Skip to content
AI supply chain term

GPU / AI accelerator

Graphics Processing Unit — the massively-parallel chip that does most AI training and inference.

What it means

A GPU, or AI accelerator, is the massively parallel chip that does most of the heavy lifting in AI — both training models and running them. Where a general-purpose processor handles a few tasks at a time, an accelerator runs thousands of simple calculations at once, which is exactly what neural networks need. These chips are the most visible and expensive part of an AI cluster, but they cannot work alone: each one needs HBM memory beside it, advanced packaging to assemble it, a leading-edge foundry to print it, and high-speed networking to link it to thousands of peers. The accelerator is the star, but its supply is governed by everything beneath it.

Why it matters to investors

Accelerators are where the demand narrative is loudest, but the supply is set by upstream constraints — memory, packaging, foundry slots. That gap between demand and what can actually ship is exactly where the AI trade gets interesting: a dominant accelerator vendor can be the most valuable name in the chain and still be capacity-limited by parts it does not make.

See GPU / AI accelerator in the AI value chainIts live model score, why it matters, and every company exposed to it.

Companies on this part of the chain

NVIDIANVDAAMDAMDBroadcomAVGO

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

ASIC (custom AI chip)Application-Specific Integrated Circuit — a custom chip built for one job, e.g. AI inference.HBM (High-Bandwidth Memory)High-Bandwidth Memory — DRAM stacked next to an accelerator to feed it data fast; often the real limit on GPU output.CoWoS (Chip-on-Wafer-on-Substrate)Chip-on-Wafer-on-Substrate — the advanced packaging that fuses compute dies and memory into one accelerator.InferenceRunning a trained model to answer requests — the recurring cost of AI in production.Training (pretraining)The large, compute-heavy run that creates a base model from scratch.FLOPs (compute throughput)Floating-point operations per second — a raw measure of compute throughput.

See GPU / AI accelerator 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.