Training Compute
Training compute is the total amount of computation used to train an AI model, usually measured in floating-point operations.
What is Training Compute?
Training compute for frontier models has grown roughly four to five times a year. Because it is measurable and linked to capability, regulators use it as a trigger for obligations.
How it works
It is calculated from the hardware used, the time and utilisation, or from model size and data volume.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions