Microsoft pushes its own AI models
Microsoft has begun rolling out its own AI models and reducing operational dependence on OpenAI, a strategic shift covered this weekend that emphasizes platform depth over a single external partner. Commentators say the move aims to deepen Azure’s ecosystem control while preserving multi‑model optionality for large customers. (tradingkey.com) (seekingalpha.com)
Microsoft is putting its own artificial intelligence models into Azure, giving customers more ways to build without relying only on OpenAI. (azure.microsoft.com) Microsoft’s Azure AI Foundry model catalog now says it includes “Microsoft AI models” alongside models from OpenAI, Anthropic, Meta, Mistral AI, xAI and others. The product page says customers can choose from more than 11,000 models and switch among them through a shared inference interface. (azure.microsoft.com) At Microsoft Build in May 2025, the company said Azure AI Foundry had grown to more than 70,000 customers, processed 100 trillion tokens in the prior quarter and powered 2 billion daily enterprise search queries. In the same announcement, Microsoft said it was expanding the catalog with new outside models and adding a router that automatically selects the best model for a prompt. (azure.microsoft.com) That router matters because it turns model choice into an infrastructure feature. Microsoft’s documentation says the router can pick among supported large language models in real time to reduce cost and latency while keeping comparable quality. (learn.microsoft.com) The shift also redraws a partnership Microsoft spent years deepening. In January 2023, Microsoft said it was entering the third phase of its OpenAI relationship, extending a multiyear, multibillion-dollar investment and keeping Azure as OpenAI’s exclusive cloud provider. (blogs.microsoft.com) That 2023 deal did not stop Microsoft from commercializing its own systems. The same announcement said each company would be able to “independently commercialize” advanced artificial intelligence technologies, leaving room for Microsoft to sell both OpenAI models and its own. (blogs.microsoft.com) Microsoft has framed Azure AI Foundry as a full stack for building agents and applications, tying models to GitHub, Visual Studio Code and Azure cloud services. Asha Sharma, Microsoft’s president for CoreAI Product, wrote in May 2025 that the platform had moved from an application layer to a “full-stack platform” for enterprise agents. (azure.microsoft.com) The company’s model page now presents that stack as explicitly multi-model: Microsoft’s own models sit beside OpenAI and rival providers in one marketplace. For large customers, that means one contract, one platform and more leverage over price, speed and compliance requirements. (azure.microsoft.com; learn.microsoft.com) Microsoft is not walking away from OpenAI in public. Its Azure site still promotes Azure OpenAI in Foundry Models, and Microsoft’s 2023 partnership post remains live, but the newer Foundry rollout shows the company building a broader platform that does not depend on a single model supplier. (azure.microsoft.com; blogs.microsoft.com; azure.microsoft.com) The result is a quieter change than a breakup headline: Microsoft is turning Azure into the place where customers can buy OpenAI, Microsoft and rival models side by side — and let Microsoft’s software decide which one runs. (azure.microsoft.com; learn.microsoft.com)
