AI Investing Glossary

What is the difference between AI training and inference?

Training is teaching an AI model, a massive one-time computation that can cost hundreds of millions of dollars. Inference is running the trained model to answer every user query, forever. Both consume compute, but inference is the recurring bill that grows with usage, and it now drives a rising share of AI infrastructure demand.

Why it matters for AI investors

The mix matters for suppliers: training concentrates in frontier labs and rewards the biggest accelerator clusters, while inference spreads across every deployed application and rewards efficiency, custom chips, and capacity everywhere. A durable inference boom is the version of the AI story in which buildout spending keeps compounding.

Tracked stocks in this layer

The largest names in the Semiconductors category, ranked by market cap:

Market data as of the close on . Prices are end of day, not live.

Related reading

Frequently asked questions

Why does the training-versus-inference mix matter for chip stocks?

Training demand is lumpy and tied to a few frontier labs' budgets, while inference demand scales with end-user adoption across thousands of applications. Chips optimized for inference, including custom silicon from cloud providers, compete hardest in that second market, so the mix shapes which vendors' revenue proves most durable.

Related terms

See how this fits the whole picture in the Start Here guide, or browse the full glossary.