OpenAI and Microsoft rework their partnership: Azure keeps priority, but loses exclusivity

On April 27, 2026, OpenAI announced a new version of its agreement with Microsoft. The changes aim to make one of the most important alliances in the AI industry simpler and more predictable: fewer ambiguous terms and clearer rules on who is responsible for what, how commercial rights are divided, and where products will be deployed. For the market, it signals that the partnership is not winding down. Instead, it is moving into a more mature phase built for scale and the long haul.

The main practical change concerns cloud infrastructure. Microsoft remains OpenAI's primary cloud partner, and OpenAI products should still launch first on Azure. But the agreement now states explicitly that OpenAI can serve its customers through any cloud provider when delivering its own products. There is only one exception: if Microsoft cannot or chooses not to support the required capabilities. For OpenAI, this significantly widens room to maneuver, especially amid rising demand for compute, capacity shortages, and large customers' desire not to depend on a single infrastructure platform.

The second set of changes concerns intellectual property and money. Microsoft keeps its license to OpenAI's models and products through 2032, but that license is no longer exclusive. This is an important shift: Microsoft remains a privileged partner, but OpenAI gets more freedom in how it commercializes its technology and with whom it builds further distribution channels.

At the same time, Microsoft will no longer pay OpenAI a share of revenue. Payments in the other direction remain, however: OpenAI will keep sharing revenue with Microsoft through 2030 at the same rate, though the payments are now subject to an overall cap and do not depend on how quickly OpenAI progresses along its technology trajectory. OpenAI also stresses that Microsoft remains a major shareholder and still participates directly in the company's growth.

The amendment did not come out of nowhere. The OpenAI–Microsoft partnership dates back to 2019 and has gone through several stages: from strategic investment and cloud exclusivity to a more complex structure involving rights to AI developments, corporate restructuring, and large commitments on computing infrastructure. As recently as October 28, 2025, the companies described a new "next stage" of cooperation, with many special provisions on rights to models, products, and future scenarios around AGI. The current update looks like an attempt to strip out unnecessary complexity and lock in a more direct arrangement: Microsoft gets long-term access and a commercial stake, while OpenAI gets more operational flexibility.

The companies themselves stress that this is not about distancing but about continuing large-scale joint work. OpenAI's statement mentions gigawatt-scale data center expansion, cooperation on the next generation of chips, and the use of AI in cybersecurity. That detail matters, because in this market a partnership is measured not only by rights to models but by the ability to build compute capacity quickly, procure hardware, optimize energy consumption, and deliver services to enterprise customers worldwide. Put simply, the contract was rewritten not as a formality but to better fit a reality in which infrastructure has become as strategic an asset as the models themselves.

What does this mean in practice? For OpenAI, less rigid attachment to a single commercial scheme and more freedom in dealing with the market and clouds. For Microsoft, it keeps its special status, access to key AI intellectual property through 2032, and a share in OpenAI's financial growth. For customers, relative predictability: Azure is not disappearing from the center of this story, but the window for more flexible ways of delivering products is getting wider.

In short, the OpenAI–Microsoft alliance is not falling apart. It is turning from a nearly exclusive pairing into a more grown-up, pragmatic partnership built for the scale and competition of the coming years.