Key points
- A plain-English glossary of the words you keep seeing in AI-stock coverage, from GPU and HBM to neocloud, 13F, and picks and shovels.
- Grouped by theme: the chips and hardware, the cloud and data centers, power, the filings, and the market terms.
- Bookmark it. Most terms link out to a fuller explainer where we have one.
AI-stock coverage comes with its own vocabulary, and a lot of it is jargon nobody bothered to explain to you. This is a running glossary of the terms we use most, in plain English, grouped by where they fit in the AI trade. Skim it, or search the page for the one word that tripped you up. For the big picture of how these pieces fit together, start with the AI Stock Map.
Chips and hardware
- GPU: graphics processing unit, the chip built for the parallel math that AI training and inference need. Nvidia (NVDA) dominates the market.
- HBM (high-bandwidth memory): fast, stacked memory that sits next to a GPU and feeds it data. Made mainly by Micron, SK Hynix, and Samsung, and a key bottleneck of the AI buildout.
- Foundry: a factory that manufactures chips other companies design. TSMC makes most of the world's leading-edge AI chips.
- Fabless: a chip company that designs chips but owns no factory, like Nvidia or AMD, and relies on a foundry to build them.
- Advanced packaging (CoWoS): the step that stitches a processor and its memory into one package. It is a real bottleneck in making AI chips.
- Training vs inference: training is the heavy compute that builds an AI model; inference is the ongoing compute that runs it for users. Both drive chip demand.
- ASIC: an application-specific chip built for a single job, like Google's TPU, as opposed to a general-purpose GPU.
- Optics and transceivers: the components that move data between chips and servers using light, which matters more as AI clusters grow.
- Supply-chain chokepoint: a single company, country, or material that a whole industry depends on. AI chips have several hidden ones.
Cloud, compute, and data centers
- Hyperscaler: a giant cloud provider that builds enormous data centers: Amazon (AWS), Microsoft (Azure), Google (Cloud), and increasingly Oracle.
- Neocloud: a newer company that rents out GPU computing specifically for AI work, such as CoreWeave, Nebius, and SharonAI, as opposed to the general-purpose hyperscalers.
- Data center: the physical building full of servers and networking gear where AI models are trained and run.
- Capex (capital expenditure): the money companies spend on long-lived assets like data centers and chips. Hyperscaler capex is the number the whole AI trade watches.
- Sovereign AI: a country building its own AI computing capacity so it does not depend on foreign data centers.
Power
- AI's power problem: data centers consume enormous amounts of electricity, so power and the grid are becoming a bottleneck for AI growth. See nuclear's big break.
- SMR (small modular reactor): a smaller, factory-built nuclear reactor pitched as a way to supply steady power to data centers.
Money and filings
- 13F: the quarterly SEC filing that lists a big investment manager's US stock holdings. Full explainer here, and you can browse them on our fund tracker.
- 13D and 13G: SEC filings triggered when an investor crosses 5 percent ownership of a company. A 13D signals an activist or influential stake and must be filed within a few business days; a 13G is the lighter short-form version for passive holders, which large institutions can file up to 45 days after the quarter ends.
- Form 4: the filing a company's own officers and directors must submit, usually within two business days, when they buy or sell their own stock.
- STOCK Act: the law that requires members of Congress to disclose their stock trades. We track those on the Congress page.
- Float: the number of a company's shares that trade freely. A small float can make a stock swing hard in both directions.
- Dilution: when a company issues new shares, shrinking each existing owner's slice. Common in fast-growing or unprofitable companies that raise money often.
Market and trading terms
- Picks and shovels: investing in the suppliers and infrastructure behind a boom rather than the headline product. In AI, that means chips, memory, power, and networking.
- Mag 7 (Magnificent Seven): the seven megacap tech stocks that drive much of the market: Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, and Tesla.
- Physical AI: AI placed inside machines that act in the real world, like humanoid robots and self-driving systems.
- Book-to-bill: the ratio of new orders to products shipped. Above 1 suggests demand is building faster than a company can ship.
- Take-or-pay: a contract where the customer pays for capacity whether or not it actually uses it, common in data center and compute deals.
- Short interest: the share of a stock that traders have bet against by selling borrowed shares. A heavily shorted stock can rip higher in a "short squeeze."
Missing a term you keep seeing? It will get added here over time. None of this is investment advice; it is just the vocabulary, explained.
