AI and Power Concentration: Why a Few Labs Matter So Much
This page is a structured working draft — real analysis, not yet expanded with the full expert sourcing given to the flagship pages. Safe to build on; treat specifics as provisional until sourced.
The mechanism
Frontier AI development requires capital, compute, and talent at a scale only a handful of companies and states can marshal. Because capability currently compounds — a more capable model earns more revenue and attracts more investment, funding the next, more capable model — early leaders tend to extend their lead rather than get overtaken by fast followers the way earlier tech waves sometimes saw.
Corporate concentration
A small number of firms (and the cloud providers and chipmakers underneath them) now sit at a genuine chokepoint for a general-purpose technology increasingly embedded in government, finance, healthcare, and media. That is a different kind of risk than a normal industry monopoly: the “product” is a capability that shapes what information reaches people and how decisions get made across many other industries at once.
State-level concentration
The same dynamic plays out between nations. Export controls on advanced AI chips are explicitly designed to slow a small number of states from compounding a lead the way companies do — with all the geopolitical friction that implies (see Geopolitical Race Dynamics).
Why this is a governance problem, not just an economic one
Concentrated power over a general-purpose reasoning technology is a standing invitation to abuse, regardless of whether any single actor currently intends to misuse it — the concern is the structural incentive, not any particular company’s stated values today. This is why several governance proposals (see Averting Loss of Control) focus on distributing frontier AI development across multiple independent, accountable labs rather than trusting any one organization’s internal restraint.