The Hidden Infrastructure Behind Artificial Intelligence
A chatbot feels weightless. Behind it sit chip factories, power plants, cooling systems and supply chains that span the planet.
Artificial intelligence is usually described as software, and the interface encourages that. You type, and an answer appears. What you do not see is the physical system that makes the answer possible.
From sand to answers
Modern models are trained on specialised chips made in a handful of advanced fabrication plants. The most advanced of those plants depend on extreme ultraviolet lithography machines that come from a single supplier. The chips are packed into data centres that need large, steady supplies of electricity and, often, water for cooling.
Each layer has its own constraints. Fabrication plants take years to build. Grid connections can take longer. Minerals for chips and power equipment are mined and refined in a small number of countries.
Intelligence at scale is an industrial process, and industrial processes have bottlenecks.
Information needs a body
Models also depend on data: text, images and code produced by people over decades. How that data is gathered, labelled and governed is a social question as much as a technical one.
Seen as a system, AI touches all five pillars. It is a technology, built on information, powered by resources, tangled in complex dependencies and shaped by society through labour markets, regulation and public trust. Understanding any one of those layers alone gives a misleading picture of where the field is going.
The useful questions are physical as well as digital. Where will the power come from? Which component is hardest to scale? Who decides how the systems are used?