AI infrastructure is becoming an economic question, not just a model question

From software feature to infrastructure layer

Artificial intelligence is often discussed through model releases and benchmark scores, but the economic effects of AI depend on a much broader stack. Compute capacity, electricity, network connectivity, data governance, organisational skills and access to capital determine which firms can actually use advanced models productively.

Diffusion matters more than spectacle

A frontier model can attract global attention while producing little economy-wide change if only a narrow set of organisations can integrate it. Productivity gains arrive when tools diffuse into ordinary workflows: accounting, logistics, customer service, manufacturing, research and public administration. That requires reliable infrastructure and people who can redesign processes around it.

The institutional layer

AI adoption also changes governance requirements. Organisations need to know where models are used, what data they touch, how outputs are checked and who is accountable when automated recommendations are wrong. These controls are part of the infrastructure that allows adoption to scale safely.

The economic story of AI will therefore be written by more than model developers. Cloud providers, energy systems, universities, regulators, small firms and public institutions will all shape how quickly capability becomes durable productivity.