Chips, data centres and energy
AI runs on physical infrastructure: chips, memory, electricity and networks.
What is it?
AI does not live in a cloud but on physical infrastructure: chips, memory, data centres, electricity and networks. Graphics processors (GPUs) and similar chips are especially suited to the massive parallel computing that models need.
An everyday example
When you ask a chatbot a question, the computation happens in a data centre that uses electricity and water for cooling, not in your phone.
Why does it matter?
Access to chips and energy decides who can build the largest models, how much using them costs and what AI’s environmental footprint is.
A common misconception
That “digital” has no physical cost. Every answer needs hardware, electricity and infrastructure.
In more detail
Why is there so much talk about GPUs?
Graphics processors are good at doing huge numbers of calculations at the same time, which is what many neural models need.
What does training a model mean?
A long process of adjusting its parameters. It is a different job from using the model later, which is called inference.
Why does AI need so much electricity?
Large models need computing, cooling, networks and data centres. Consumption varies with the hardware, the task and how the model is used.
Can AI run locally?
Yes. Some smaller models run on a phone or a computer without a constant cloud connection. Their abilities and costs vary.
Remember: “Digital” does not mean free of physical costs.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.