Skip to content
AI supply chain term

Model FLOPs Utilization (MFU)

Model FLOPs Utilization is the fraction of a chip's peak arithmetic throughput that a training run actually uses.

What it means

MFU is the ratio between the useful floating-point operations a model training run achieves and the theoretical peak the hardware could deliver. Real training almost never reaches peak, because time is lost to memory access, communication between accelerators, pipeline stalls and less-than-ideal parallelism. A run at higher MFU squeezes more effective compute out of the same GPUs, which means more model capability per dollar and per megawatt. In the economics of the AI supply chain, MFU is the efficiency dial on pretraining, where compute capacity is converted into base-model capability. Raising it is largely a systems-engineering problem, spanning kernels, networking and parallelism strategy, and small improvements can meaningfully change the cost of training a frontier model.

Why it matters to investors

MFU determines how much of expensive compute capacity actually turns into model capability, so low utilization quietly wastes capital while high utilization stretches the same GPU spend further. Labs training at scale, such as DeepSeek, Moonshot AI and Zhipu AI, treat it as a core efficiency lever.

Companies on this part of the chain

DeepSeekMoonshot AIZhipu AI (Z.ai)Scale AI

Named to show where the term sits in the AI supply chain — research, not advice, and never a recommendation to buy or sell.

Related terms

FLOPs (compute throughput)Floating-point operations per second — a raw measure of compute throughput.PretrainingPretraining is the first and most compute-intensive stage of building an AI model, teaching it general skills from huge amounts of data.Collective communication (all-reduce)All-reduce is the collective communication step in which every chip in a cluster combines its partial results and ends up with the same final total — the core exchange that keeps distributed AI training in sync.GPU / AI acceleratorGraphics Processing Unit — the massively-parallel chip that does most AI training and inference.

See Model FLOPs Utilization in the live AI chain.

THE ENTITY maps every constraint onto one live model — which part is tight now, who owns it, and who gets squeezed when it moves. Plain-English reads you can check.

Explore the features
All glossary terms

THE ENTITY is an educational read on the AI supply chain — research, not investment advice. It explains how the chain works and who sits where, never price targets or buy/sell calls.