What companies actually say about AI spending on earnings calls

Three kinds of AI mention, and only one of them is news

The first kind is positioning: "we are leveraging AI across the organisation." It appears in the prepared remarks, carries no figure, and is aimed at the share price rather than at anyone who will check. It is close to free to say, which is exactly why it is said so often.

The second is spending: a capex line, a headcount, a named project with a budget. This is costly to say, because the money will be visible in the cash flow statement and management will be asked about the return in two quarters. Companies do not put numbers on ambitions they intend to quietly drop.

The third, and rarest, is return: revenue attributed to an AI product, a margin improvement traced to automation, a cost line that fell for a stated reason. When you find it, read it carefully, because it is the only one of the three that has already been tested against reality.

Where AI is actually being discussed, and it is not where you think

We counted the calls in our current-quarter coverage whose extracted evidence mentions AI in any form. It comes to 1,613 of them, and the sector distribution is the interesting part.

Information Technology leads with 466, which surprises nobody. Financials is second with 341, ahead of Industrials at 227, Consumer Discretionary at 158, Communication Services at 151 and Health Care at 127. Banks and insurers are discussing AI on their earnings calls more than manufacturers are, by a margin of fifty per cent.

That ordering is a reasonable proxy for where the deployment conversation has actually reached management level. Financial services has the two ingredients that make AI a board topic early: large volumes of text and numbers to process, and a cost base made mostly of people. Industrials have neither in the same concentration, which is why their AI language still tends toward the product rather than the operation.

The practical use of the ranking is as a starting scope. If you want to read serious AI deployment language rather than positioning, the density is highest in the two sectors at the top, and the third place is where the story is still forming.

The questions that separate them

• "Which companies quantified AI spending this quarter, and how much?" Forces a number and filters out the positioning.
• "Who attributed revenue to an AI product, in their own words?" The rare third kind, and worth reading in full.
• "Where did AI come up in the Q&A rather than the prepared remarks?" Analyst questions are unscripted, and the answers are more revealing than the script.
• "Which management teams were asked about AI return on investment and did not answer?" A non-answer on the record is itself information.
• "Compare what two competitors said about AI capex this quarter." Peer framing turns a number into a judgement.

Why the Q&A matters more than the script

Prepared remarks are written, reviewed and legally checked. The Q&A is not. When an analyst asks a CFO to break out AI spending and the answer becomes vague, the vagueness is a finding. When the answer is precise, that precision was prepared for, which means the company expects the question and is comfortable with the number.

This is the part of an earnings call that rewards reading and punishes summaries. A headline says "company X talks up AI." The transcript says whether the CEO named a figure, deferred to next quarter, or redirected to a different topic. Only one of those is worth acting on.

Watching the theme move across a sector

A single company spending on AI is a company story. A sector where the spending language shifts from "evaluating" to "deploying" inside two quarters is a sector story, and it usually shows up in the transcripts before it shows up in the numbers.

The way to catch it is to ask the same question across a whole sector rather than company by company, then follow the ones whose wording changed. That is a mechanical task over hundreds of calls, which is precisely what it makes sense to hand to a machine that has read all of them.

How earnings.chat helps

Ask about AI spending across a sector, a country or a quarter, and the answer comes back with the passages behind it: who said it, on which call, in what words. Every figure is quoted rather than summarised, so you can tell a capital commitment from a press release before you decide whether to care.

The archive covers more than 253,000 earnings calls back to 2020, which means the same question also works backwards. "How did this company talk about AI two years ago?" is the comparison that turns a claim into a track record.