Microsoft Unveils 7 MAI Models, Targets AI Independence Through 2032 OpenAI Pact
Updated
Updated · Stratechery by Ben Thompson · Jun 4
Microsoft Unveils 7 MAI Models, Targets AI Independence Through 2032 OpenAI Pact
3 articles · Updated · Stratechery by Ben Thompson · Jun 4
Summary
Seven new MAI models are central to Microsoft’s push to give enterprises customizable AI it controls end-to-end, with Satya Nadella saying customers need their own “hill-climbing” systems rather than just API access.
Nadella said Microsoft will keep using OpenAI’s technology through their partnership to 2032, but wants MAI, OpenAI and other models to compete inside Microsoft harnesses on customer evals, routing work to the best model.
Azure capacity remains supply-constrained, and Nadella defended disciplined AI spending over matching rivals quarter to quarter, saying Microsoft is prioritizing enterprise demand, internal applications and research compute instead of selling easy GPU capacity.
Project Solara extends that strategy beyond Windows PCs into cloud-linked agent devices for enterprises, while Nadella argued software will shift to hybrid pricing—per-seat plus usage—as autonomous agents drive real marginal compute costs.
The broader bet is that Microsoft’s edge in AI will come from operating a platform across coding, security and knowledge work, not from relying on any single frontier model or device ecosystem.
With its own models and hardware, is Microsoft building a new closed AI ecosystem to replace the open PC era?
As AI agents shift software to a consumption model, how will businesses prevent unpredictable, spiraling operational costs?
Beyond taxes, what concrete benefits will Microsoft offer communities to justify its massive data center expansion?
Microsoft’s MAI Models Debut: Strategic Break from OpenAI Promises 10x Cost Savings and Enhanced Data Sovereignty
Overview
On June 3, 2026, Microsoft unveiled MAI, a family of seven in-house AI models, marking a major shift toward self-sufficiency and less reliance on external providers. This move is a direct effort to compete with other proprietary models and was highlighted by CEO Satya Nadella as a turning point for the company. The flagship model, MAI-Thinking-1, was built from scratch without using other companies’ AI, reassuring enterprise customers about data trust and lineage. Independent raters preferred MAI-Thinking-1 over Anthropic’s Claude Sonnet 4.6, and it matched Claude Opus 4.6 on key coding benchmarks, showing Microsoft’s strong progress in AI.