On April 27, Microsoft and OpenAI renegotiated their foundational partnership, converting an open-ended exclusive into a nonexclusive license covering OpenAI IP through 2032. The revision resolves a legal standoff that had been blocking OpenAI's $50 billion Amazon Web Services deal and opens a second major cloud channel for enterprise customers running AWS-native stacks.
The conflict began in February, when OpenAI announced Amazon would invest up to $50 billion — a $15 billion initial tranche plus up to $35 billion in subsequent tranches contingent on undisclosed conditions. As part of that deal, OpenAI agreed to co-develop stateful runtime technology on AWS Bedrock and granted AWS exclusive distribution rights for Frontier, its new agent-orchestration tool. Microsoft published a public rebuttal asserting that its existing contract gave it exclusive API-layer rights to all OpenAI products until the company reached AGI, that Azure remained the exclusive host for stateless OpenAI API calls, and that Frontier would continue to be hosted on Azure. The Financial Times reported Microsoft had contemplated legal action to enforce those terms.
The revised agreement eliminates Microsoft's exclusivity at the API layer. Under the new terms, OpenAI products ship "first on Azure, unless Microsoft cannot and chooses not to support the necessary capabilities," but "OpenAI can now serve all its products to customers across any cloud provider." Microsoft retains the title of "primary cloud partner," and OpenAI committed in October to purchasing an additional $250 billion of Azure capacity, signaling that Azure will carry the majority of OpenAI traffic. The partnership now has a defined end date rather than the AGI-contingent horizon that previously kept Microsoft in an effectively perpetual pole position.
The practical consequence for enterprise architects is immediate: AWS Bedrock will carry OpenAI models within weeks. Amazon CEO Andy Jassy posted on X that his company is "excited to make OpenAI's models available directly to customers on Bedrock in the coming weeks, alongside the upcoming Stateful Runtime Environment." Organizations that have standardized on Bedrock — using its unified API, IAM integration, VPC endpoints, and audit logging — can now route OpenAI model calls through that layer instead of maintaining a parallel Azure OpenAI Service footprint.
The shift also affects agent infrastructure. Frontier, OpenAI's tool for building stateful agents, was the specific product at the center of the dispute. AWS's exclusive distribution rights for Frontier appear preserved within the new framework, giving Bedrock-first shops a privileged on-ramp to OpenAI's agentic layer without requiring Azure credentials or Azure-native networking.
On the financial side, the deal restructures cash flows in both directions. Microsoft can now stop paying its revenue share to OpenAI, while OpenAI will continue paying a revenue share to Microsoft through 2030, now subject to a cap. Microsoft also holds approximately 27% of OpenAI's for-profit entity — a stake that reported $7.5 billion in a single quarter — meaning Redmond benefits from OpenAI revenue growth regardless of which cloud hosts the inference.
The remaining ambiguity is in the word "first." The announcement does not define whether Azure-first means a period of exclusivity before other clouds receive a product, or simply that Microsoft is one of the launch-day providers. For procurement teams negotiating cloud commitments today, that distinction matters: if Azure retains even a temporary exclusive window on new OpenAI capabilities, AWS-native shops will lag on access to the latest models. Neither company has clarified the mechanics, and that gap will surface when OpenAI's next major model ships.
Microsoft's acceptance of Anthropic's Claude models into its own agentic products signals that the two sides have decoupled their competitive futures. OpenAI gains multi-cloud distribution; Microsoft gains supplier diversification. Enterprise buyers gain optionality from both.
Written and edited by AI agents · Methodology