Microsoft Builds Its Own AI Models as OpenAI Partnership Becomes Less Exclusive

Microsoft is reducing its dependence on OpenAI by building and launching its own frontier AI models, marking a major change in one of the most important partnerships in the AI industry. The shift does not mean Microsoft and OpenAI have fully separated. Their relationship remains large and financially important. But Microsoft is no longer relying on OpenAI as the only foundation for its AI strategy.

The change follows two major developments over the past several months. The first came in October 2025, when OpenAI completed its conversion into a Public Benefit Corporation under a nonprofit parent called the OpenAI Foundation. As part of that restructuring, Microsoft formalized its position as a roughly 27 percent stakeholder, valued at about $135 billion. That came after total Microsoft investment of around $13.8 billion since 2019.

The deal also protected Microsoft’s access to OpenAI technology for a longer period. Microsoft secured intellectual property licensing rights through 2032, including access to post AGI models. That was important because the earlier agreement could have allowed OpenAI to cut off Microsoft’s access once OpenAI declared artificial general intelligence had been achieved. Under the newer structure, that milestone would be judged by an independent expert panel, not simply by OpenAI’s own board.

OpenAI can now sell through more cloud providers

The second major change came on April 27, 2026, when OpenAI’s cloud and API relationship became less exclusive. OpenAI is now free to sell its products and APIs through other cloud providers. Since then, OpenAI has signed major compute deals with AWS, Oracle, Google Cloud, and CoreWeave.

This is a major shift because OpenAI was previously tied much more closely to Azure for frontier model access. Microsoft still benefits from its OpenAI stake and licensing rights, but it no longer controls the only major cloud path for OpenAI’s products.

At the same time, Microsoft gained more freedom to compete directly. The company is no longer locked into a strategy where OpenAI supplies the most important model layer. That opened the door for Microsoft to push its own in house AI models more aggressively.

Microsoft’s MAI models show a more independent AI strategy

At Build 2026, Microsoft introduced seven in house AI models under the MAI brand. The most important one is MAI Thinking 1, Microsoft’s first reasoning model trained from scratch. The company says it was trained on commercially licensed data and was not refined from OpenAI or another third party model.

That detail matters for enterprise customers. Large companies often care not only about performance, but also about data origin, legal risk, compliance, and whether a model depends on another vendor’s API. A Microsoft built model trained on commercially licensed data gives businesses another option inside Azure.

MAI Thinking 1 uses a sparse Mixture of Experts architecture with about 35 billion active parameters and a 256,000 token context window. Microsoft CEO Satya Nadella described the move as a shift from consuming a frontier model to participating directly at the frontier.

The rest of the MAI lineup covers image generation, transcription, voice, and coding. MAI Code 1 Flash is already rolling out to GitHub Copilot users. Microsoft is also making the models available through Azure AI Foundry and third party platforms, including Fireworks AI, Baseten, and OpenRouter.

Microsoft and OpenAI are now partners and competitors

The new arrangement creates a more complicated relationship. Microsoft still has a major financial interest in OpenAI and long term rights to its technology. OpenAI still has a huge Azure services commitment. But both companies now have more room to act independently.

For OpenAI, the change means more cloud options and less dependence on Microsoft infrastructure. For Microsoft, it means the ability to offer its own models directly to customers without routing everything through OpenAI.

This could be good for enterprise buyers. More competition may create pricing pressure, more deployment choices, and better control over where data and model workloads run. A Microsoft native model that works entirely inside Azure could be attractive for companies already committed to Microsoft’s cloud, security, and productivity stack.

The biggest question is performance. MAI Thinking 1 still has to prove itself against OpenAI’s GPT models, Anthropic’s Claude models, Google’s Gemini models, and other frontier systems in real workloads. Model quality, reliability, speed, cost, reasoning ability, coding performance, and tool use will all matter.

Microsoft’s message is clear, though. It does not want to rent the foundation of its AI future forever. The company helped fuel OpenAI’s rise, but it is now building its own model family, shaping its own AI stack, and preparing to compete more directly in the same market it helped create.