There is a growing assumption that artificial intelligence can be adapted to a shrinking economy by making it smaller, slower and local. The idea is that large data centres disappear, but modest local systems continue to operate within communities. Once declining materials, declining know-how and the collapse of capital investment are taken seriously, that assumption no longer holds.

Modern AI depends on hardware that cannot be produced locally. Even the smallest systems rely on ultra-pure silicon, rare metals, chemically complex processes and precision manufacturing. These are not craft activities that can be recreated at community scale. They depend on long supply chains, high energy use and continuous industrial coordination. As those conditions weaken, hardware replacement becomes impossible. Existing machines are kept running only by scavenging parts from others, until a single failed component ends the system entirely.

Know-how declines even faster than hardware. The knowledge required to design, fabricate and maintain advanced electronics is deeply specialised and distributed across global institutions. It depends on formal education, constant practice and large teams working together. In a shrinking economy, people move into roles that meet immediate needs such as food production, care and repair. Training pipelines collapse. Tacit knowledge disappears first. What remains are machines that exist physically but cannot be meaningfully understood or repaired.

Capital investment is the third constraint, and it is decisive. Even if some hardware and some expertise remain, a contracting economy cannot justify investing scarce resources in systems that offer indirect or marginal benefit. Capital flows toward things that are essential, durable and repairable. AI hardware fails that test. It requires controlled environments, specialist components and planned redundancy. When it breaks, it is not worth replacing.

Taken together, these forces do not produce a gentle scaling down of AI. They produce a threshold effect. Once materials, knowledge and capital fall below a certain level, AI systems are no longer reproducible. From that point onward, they are not tools but legacy artefacts. They may continue to function briefly, in isolated pockets, but they have no future path.

This distinction matters. Technologies that survive deep economic contraction are those that can be remade from what is locally available. They use materials that can be sourced nearby. They rely on skills that can be passed from person to person. They require little capital and reward repair rather than replacement. Mills, hand tools, simple buildings and food systems meet these conditions. AI hardware does not.

In that sense, artificial intelligence belongs to a surplus phase of civilisation. It externalises knowledge into machines that only function while energy, materials, expertise and investment remain abundant. As surplus disappears, knowledge moves back into people, into practice, habit and shared experience.

AI therefore does not become local in a truly shrinking economy. It becomes historical. Not because it was a mistake, but because it depends on conditions that no longer exist.

Localism is not about finding smaller versions of global systems. It is about rebuilding ways of living that can be sustained, repaired and passed on using what is at hand. Once that standard is applied, the long-term fate of AI is clear.