Microsoft AI chief says company was 'set free' from OpenAI to pursue superintelligence

Microsoft AI chief Mustafa Suleyman publicly reframed Microsoft's relationship with OpenAI, saying the company has been 'set free' to pursue superintelligence on its own — outlining a roughly five-year plan to generate frontier AI capability internally rather than rent it. The framing is a notable shift from the years when Azure's AI story was essentially OpenAI's models on Microsoft infrastructure.

The substance behind the rhetoric is the MAI model suite unveiled at Build (MAI-Thinking-1, MAI-Code-1-Flash, MAI-Voice-2, MAI-Image-2.5, MAI-Transcribe-1.5, Alon-1.0) plus Frontier Tuning. Suleyman pointed to Artificial Analysis benchmarks showing MAI-Transcribe-1.5 leading its category as early proof the in-house effort is bearing fruit.

Strategically, this is Microsoft hedging a dependency that became a liability as OpenAI pursued its own consumer superapp and listing ambitions. Owning the model stack lets Microsoft control cost, roadmap, and differentiation across Copilot and Azure — and the redesigned Microsoft 365 Copilot, which loads twice as fast and surfaces tools contextually, is the consumer-facing expression of that control.

Caveats: 'self-sufficient superpower in five years' is a bold claim from a company still deeply intertwined with OpenAI commercially, and MAI models have yet to prove themselves against GPT-5.5, Gemini 3.5, and Claude in broad independent testing. What to watch: whether Microsoft reduces OpenAI reliance in flagship Copilot features, and how the partnership's economics evolve as both sides build competing consumer products.