Mustafa Suleyman, head of Microsoft AI, told VentureBeat at Build 2026 that a contract change with OpenAI about six months ago gave his division the formal freedom to pursue "superintelligence" using Microsoft’s own researchers, data pipelines and custom silicon. The shift marks the first time the company can develop large‑scale models without the restrictions that previously limited its own AGI work.
At the same event Microsoft unveiled a family of seven in‑house models under the MAI brand, covering reasoning, code generation, image creation, transcription and voice synthesis. The flagship MAI‑Thinking‑1 is a 35‑billion‑parameter reasoning model that matches peers on software‑engineering benchmarks, and all models were trained from scratch on clean, licensed data rather than distilled from external systems. Suleyman stressed that the seven models are a proof‑of‑concept; the longer‑term goal is a dedicated lab that can produce frontier‑scale AI independent of third‑party providers.
Alongside the models, Microsoft introduced "Frontier Tuning," a service that lets enterprises fine‑tune MAI models with proprietary data inside secure boundaries. Early pilots with Mayo Clinic, EY, Land O'Lakes and Pearson show cost‑effective performance gains. New products such as the Scout autopilot agent, Windows 365 for Agents, and updates to the Foundry platform illustrate Microsoft’s push from conversational IQ toward actionable AQ agents that can operate across enterprise software.
