Stock-market and AI infrastructure glossary

AI hardware and memory

Accelerator

A processor designed to perform particular workloads more efficiently than a general-purpose CPU. GPUs and purpose-built AI chips are accelerators. In AI systems they perform large volumes of parallel math for training and inference. Performance depends on the surrounding memory, networking, software, and power—not just the processor’s peak specification.

Artificial intelligence training

The compute-intensive process of adjusting a model’s parameters using large datasets. Training can require many accelerators operating together for extended periods, which increases demand for high-speed memory, networking, storage, electricity, and cooling. Bigger is not automatically better, but it is reliably more expensive.

Inference

Running a trained model to produce an answer, classification, image, recommendation, or other output. A single inference may be lighter than training, but serving millions of users continuously can create enormous aggregate infrastructure demand. Cost per query and utilization matter to the economics.

GPU (graphics processing unit)

A processor originally associated with graphics and now widely used for parallel computing. Its architecture is well suited to much of the matrix math used in AI. A GPU investment thesis should consider hardware demand, software ecosystem, product transitions, competition, customer concentration, and the returns customers earn from deployment.

CPU (central processing unit)

The general-purpose processor that runs operating systems and coordinates a wide range of computing tasks. AI servers often pair CPUs with accelerators. CPUs manage orchestration, data preparation, and workloads that do not belong on a GPU. They remain infrastructure even when keynotes give them fewer leather jackets.

HBM (high-bandwidth memory)

A form of high-performance DRAM packaged close to an accelerator to move large amounts of data quickly. HBM is important to AI computing because processors can sit idle if memory cannot supply data fast enough. Manufacturing yields, advanced packaging, customer qualification, capacity, and price are central business variables.

DRAM (dynamic random-access memory)

Volatile memory used for active data and working operations in servers, PCs, and devices. “Volatile” means data disappears without power, not that the chip has seen its brokerage statement. DRAM is cyclical: supply discipline, demand growth, inventory, and pricing can cause large swings in producer profitability.

NAND flash

Non-volatile memory used in solid-state drives and many storage devices. It retains data without power. AI systems can increase storage needs through datasets, checkpoints, logs, embeddings, and generated content. NAND has its own supply-and-price cycle and should not be treated as interchangeable with HBM or DRAM.

Enterprise SSD

A solid-state drive designed for servers and data centers, with requirements for performance, endurance, reliability, and management. Enterprise SSD demand can connect AI growth to NAND suppliers, but outcomes depend on product mix, customer qualification, capacity, pricing, and competitive behavior.

Advanced packaging

Manufacturing techniques that connect multiple chips or memory stacks into a high-performance package. Packaging can affect bandwidth, power efficiency, yields, and available supply. A shortage of packaging capacity can constrain finished accelerators even when individual components exist—a supply chain group project with one missing stapler.

Data-center utilization

The portion of installed computing capacity that is actively used. High utilization can support revenue and investment returns; low utilization may signal excess capacity, delayed customer demand, or inefficient deployment. Announced capacity is not the same as installed capacity, and installed capacity is not the same as profitable usage.

Company finance and earnings

Capital expenditure (capex)

Money spent on long-lived assets such as factories, servers, data centers, networking, and equipment. Rising AI capex can benefit infrastructure suppliers, while also increasing depreciation and cash requirements for buyers. The important questions are where the money goes, when capacity becomes productive, and whether it earns an adequate return.

Revenue

The money a company records from selling goods or services before subtracting operating costs. Revenue growth can come from higher unit volume, higher prices, acquisitions, or currency effects. It is the top line, not the ending of the story.

Gross margin

Revenue minus the direct cost of producing goods or services, expressed as a percentage of revenue. For memory suppliers, higher selling prices and richer product mix can lift gross margin, while weak pricing, underused factories, or poor yields can reduce it. Gross margin helps reveal whether growth has economic quality.

Operating leverage

The tendency for profit to change faster than revenue because part of a company’s cost base is fixed. Semiconductor manufacturers can have high operating leverage: when factories fill and prices rise, incremental revenue may produce large profit gains; when demand falls, fixed costs remain helpfully committed to the bit.

