's favor on percentage and against it on cap. The exclusivity lever is gone.

What Microsoft Gave Up Versus What It Kept

Microsoft surrendered the asset that made OpenAI a Microsoft moat. Azure was the only place enterprises could run frontier OpenAI models at scale. That advantage is now gone. AWS announced Bedrock managed agents with OpenAI in the same week, a partnership the old contract did not allow.

What Microsoft kept is the upside. Equity in a company insiders value at over $850 billion. A 2032 license that lets Microsoft ship products with OpenAI IP for the next six years. A revenue stream from OpenAI that continues even if Anthropic wins the next benchmark cycle.

This is not a divorce. It is a renegotiation Microsoft accepted because the alternative was worse. Holding OpenAI to exclusivity was no longer realistic. Anthropic's enterprise deals are scaling. Google's Gemini is in production at large enterprises that want a non-Microsoft option. The exclusivity lever was already losing value. Microsoft traded an asset that was depreciating for a longer license and a revenue cap.

Why OpenAI Needed The Change

OpenAI's pressure is competitive, not contractual. Anthropic's enterprise traction is real. Google's vertical integration of Gemini into the rest of its stack is real. AWS has a customer base Microsoft cannot reach, and the FedRAMP Moderate authorization OpenAI just cleared opens government procurement that Azure-only access could not unlock.

The Bedrock partnership is the immediate beneficiary. Sam Altman and Matt Garman framed it as managed agents for the AWS customer base, with OpenAI models as one option among several. That is exactly what Microsoft's old exclusivity prevented.

OpenAI needed to be on every cloud because every cloud is where its enterprise customers already live. The new agreement is the contractual move that lets it follow them.

The Commerce Angle

This is where it stops being an AI industry story and starts being a payments story.

Agentic commerce infrastructure runs on cloud. Stripe's checkout extensions for OpenAI, the Google partnership, Adyen's agent integrations, the issuer-side agent rails Visa and Mastercard are building. All of it lives on top of the cloud the merchant chose. When the AI model layer was Azure-only, the integrations narrowed. With portability, the integrations widen.

A merchant on AWS no longer has to route through Azure to reach OpenAI agents. A bank on Google Cloud no longer has to dual-host. The agent layer follows the data, not the contract.

This matters because agentic commerce volume is going to land disproportionately at large enterprises. Those enterprises chose their cloud years ago. Forcing them to integrate to a model layer hosted somewhere else was friction the agentic stack could not afford. The amendment removes that friction.

What To Watch

Three signals will show whether the unlock is real.

The first is enterprise contracts. Watch which large merchants and issuers move to OpenAI agents on AWS or Google Cloud over the next two quarters. The volume tells the story the press release cannot.

The second is the cloud agent SDKs. AWS Bedrock managed agents, Google's Vertex agent tooling, Microsoft's Copilot stack. Each cloud is now competing on agent ergonomics, not on access. The cloud that ships the most usable agent runtime will end up hosting the most agentic commerce volume.

The third is the liability question. When an agent runs on AWS Bedrock and transacts on a Stripe rail and authorizes against a Mastercard token, the failure mode crosses three contracts. The amendment opens that door wider. None of the three parties has yet said how they will close it.

Sources

When the agent runs on every cloud, which one carries the liability for what it spends?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.