Every technology cycle produces two kinds of companies: those whose economics are changed by the technology, and those whose press releases are. The difference is rarely visible in a quarterly headline and almost always visible in the filings. A disciplined process for evaluating AI exposure starts by refusing to accept the label and asking where, specifically, the money moves.
1. Locate the revenue line
Begin with segment reporting in the annual report (Form 10-K for US issuers). Ask whether the company discloses AI-related revenue separately, folds it into a broader segment, or only references it in management commentary. Commentary is not disclosure. If a company cannot or will not quantify the contribution, treat the narrative as an option, not an asset.
2. Separate the picks from the miners
Suppliers of compute, memory, networking and power sell to AI builders regardless of which model wins. Application companies must win customers against every other application built on the same foundation models. These are different risk profiles. Suppliers carry cyclical and concentration risk; application companies carry competitive and pricing risk.
3. Read customer concentration notes
Filings typically disclose customers representing more than ten percent of revenue. In AI infrastructure, a small group of hyperscale buyers can account for a large share of demand. Concentration is not disqualifying, but it means the stock is partly a bet on the capital expenditure plans of companies you do not own.
4. Follow the margin, not the growth
Rapid growth funded by discounting or by subsidised compute is fragile. The more useful signal is whether incremental revenue arrives at an equal or higher gross margin than the existing business. Rising margins suggest pricing power; falling margins suggest the company is paying to participate.
5. Price the downside first
Before modelling the optimistic case, write down what the business is worth if AI revenue merely stabilises. If the answer implies a large drawdown from the current price, position size should reflect that, regardless of conviction.
Commentary is not disclosure. Treat an unquantified AI narrative as an option, not an asset.
None of this predicts which companies will outperform. It is a way to make sure that when you own one, you know which assumptions you are paying for. This article is educational and does not constitute investment advice.
