Chamath's megacorp thesis and Soren Larson's case that labs will become conglomerates

ericwdolan · x · 2026-11-22

Chamath argues 9–10 megacorps will be built on AI over the next decade, each going "from the application layer all the way down to the dirt" — citing Anthropic, OpenAI, SpaceX (which owns Cursor), and Meta. Soren Larson's essay "The Conglomerate Premium" backs this up: Google reportedly beat Mercor's $7.5M bid with $10M for Spirit Airlines' full operational dataset (docs, email, workflows, codebase), after reportedly acquiring RL-env developer Mechanize. Frontier labs vertically integrating over data supply chains is the dominant strategy — paying Mercor/Handshake both capitalizes competitors and depreciates your own models' data alpha. The "Coasean Singularity" is wrong: the only practical way to hill-climb long-duration, hard-to-verify tasks is to own the operating companies and run them as firm-scale RL environments, so labs will look like chaebols/diversified conglomerates. Author remains bearish on data intermediaries.

Original post →

More from AGI Musings

AGI Musings channel →