Microsoft seeks AI independence

Microsoft is using its Build developer conference this week to show customers and investors that its AI business can rely less on OpenAI and more on Microsoft-built models, software and cloud infrastructure. The shift follows an amended agreement between Microsoft and OpenAI that OpenAI said last month “simplifies the partnership” and adds long-term clarity. The Information reported on May 31 that Microsoft was treating the moment as an “AI Independence Day,” two months after what it called a “conscious uncoupling” from the ChatGPT maker. Wells Fargo, in a note cited by CNBC on June 1, said Microsoft’s homegrown AI push could help lift the stock to new highs. ### What changed between Microsoft and OpenAI? OpenAI said in a post last month that the companies had reached an amended agreement to simplify their partnership and support continued AI development at scale. The company did not describe that change as a breakup, and another joint statement published earlier said the partnership remained strong and that Microsoft’s intellectual-property access continued unchanged. But The Information reported that Microsoft had been reducing its dependence on OpenAI technology as the relationship was restructured. (theinformation.com) The Information also reported that Microsoft planned to use Build to prove it could operate as an AI provider without depending on OpenAI’s technology for every layer of the stack. That framing matters because Microsoft still sells OpenAI models through Azure, even as it expands Microsoft-branded offerings. ### What is Microsoft building for itself? Microsoft’s own product pages show that Azure AI Foundry has become the company’s main control plane for model access, evaluation and deployment. (openai.com) Microsoft says Foundry Models offers more than 11,000 models and includes a model router designed to optimize cost and performance at runtime. Separate Microsoft documentation published last month shows MAI image-generation models available in preview through Foundry. (theinformation.com) The Information separately reported on May 28 that Microsoft was preparing to release a suite of homegrown AI models at Build, including a coding model intended to strengthen GitHub Copilot. Microsoft has already been positioning Foundry as a place where customers can mix Microsoft, OpenAI and other models rather than commit to one provider. ### Why are investors paying attention now? (azure.microsoft.com) CNBC reported on June 1 that Wells Fargo sees upside in Microsoft’s homegrown AI push. The argument, as described by CNBC, is that Microsoft could capture more value if it owns more of the model layer instead of depending as heavily on an outside supplier. Microsoft’s scale gives that argument weight. CNBC reported after Microsoft’s April 29 earnings that the company expects $190 billion in 2026 capital spending, while Azure growth remained a central focus for investors. (theinformation.com) A company spending at that level has reason to direct demand toward its own models, routing software and cloud services where possible. (cnbc.com) ### Why does this look bigger than a model launch? Microsoft’s own marketing around Foundry emphasizes model choice, routing and governance, not just raw model quality. The platform pitches evaluation tools, deployment controls and runtime routing as core features. That suggests Microsoft is trying to win at the orchestration layer as much as at the model layer. (cnbc.com) The Information’s reporting points in the same direction. Its Build preview described Microsoft’s effort as a test of whether developers will adopt a broader Microsoft AI stack, including coding tools and infrastructure, rather than simply buying access to OpenAI models. In practice, that means hosting, routing, enterprise integration and developer workflow become part of the competition. (azure.microsoft.com) ### What should readers watch at Build? Microsoft says Build runs on June 2 and June 3 in San Francisco and online. The company’s conference site says the event will focus on “real code and real systems” from teams building and scaling AI at Microsoft. The clearest next test is whether Microsoft uses the conference to introduce additional MAI models, expand GitHub Copilot’s in-house model options, or push Azure AI Foundry further as the default layer for selecting and managing models. (theinformation.com) Build sessions and product announcements beginning June 2 should show how much of Microsoft’s AI stack it now wants customers to treat as Microsoft-native. (build.microsoft.com)