The AI Capex Supercycle, as Told by 1,400 Earnings Calls

I spend my days inside a database of 253,129 earnings call transcripts — 12,799 companies, 11.99 million speaker segments, everything from January 2020 to today, with roughly 10,000 new calls landing every quarter. When a macro story is real, it shows up in this corpus before it shows up almost anywhere else, because earnings calls are where the people spending the money have to explain themselves, on the record, to the people who gave it to them.

The macro story of 2026 is AI capex. Not as a think-piece abstraction — as a line item. In the second quarter of 2026, 1,447 companies held an earnings call that matched a full-text search for "AI capex". Two years earlier, in Q2 2024, that number was 781.

And the companies doing the most talking are not the ones you'd guess.

Here's what the corpus says, as of August 7, 2026.

The curve, quarter by quarter

I ran the same search — the exact phrase "AI capex" — across every quarter of calls since the start of 2024 and counted distinct companies per quarter. This is the result:

A quick word on methodology, once, so we don't have to keep circling back: each count is the number of companies whose earnings call in that quarter matched a full-text search. Keyword matching is blunt. An analyst asking a supplier about a customer's AI capex counts exactly the same as a CFO announcing their own, so treat every figure here as an upper bound that includes incidental co-occurrences. The level is fuzzy; the trend is the signal.

And the trend is not subtle. From 765 companies in Q1 2024 to 1,757 in Q1 2026 — the count has more than doubled in two years, and the growth is accelerating, not fading. The year-over-year change ran +43% in early 2025 and hit +85% in Q2 2026, the steepest reading in the series.

Two honest caveats about the shape of that curve. The Q2 dips are partly mechanical: second quarters simply carry fewer calls in this corpus — 10,053 calls in Q2 2026 versus 10,234 in Q1 — and 2025 shows the same seasonal sag. And the one flat reading, Q2 2025, sits exactly in that thin quarter. Every other reading climbs. Which is why the YoY column is the honest comparison, and why the latest number in it is the one I keep staring at.

Industrials out-talk tech now

The quarterly counts tell you the theme is spreading. The sector split tells you where — and this is the datapoint I think most people have wrong.

Here's the Q2 2026 breakdown of companies matching "AI capex", by sector:

• Industrials: 461 companies
• Information Technology: 378
• Consumer Discretionary: 281
• Communication Services: 183
• Health Care: 156
• Financials: 155

Read that again. Industrials lead. Roughly one in three companies talking about AI capex last quarter was an industrial — the companies building, powering, cooling and wiring data centers now discuss AI capex more than the technology sector does.

The bottom of the list is worth a glance too. Health Care at 156 and Financials at 155 are not building data centers — they're buying the output. When hospital operators and banks field AI capex questions on their earnings calls, the theme has crossed over from the producers of compute to its consumers, and that's the part of the diffusion that keyword counts capture better than any sell-side model.

This is what a supercycle looks like from the inside. The first wave of AI talk on earnings calls was software companies describing their models. The second wave was hyperscalers defending their budgets. The third wave — the one we're in — is gen-set manufacturers, turbine makers, electrical-equipment suppliers, HVAC companies and engineering firms fielding analyst questions about data centers. The supply chain starts talking about a cycle before the cycle fully shows up in its own backlog, because analysts ask about demand the moment they see it forming somewhere else. I saw the same pattern when I tracked grid constraints in why power is the new bottleneck in earnings calls — the electricity problem surfaced in industrial and utility transcripts quarters before it became a consensus headline.

Caterpillar's call this week is the cleanest specimen I've seen. On the August 4 call, management reported: "In Power and Energy, sales to users grew a robust 33%. Power generation grew 72%, driven by very strong demand for large gen sets and turbines used in data center applications."

Sit with that. A company most people file under bulldozers and mining trucks just reported 72% growth in power generation — driven by data centers. That's not Caterpillar making an AI pivot. That's AI capex arriving at Caterpillar's loading dock, whether they asked for it or not.

