The cap table became the competition

At Microsoft's Build conference in San Francisco in early June, Satya Nadella, the company's chief executive, told a room of developers that the age of merely renting intelligence was ending. "We believe the time has come," he said, "for every company to move from consuming a frontier model to fully participating at the frontier." The remark was framed as an invitation, though it read more like a declaration of independence. Microsoft has put roughly $13 billion into OpenAI and as much as $5 billion into Anthropic. Minutes earlier, it had unveiled seven of its own models built to compete with theirs.

The seven, badged MAI, were trained in-house; the flagship, a reasoning system called MAI-Thinking-1, was built from scratch rather than distilled from a larger rival, and Mustafa Suleyman, who runs Microsoft's AI division, claimed that a version tuned for the consultancy McKinsey had bested OpenAI's GPT-5.5 on quality at roughly a tenth of the cost. Whether that figure survives contact with independent benchmarks is a separate question. The signal was the point: the largest financial backer of the two leading AI labs now intends to sell against them, on its own cloud, at a price they will struggle to match.