Microsoft replaces OpenAI and Anthropic models with internal MAI stack

The detail that tens of thousands of weekly queries are already routed through MAI models suggests this is not a roadmap announcement but an active migration already underway, making the capability trade-off a present-tense user experience question rather than a hypothetical.

The timing is notable given that Anthropic spent much of early July navigating U.S. export restrictions on its Fable and Mythos models (covered across multiple sources around July 1), only to now face a major distribution partner actively routing volume away from third-party models entirely. Microsoft losing appetite for external model spend compounds Anthropic's challenge: regulatory clearance matters less if the enterprise channel it was meant to serve is being quietly closed off. OpenAI faces the same structural pressure, and the GPT-5.6 tiering story from The Decoder around July 1 looks different in this light: fragmenting into multiple variants may be partly a hedge against large customers like Microsoft reducing their dependency on any single external model.

Watch whether Anthropic's enterprise contract renewal rates with Microsoft surface in either company's next earnings commentary. If Microsoft discloses a material reduction in third-party model spend before end of Q3 2026, that confirms the MAI migration is moving faster than Suleyman's public framing suggests.