The Physical Economy Behind AI

A digital service needs physical capacity

Training and serving AI models require chips, networking, buildings, cooling and dependable electricity. That puts physical infrastructure inside a digital investment thesis.

In its 2025 Energy and AI base case, the International Energy Agency projected global data-centre electricity use to roughly double to 945 TWh by 2030. That is a scenario, not a guaranteed demand order for utilities or equipment manufacturers.

Follow the bottlenecks

A company may announce a large project well before it produces revenue. Investors should compare orders with backlog conversion, capital expenditure and cash collection. They should also distinguish a one-time construction benefit from a recurring service stream.

A broader opportunity, with shared risk

Owning a semiconductor company, an electrical-equipment manufacturer and a utility may look diversified by sector. All three can still depend on the same data-centre spending cycle. The SEC's guidance on diversification is a useful reminder to inspect common drivers, not just labels.

Model who earns money if AI spending grows more slowly than expected. Contractors with firm orders, regulated utilities and speculative capacity builders may respond very differently.