The Physical Limits of AI Are Political Limits Too

The first mistake in thinking about artificial intelligence is to imagine it as pure mind. AI appears as language, images, code and prediction, so it tempts us to treat it as immaterial. But every answer rests on a chain of mines, fabs, data centres, transmission lines, cooling systems, logistics contracts and political permissions.

This is not new. The printing press required paper, metal type, roads and literate markets. Telegraphy required copper, poles, undersea cables and empires willing to guard them. The internet required fibre, routers, satellites and warehouses of servers. AI belongs to that history: symbolic power carried by physical systems.

The physical limit begins with computation. More parameters, more data and more inference mean more operations. Those operations require chips whose supply chains are geographically concentrated and politically exposed. A model can be copied, but frontier capacity cannot be copied without fabrication plants, lithography equipment, packaging, memory and electricity.

There is also a thermodynamic limit. Information processing consumes energy and produces heat. Engineers can improve efficiency, but they cannot abolish physics. Landauer's principle is often discussed as a theoretical floor, yet the political point is simpler: useful AI needs practical machines, and practical machines must be powered, cooled and maintained in real places.

That turns AI into an energy question. Data centres compete with households, factories, electric vehicles and public infrastructure for grid capacity. In places where power is cheap and abundant, AI is easier to host. Where grids are fragile, AI expansion becomes a social conflict over whose electricity, water, land and noise are used to produce whose intelligence.

The economic consequence is that AI may strengthen large firms before it democratizes production. The marginal cost of text can look tiny to the user, while the fixed cost of infrastructure is enormous. Railways, steel, oil, semiconductors and cloud platforms all rewarded scale before equality. AI may follow the same pattern unless public policy treats compute as infrastructure.

The philosophical lesson is modesty. AI will enlarge what institutions can perceive and automate, but it will not repeal scarcity. The future will belong not only to whoever has the cleverest model, but to whoever can organize matter around mind.