Two sides of one trade

Every capex dollar is someone else's revenue dollar. That's the accounting identity underneath this whole story, and it's why the same theme reads so differently depending on which side of the invoice a company sits on.

On the spending side, the numbers have left the realm of normal corporate budgeting. Per sell-side estimates circulating this summer, the top hyperscalers are expected to nearly double their capex in 2026, with combined spend heading toward $1 trillion in 2027. A trillion dollars of spend is not a product decision anymore; it's an industrial policy executed by four or five procurement departments.

On the collecting side, this week gave me a near-perfect specimen too. Western Digital, on its August 5 call, reported: "Operating cash flow was $1.4 billion and CapEx was $108 million. This resulted in free cash flow generation of $1.3 billion for the quarter and a strong free cash flow margin of 34%." Earnings per share came in at $3.56, up 109% year over year, per the call.

Look at the ratio. $108 million of capex against $1.4 billion of operating cash flow. Western Digital is a direct beneficiary of AI demand — the training clusters and inference farms all need storage — but it isn't spending its way into the boom. Storage pricing is doing the work; the capex is happening on someone else's balance sheet. That is what the good side of a capex supercycle looks like: your customers build at a historic pace, and you convert their urgency into a 34% free cash flow margin.

The same week, the same theme, two transcripts: Caterpillar selling turbines into the buildout, Western Digital harvesting cash flow from it — and the hyperscalers writing the checks that fund both. This dispersion is the real Q2 story, and it's invisible at the index level. Per public reporting, more than 80% of S&P 500 companies beat Q2 estimates — the aggregate looks placid. The action is under the surface, between the companies spending the money and the companies collecting it. If you only track the beat rate, you miss the entire trade.

The debt angle is the one to watch next

One more piece of context from public reporting this year, and I think it's the most underrated: AI-related issuance has become a mid-to-high-teens share of US investment-grade and high-yield issuance in 2026. The buildout is increasingly debt-financed.

That changes what earnings calls are for. When capex runs through operating cash flow, the interesting disclosures are budget numbers. When capex moves to the bond market, the interesting disclosures move to the language around the numbers — how management talks about financing, leases, commitments, off-balance-sheet structures and take-or-pay arrangements. That vocabulary is the early-warning channel. Ratings and covenants react in quarters; call language shifts in weeks, because executives start pre-empting the questions they know are coming.

I'm not going to hang numbers on this yet — I haven't run the full study, and inventing figures is exactly what this corpus exists to prevent. But the search strategy is straightforward: take the phrases "financing", "lease" and "commitments", scope them to the names in the data-center supply chain, and watch the frequency and the surrounding context quarter over quarter. Language-level signals like this are bread and butter for systematic readers — I covered the general toolkit in five text signals quants pull from earnings calls. The financing vocabulary around AI capex is, in my view, the next one worth adding to that list.

Track it yourself

Everything in this post is one API call away. Here's the exact query for "AI capex" mentions from Q2 2026 onward:

curl 'https://earningscalls.dev/api/v1/search?q="AI capex"&type=transcripts&date_from=2026-04-01' -H "X-API-Key: $KEY"

The q parameter supports "exact phrase" matching, AND/OR operators and -negation, so you can tighten the query as much as you like — "AI capex" AND "data center" -crypto is a perfectly legal search. Bucket the results by quarter and sector and you've rebuilt my table; the API docs cover the full parameter set, and if you'd rather have Claude run the queries conversationally, the MCP server exposes the same search as a tool. For a complete worked example — quarterly counts, plotting, the whole loop — I wrote up the pattern in build an earnings call theme tracker in Python; swap in "AI capex" as the theme and you have this post as a living dashboard.

My honest read on where this stands: the mention curve is still accelerating, the theme has jumped the fence from tech into the industrial economy, and the financing of the buildout is migrating to the bond market — which means the next chapter of this story will be written in the careful, lawyered language of CFOs discussing commitments. That language lands in this database within hours of every call. I'd rather read it there than wait for it to become a headline.

Search all 253,129 transcripts yourself — full-text API, MCP server and web search at earningscalls.dev.