Microsoft’s sudden embrace of open-weight AI isn’t about altruism. It’s about Azure lock-in and fattening margins, and the company isn’t even trying to hide it. A new open letter, co-signed by Meta, Nvidia, and Hugging Face, frames open models as a patriotic bulwark against Chinese AI dominance while defending distillation as legitimate innovation. The timing is no accident. The Trump administration is reportedly eyeing restrictions on Chinese open-weight models, and Microsoft is scrambling to protect a pipeline it now depends on.
But the real story is what’s happening inside Microsoft 365. The company is quietly swapping OpenAI and Anthropic models in Copilot, Excel, and Outlook for its own MAI family of models. CEO Satya Nadella pitched this as using “the right model for each task” and keeping costs down. What he didn’t say is that independent benchmarks place MAI’s performance closer to DeepSeek V3.2 than to anything from OpenAI. Customers are getting weaker AI for the same price, while Microsoft pockets the savings from running inference on older Nvidia H100 and A100 GPUs.
Microsoft’s own blog admits this swap “significantly lowers the cost of deployment for Microsoft.” That’s an honest sentence buried in a sea of spin. The company’s superintelligence team ran benchmarks comparing MAI only against smaller models like GPT-5.4 Mini and Anthropic’s Haiku — and when they cite “GPT-5.6,” they conveniently omit whether that means Sol, Terra, Luna, or what reasoning mode was active. Anyone who’s run a real eval knows that’s not how you do benchmarks. Microsoft certainly knows it.
The broader strategy is classic Redmond. A fragmented model market means no single AI lab — not OpenAI, not Anthropic — grows powerful enough to threaten the Windows and Office ecosystem. More models on Azure give enterprises fewer reasons to look elsewhere. What looks like an open-weight love letter is really an insurance policy. The company that invented software lock-in hasn’t changed. It just found a new way to write the contract.
