Glossary term
Compute (AI)
The general term for processing capacity used to train or run AI models; measured in FLOPS, TOPS, or GPU-hours.
Compute is the general term for the processing capacity used to train or run AI models, measured in FLOPS, TOPS, or GPU-hours depending on context. It is the central input to the scaling-laws thesis, where more compute reliably yields more capability. Compute demand splits into training (one-time, capex-heavy) and inference (recurring, opex-heavy), with the balance shifting toward inference as agentic workloads grow. See Inference Economics 101.