Guide · Markets
AI Capex Explained: How Data-Center Spending Moves Stocks
AI capex is the money cloud and AI companies spend on data centres, chips, servers and networking, and it becomes the revenue of the suppliers. Here is how to follow the chain, using AMD, Arista and Supermicro's reports of November 3, 2026.
On the evening of Tuesday, November 3, 2026, three suppliers of AI data centres reported within an hour of each other. AMD said its data-center revenue had risen 97% to $8.5 billion in the third quarter and guided total fourth-quarter revenue to about $17 billion, against $16.2 billion expected. Arista Networks guided to $3.9 billion to $4.0 billion against $3.60 billion. Super Micro Computer reported quarterly revenue of $19.95 billion, up 298%, and raised its fiscal-year range to $72 billion to $82 billion. All three rose after hours. A few hours later, as Democrats were projected to win Congress, Tokyo's AI suppliers fell hard on fears of tighter rules for data centres, and the Nikkei was down 3.56% by Wednesday's midday break.
Capex, short for capital expenditure, is money a company spends on long-lived assets such as buildings and equipment instead of on running costs. AI capex is the part spent on data centres for training and running AI models: the buildings and the power connections, the accelerator chips and processors, the servers they sit in, the memory, and the switches and cables that tie thousands of servers together. It is reported in the cash-flow statements of the buyers, mostly the large cloud companies, and their guidance for next year's capex is one of the most watched numbers of every earnings season.
The useful way to read it is as a chain in which one company's spending is another's revenue. At the top sit the buyers: cloud providers, AI labs and the companies that lease capacity to them. Below them are the chip designers such as Nvidia and AMD, the server builders such as Supermicro, the networking suppliers such as Arista, the memory makers, the equipment makers that supply the chip factories, and the power and cooling companies. On November 3 the chain moved together in New York: besides the chipmakers, Eaton rose 7.0%, GE Vernova 6.5% and Vertiv 5.2%. In Tokyo the next morning it fell together, from Advantest and Tokyo Electron to the cable makers Fujikura and Furukawa Electric.
Position in the chain decides how much of each dollar a supplier keeps. Supermicro's gross margin was 12.8% in the quarter, because most of a server's value is the chips it buys from others; Arista's adjusted gross margin was about 64%, because it designs its own hardware and software; AMD's was 56%. Higher-margin suppliers turn growth into profit faster, while low-margin assemblers need very large revenue gains to make the same difference, which is why investors watch Supermicro's margin almost as closely as its sales.
Three features make AI-capex stocks swing more than their earnings suggest. First, they are priced on the rate of growth rather than its level, so a forecast that is merely good can sink a stock that was priced for great. Second, spending is concentrated in a handful of buyers, so one cloud company trimming its budget can move every supplier at once. Third, the spending depends on things outside the companies' control: financing costs, power supply and, as Tuesday night showed, politics, including zoning, energy and planning rules at state level and the possibility of new federal oversight.
To follow the theme, read the buyers before the suppliers. Note each large cloud company's capex guidance for the coming year and whether it was raised or cut, then check how the suppliers' guidance compares; when suppliers guide above consensus while buyers hold their budgets steady, someone is gaining share. Watch the margins as well as the revenue, and keep a list of the policy events that could slow construction. A strong quarter tells you about orders already placed; the buyers' plans and the rules around them tell you about the next ones.