Free cash flow

Cash generated by operations after capital expenditures, subject to the exact definition used. It shows what remains after funding much of the infrastructure needed to run and grow the business. For capital-intensive AI companies, accounting earnings and free cash flow can tell noticeably different stories.

Guidance

Management’s forecast or range for future revenue, profit, margins, spending, or other metrics. Investors compare guidance with prior guidance and market expectations. Guidance is informed by company knowledge but remains a forecast, delivered by people who also prefer the stock to remain emotionally stable.

Earnings revision

A change to analysts’ forecasts for a company’s future revenue or profit. A pattern of upward revisions can show that results and guidance are exceeding prior assumptions. Downward revisions can indicate weakening demand, pricing, margins, or execution. Revisions matter because stocks respond to the gap between reality and expectations.

Inventory

Goods and materials held by a supplier, distributor, or customer. Rising inventory can reflect preparation for growth or products that are not selling; context decides which. In semiconductor cycles, inventory across the channel helps distinguish genuine end demand from temporary restocking.

Product mix

The proportion of revenue coming from different products. A shift toward higher-value HBM or enterprise products can lift average selling price and margin even if total unit volume grows modestly. Mix can also reverse, because spreadsheets enjoy plot development.

Markets, valuation, and risk

Valuation

The price investors assign to a company relative to earnings, sales, cash flow, assets, or a modeled future. A strong business can be a poor investment if the price embeds outcomes it cannot exceed. A low multiple can signal opportunity or a problem visible from space. Valuation needs growth, risk, quality, and cycle context.

Multiple

A ratio such as price-to-earnings or enterprise-value-to-sales used to compare market value with a financial metric. “Multiple expansion” means investors pay more for each unit of the metric; contraction means they pay less. Neither movement requires the underlying company to change on the same day.

Cyclical company

A business whose demand, pricing, and profits fluctuate meaningfully with economic or industry cycles. Memory manufacturers are classic examples because supply additions and demand changes can create shortages and gluts. Using peak earnings as if they are permanent is a popular method of discovering cyclicality.

Pricing power

The ability to raise or maintain prices without losing unacceptable demand. Scarcity, product differentiation, switching costs, and limited competition can create pricing power. New capacity, substitutes, or weaker demand can remove it with limited notice.

Supply tightness

A market condition in which available qualified supply struggles to meet demand. Tightness may improve price and supplier margins, but the duration matters. Customers can redesign products, suppliers can add capacity, and demand can slow. A shortage is a condition, not a personality.

Hyperscaler

A very large cloud or internet company operating data centers at enormous scale. Hyperscalers are major buyers of AI accelerators, servers, networking, storage, and power infrastructure. Their capital-spending plans can influence demand across the entire supply chain.

Market capitalization

The share price multiplied by shares outstanding. It measures the equity market value of a public company, not its revenue, enterprise value, economic importance, or cash in the bank. Share-price comparisons without share counts are therefore mostly decorative.

Volatility

The magnitude of price fluctuations over time, often measured statistically. Volatility is not identical to permanent loss, though it can produce one if leverage, forced selling, or panic enters the chat. Historical volatility also does not fully capture business, liquidity, or event risk.

Position sizing

Choosing how much capital to allocate to an investment. It connects conviction with uncertainty, downside tolerance, liquidity, and portfolio concentration. A thesis can be directionally correct and still cause damage if the position is too large or the timing requires money the investor cannot leave at risk.

Thesis invalidator

An observable condition that would weaken or disprove the reason for holding an investment. Good invalidators are defined before adverse news and focus on business evidence: falling demand, oversupply, lost competitiveness, worsening unit economics, or collapsing estimates. “The stock has hurt my feelings” is data, but not sufficient analysis.

How to use these terms

Definitions are the starting point. When reading an earnings release or market thesis, identify the metric’s period, source, units, comparison base, and relationship to expectations. Ask whether a claim concerns demand, shipments, revenue, profit, cash flow, or merely an announcement. Those words can describe different points in the same pipeline.

Then look for the counterweight. Strong capex may create supplier revenue and buyer depreciation. Tight supply may raise margins and encourage new capacity. High utilization may validate demand and accelerate competition. Every clean narrative is usually standing on several messy variables.