Microsoft and OpenAI have reset the terms of one of the most important partnerships in artificial intelligence. The new agreement gives OpenAI more operational independence while preserving major rights for Microsoft around models, products, cloud access, and AGI.
The deal follows OpenAI's recent restructuring and places Microsoft at about a 27 percent stake in the restructured OpenAI Group Public Benefit Corporation (PBC). According to Microsoft's announcement, that stake is valued at roughly $135 billion.
What stays in place
The new agreement changes the balance of control, but it does not break the core partnership. OpenAI remains Microsoft's partner for so-called frontier models, keeping the relationship central to both companies' AI strategies.
Microsoft also keeps exclusive IP rights and exclusive access to the Azure API until OpenAI reaches artificial general intelligence, or AGI. That AGI clause had been a major sticking point in the negotiations, along with OpenAI's push for more independence.
The new structure appears designed to keep the commercial engine running while reducing some of the constraints around OpenAI's business. In practical terms, Microsoft continues to hold important rights tied to OpenAI technology, while OpenAI gains more flexibility in how it builds, partners, and deploys certain products.
Where OpenAI gains more room
OpenAI now has broader freedom to work with other companies. The company can co-develop products with third parties, which gives it more space to build beyond the Microsoft relationship.
Cloud rules are now more divided by product type. API-based products still have to run exclusively on Azure. Non-API products, however, can be hosted on any cloud provider.
That distinction matters because it separates OpenAI's API business from other product categories. The agreement keeps Azure central to API delivery, but it no longer makes Azure the only possible cloud home for every OpenAI product.
OpenAI also gains room in government-facing work. The company can offer API access to US government clients in the national security space, even when those clients use a cloud provider other than Azure.
The company is also allowed to release open-weight models, provided those models meet certain criteria. The source does not specify those criteria, but the permission itself marks another area where OpenAI now has more flexibility.
Another major change concerns compute. Microsoft no longer has the right of first refusal as OpenAI's compute provider. In return, OpenAI has committed to buying an additional $250 billion worth of Azure services.
What Microsoft receives
Microsoft's position also changes in important ways. Its IP rights to OpenAI models and products now extend through 2032. Those rights also cover models developed after AGI is achieved, with safety requirements attached.
Microsoft's rights to confidential research methods, described as "Research IP," last until AGI is officially recognized or until 2030. There are exceptions: model architectures and weights are among the areas where Microsoft still has rights.
Consumer hardware from OpenAI is explicitly carved out of these IP rights. That exclusion creates a boundary around what Microsoft can claim under the agreement, even as its model and product rights remain broad.
The deal also gives Microsoft a new path on AGI. Microsoft can now pursue its own AGI development independently or with partners. The source states that this option was previously off the table until 2030.
There is still a constraint if Microsoft uses OpenAI technology before AGI is formally declared. In that case, the agreement sets strict compute thresholds for those models.
Who decides when AGI has arrived
One of the most consequential changes is about the definition of the milestone itself. AGI status will no longer be determined by OpenAI alone.
Instead, an independent panel of experts will make the decision. That change shifts the AGI trigger away from a single company's internal judgment and places it in the hands of a separate review process.
This matters because AGI is not just a technical label inside the agreement. It affects rights, access, revenue-sharing, and the timeline for parts of the Microsoft-OpenAI relationship.
Revenue-sharing between the companies continues until AGI is verified. The payments are now structured to stretch over a longer period, according to the source.
Why the deal matters
The agreement is a compromise between independence and continuity. OpenAI receives more freedom to work with third parties, host some products outside Azure, serve certain US government clients across cloud environments, and release open-weight models under criteria.
Microsoft, meanwhile, keeps a large financial stake, long-running IP rights, exclusive Azure API access until AGI, and new freedom to pursue AGI on its own or with partners. The company also secures a major Azure commitment from OpenAI.
The clearest theme is that both companies wanted more optionality without ending the relationship. OpenAI gets broader operating room after restructuring. Microsoft keeps deep access to the technology and a stronger set of rights extending into the next phase of the partnership.
The expert panel may become the most important part of the arrangement. By moving the AGI determination outside OpenAI alone, the agreement ties a defining milestone to a separate decision process. For a partnership built around frontier models, cloud infrastructure, IP rights, and AGI, that change could shape how both companies plan their next moves.