Concentration of Power

Democratic theory rests on a foundational premise: political and economic power should not be so concentrated in any single actor — individual, corporation, or state — that it becomes impossible to challenge or correct through legitimate democratic means. The history of democratic institutions is partly the history of mechanisms designed to prevent excessive concentration: antitrust law, separation of powers, freedom of the press, voting rights, independent courts. These mechanisms are never perfect, but they embody hard-won understanding that unchecked power corrupts and that diffusion of power is a precondition for collective self-governance. Advanced AI systems are creating new vectors for power concentration that existing institutional safeguards were not designed to address. This lesson examines what those vectors are, who currently controls them, and what the implications are for democratic equality.

The Economics of AI and Winner-Take-Most Dynamics

Building and operating frontier AI systems requires extraordinary resources: billions of dollars of specialized hardware (GPUs and custom AI chips), vast quantities of training data, large teams of highly skilled researchers and engineers, and significant ongoing compute for inference. These costs create substantial barriers to entry. Unlike software that can be copied infinitely at near-zero cost, training a frontier AI model requires capital investment that few organizations in the world can make. This creates what economists call winner-take-most dynamics. A small number of organizations — primarily a handful of large technology companies and a small number of well-capitalized AI research laboratories — are capable of training the most capable models. Organizations that cannot afford this compute and data advantage must use the models produced by these few actors, through API access or licensing. This makes the AI layer of the economy structurally different from many prior technological layers: it is not merely that large incumbents are ahead; the capital requirements make it structurally difficult for newcomers to catch up. The economic concentration is reinforced by network effects and data advantages. The company that deploys a widely-used product gathers more behavioral data, which improves its models, which attracts more users, which generates more data. This self-reinforcing loop is familiar from earlier technology platform economics — search engines, social media — but in AI it operates at a layer closer to foundational infrastructure.

AI and Political Power Concentration

Economic concentration and political power concentration are related but distinct. A government that gains access to AI-enabled surveillance, predictive analytics, and information-environment control at scale acquires tools with no historical precedent for maintaining political dominance. Several existing dynamics illustrate this. First, AI-powered surveillance combined with predictive analytics can enable a government to identify dissent before it organizes — monitoring communications for signals of discontent, identifying individuals who associate with opposition figures, and intervening before protest becomes collective action. This is qualitatively different from historical surveillance because it is not reactive; it is anticipatory. Second, AI-generated content combined with algorithmic distribution gives a government that controls those systems extraordinary influence over the information environment its citizens inhabit. A government that controls what its citizens know and believe does not need to coerce as much — consent can be manufactured more efficiently through information control. Third, concentration of AI capability in a small number of companies creates a situation where governments can compel those companies to serve state interests. In authoritarian contexts this compulsion is direct. In democratic contexts it takes the form of legal processes, national security authorities, and the political leverage that comes from threatening regulatory action. The relationship between state power and private AI capability is complex and contested in every country. None of these dynamics is uniquely caused by AI. But AI amplifies each of them substantially, and does so at a pace that outstrips the capacity of existing institutions to respond.