🔌

CAT

💡
Thesis. Power generation sales to users up 72% and a record $72B backlog growing 92% — with orders extending to 2030 and management stating demand is not the constraint, capacity is. Record margins alongside record volume is the margin test passing.
Reviewed 19 Sep 2026, after Q2 2026 results. Next review after Q3 2026 earnings on 28 Oct 2026.

What it does

Caterpillar makes reciprocating engines, gensets and Solar Turbines for on-site power — alongside the construction and mining equipment it is better known for.
A genset is a large piston engine driving a generator: think an enormous truck engine. The distinction that matters is how it is used. Standby power runs only when the grid fails. Prime power runs continuously as the main source. Datacentres are shifting from standby to prime, and prime is a far larger market per site — the engine runs all the time instead of a few hours a year, which changes both the unit economics and the service revenue behind it.
This is simultaneously an industrial-cycle stock and an AI power name. Reading only the total revenue will tell you about the first and hide the second.

Bull case

  • Q2 2026 revenue $20.5B, +24% — the first $20B quarter in company history
  • Adjusted EPS $8.17, +73%, against ~$6.20 expected. Stock +9% premarket
  • Adjusted operating margin 21.9%, up 430bps, with record MP&E free cash flow of $5.1B. Record margins on record volume
  • Record backlog $72B, +$9B sequentially and ~92% YoY. 59% delivers within 12 months, and some customers are placing orders as far out as 2030
  • Power generation sales to users +72%. Power & Energy sales to users +33%; segment revenue $8.2B of the $20.5B total
  • Construction Industries $8.3B, +35%, with North American demand +50%. Oil and gas backlog nearly doubled
  • FY26 outlook raised to mid-to-high-teens growth from low double digits; free cash flow expected in the top half of the $6–15B range
  • Tariff costs came in at ~$400M against a $700M estimate, helped by $392M of IEEPA recoveries
  • Demand is explicitly not the limiting factor — management names manufacturing capacity, lead times and site-level power availability. Capacity additions are additive to the prior 65 GW target. CEO Joe Creed on AI power demand: "no one is slowing down"

Bear case

  • Construction and mining remain the majority of the business — genuine industrial cyclicality that dilutes the AI signal, and a downturn there would swamp the power story
  • Tariff exposure ~$2.2B for the full year, against which Q2's ~$400M was flattered by one-off recoveries
  • Megacap size dampens the signal. The stock moves 1% on days when smaller power names move 8%, so it will drown a category average rather than express it
  • Backlog is partly a function of extended lead times, not purely demand growth — a $72B figure built on long queues behaves differently from one built on new orders
  • If gas turbine supply loosens, some demand that went to reciprocating engines could shift back to turbines at

Major customers

  • Datacentre operators and hyperscalers — gensets and prime power. Power generation sales to users +72%
  • Neoclouds and colocation builders —
    ☁️
    CRWV    ,
    ☁️
    IREN     and similar, fitting out capacity
  • Construction contractors and equipment dealers — Construction Industries, +35%
  • Mining companies — Resource Industries
  • Oil and gas operators — backlog nearly doubled YoY
  • Sold through an unmatched global dealer and service network, which is the real moat
🔗
Read-through. ⚠️ An earlier note in this workspace described Caterpillar's AI exposure as "a small slice — an industrial cycle stock in an AI basket." That was wrong. Power generation sales to users rose 72%, Power & Energy is $8.2B of $20.5B of revenue, and management named datacentre demand as a primary driver. It competes with where customers choose turbines over reciprocating engines and with
🔌
BE where they choose fuel cells for speed of deployment — so the three should be read as one contest over how a site gets powered before the grid can reach it. Caterpillar's statement that site-level power availability is a binding constraint is independent confirmation of what the grid contractors at
⚡
PWR and
⚡
STRL report from the other side.

What would change the view

  1. Power generation sales to users — currently +72%. The AI metric, tracked in isolation from the total
  2. Backlog trajectory from $72B, and whether growth is orders or lead times
  3. Whether construction and mining turn down, which would expose the cyclicality the AI story is currently masking
  4. Tariff costs against the ~$2.2B full-year estimate
  5. Capacity additions beyond the 65 GW target
  6. Whether orders keep extending toward 2030

Update log

19 Sep 2026 — Converted to the standard template. Current through Q2; no new disclosures since. Kept the correction of the earlier "small slice of AI exposure" note, since it is the kind of error a reader should see acknowledged rather than quietly removed.
Q2 2026 — Revenue $20.5B (+24%), the first $20B quarter. Adjusted EPS $8.17 (+73%) against ~$6.20 expected; adjusted operating margin 21.9% (+430bps); MP&E free cash flow $5.1B. Record backlog $72B (+92% YoY, +$9B QoQ). Power generation sales to users +72%; Power & Energy segment $8.2B. FY26 outlook raised to mid-to-high-teens growth.
Research and education only — not investment advice.