Talent, Research Culture and Building an AI Company

In most industries capital is scarce and people are replaceable. At the frontier of AI in 2023–26 the reverse often held: money was abundant and a few hundred researchers who knew how to train and tune the largest models were not. That scarcity produced pay packages reported in the hundreds of millions of dollars, licence-and-hire deals in which Big Tech paid billions mainly for teams, seed rounds of $1–2bn for labs with no product, and a summer in 2025 when Meta's chief executive personally recruited rivals' researchers. It also made governance fragile. OpenAI's board fired its chief executive in November 2023 and reversed itself within five days when nearly all its staff threatened to leave, and the episode changed how every lab thinks about mission, control and retention. This section explains how the AI talent market works and why it pays what it does, how labs are governed and why their unusual structures exist, how research culture differs from product culture, and how founders build AI companies, from the team to compute access and the order of hires, with the 2025–26 cases as evidence.