Infrastructure · Fast-moving · Intermediate

AI Compute

The total processing capacity used to train and run AI models, and one of the field's defining constraints.

What AI Compute is

Compute is measured in floating point operations for training runs and in throughput for serving. It is the resource that scaling laws consume and the one that determines who can build frontier models.

How it works

Organisations buy, rent or reserve accelerator capacity in data centres, plan around power and cooling limits, and optimise utilisation because idle accelerators are the dominant waste.

Why it matters

Compute cost, energy use and export controls have turned an engineering variable into an economic and policy issue.

Common uses

→Training budget planning
→Capacity reservation
→Cost modelling for AI products

Watch for

✓Significant energy and water footprint
✓Concentrated supply chain

Continue exploring

Scaling Laws Pretraining Inference AI Energy and Environmental Impact