Nadella Pushes Multi-Model Enterprise AI as Azure OpenAI Assistants API Retires

Nadella's multi-model pitch frames Microsoft Foundry as a marketplace of thousands of models, arguing that enterprises betting everything on a single model risk being locked out of better options — an implicit jab at Sundar Pichai's Gemini-centric strategy. Microsoft reported AI demand broadening beyond early adopters and 'wall-to-wall' Copilot deployments across information workers, backed by Azure annual revenue surpassing $100B and a $678B remaining-performance-obligations backlog.

The strategy showed up concretely in Foundry adding five new Claude capabilities — Structured Outputs, Web Search, Web Fetch, MCP Connector, and Tool Search — and in Copilot's August update introducing model choice across OpenAI and Anthropic models, Cowork workflow automation, and Copilot Notebooks. The flip side: the Azure OpenAI Assistants API was deprecated and retired on August 26, forcing teams to migrate to Foundry agent builders.

The retirement is a real migration burden that some developers will resent, but it reflects Microsoft consolidating around its agent platform. Competitively, the multi-model bet hedges against any single lab's stumbles — relevant given Anthropic withholding Model 2 and reported Opus 5 disappointment. Watch whether enterprises value model optionality enough to tolerate the added complexity, and how the Assistants API sunset affects existing production apps.