Microsoft AI forms the MAI Superintelligence Team

Background

Microsoft AI had released its first in-house models a few months earlier, in August 2025, and its first image model in October 2025.^[2] On 6 November 2025 the company announced a dedicated team to pursue longer-range goals.^[1]

What happened

On 6 November 2025 Mustafa Suleyman announced the MAI Superintelligence Team within Microsoft AI. He led the team himself.^[1][2] Superintelligence is a term for AI that would surpass human ability across most tasks. The team's stated goal, in Microsoft's words, is "humanist superintelligence": AI that is focused on specific problems and domains, with limits on autonomy so that humans stay in control.^[1]

The phrase sets the team apart from a general push for the most capable system possible. It asks that the systems be built around particular purposes and that human oversight be part of their design.^[1]

Key details

Suleyman named medical superintelligence as one of the goals. He described it as AI with expert-level diagnosis and clinical planning, and he expected it to arrive "in the next few years".^[1] As evidence of progress in this area he cited Microsoft's MAI-DxO orchestrator, which reached 85% accuracy on New England Journal of Medicine case challenges.^[1] An orchestrator here is a system that coordinates several AI components to work on one problem together.

Two other domains were named in the announcement. One was a personal AI companion for everyone. The other was plentiful clean energy.^[1] Both were presented as areas where the same approach could be applied, not as products already in use.

Suleyman also said the team had a next-generation GB200 compute cluster in operation.^[1] Computing capacity of this kind is the precondition for training large models, so the statement tied the team's goals to physical infrastructure that was already running.

Impact

The team's creation gave Microsoft AI a named unit with a long-range mission, separate from the day-to-day release of models such as MAI-1-preview and the image and speech models. Its stated framing, problem-oriented systems with human control built in, is a position that readers can compare with other companies' descriptions of their own goals.^[1][2]

The announcement is also a statement of intent. Several of the goals it names were, at the time, described as things expected in future years rather than achievements. Readers should keep that distinction in mind when they read the claims about medicine, companions and energy.^[1]