What it means
Mixture of Experts (MoE) is a neural-network architecture that splits a model into many specialized sub-networks, or experts, and activates only a few of them for each input rather than the whole model. A small routing network decides which experts handle each token, so a model can hold a very large total number of parameters while using only a fraction of them per step. This makes MoE models cheaper to run for their size and improves the balance between training cost and inference cost, a core reason several recent efficient models adopt it. In the AI stack it is a design lever that raises capability per unit of compute, but it complicates memory, networking, and load-balancing because the experts must be distributed across many chips and tokens routed between them. That places fresh demands on interconnect and memory bandwidth.
Why it matters to investors
MoE improves capability per unit of compute, which is why cost-efficient labs like DeepSeek and Moonshot AI have leaned on it. It shifts some pressure from raw FLOPs toward memory bandwidth and interconnect, since experts are spread across many chips and tokens must be routed between them.
Companies on this part of the chain
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
See Mixture of Experts 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.
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.