Caterpillar Inc. CAT

Key data

The business

Caterpillar sells the machines that dig, move, and haul: excavators, bulldozers, wheel loaders, and the giant mining trucks that carry ore out of open pits. It also builds the reciprocating engines and gas turbines that generate power, and it runs a captive bank, Financial Products, that finances the dealers and buyers. As of this year the company reports in two lines, Machinery, Power & Energy and Financial Products. The first is 95% of revenue and did $19.6 billion last quarter. The moat is the dealer network, a few hundred independent dealers who sell, service, and stock parts, so a mine in Chile or a contractor in Texas buys parts and service from the same yellow brand for the machine's whole life. That aftermarket annuity carries CAT through the down years when nobody orders a new dozer. The new twist is power: data centers being built for AI need backup and prime power, and CAT's engines and turbines are suddenly a growth business inside a company most people still picture as a construction cyclical.

The numbers

The sequence shows a company that stumbled in 2025 and came roaring back in 2026.

The line to read is operating income against revenue. Through 2025 sales grew while profit did not, tariffs and cost inflation eating the margin, and Q4 2025 margin fell to 13.9%. Then Q2 2026 broke the pattern: revenue up 24% year over year, operating profit up 50%, the biggest quarter Caterpillar has ever printed. Adjusted earnings of $8.17 ran well past the $6.22 the Street looked for.

Step back to the fiscal years and 2025 is the dent, EPS falling to $18.81 from $22.05 as margins gave back what 2023 and 2024 had won. The first half of 2026 has already recovered $13.24 of it. The question is which number is real. Every incoming dollar of sales in Q2 turned into 36 cents of operating profit, a rate you only get at the strong part of a cycle. The market is capitalizing it as permanent: EV/EBITDA of 26.6 sits above the 11.5 to 17.4 the stock held for most of twelve years, and above the roughly 18 peers fetch. Consensus has EPS at $32.41 in 2027 and $38.72 in 2028, which requires the margin to hold and data-center power demand to keep building. What this memo doubts the market is pricing carefully is that the incremental margin survives a construction slowdown. The print that settles it is the order rate, the next two quarters of book-to-bill against a decelerating machinery cycle.

That table is the whole argument in three rows: 2025 was decremental, every extra sale losing money at the margin, and 2026 flipped hard positive. A day ago the read here was that the multiple, not the machines, was the risk. Q2 confirmed it. The machines ran hot while the multiple stayed at the top of its history.

Management

Management's own money tells a one-sided story. Over the last twelve months insiders bought once, for $219 thousand, and sold thirty-five times for $114 million, clustered in May 2026. The CFO Andrew Bonfield sold $14.4 million, group president Anthony Fassino $14.2 million, Bob De Lange $13.3 million, all within days. The feed does not carry 10b5-1 status, so plan status is not disclosed on these, read them as a cluster rather than a signal, but a cluster that leaned into the run. Capital allocation reads better: Caterpillar spent $5.0 billion on buybacks in Q1 2026 near the year's lows around $416, then watched the stock reach $1,073, good timing on its own shares for once.

How it fails or surprises you

The cycle rolls over. Machinery is cyclical and you are paying peak earnings for it. If construction and mining capex decelerates, the same leverage that gave you 36-cent incremental margins runs backward, exactly as in 2025 when Q3 operating profit fell on higher sales. Watch dealer order rates and book-to-bill. A reading below one is the downturn already booked.

The multiple, not the machines. The fact this read explains least is the multiple. Booming earnings explain the price, not why EV/EBITDA sits 50% above its twelve-year norm. The market is pricing data-center power as a secular annuity. If that demand proves cyclical, earnings and multiple compress together, the double unwind that takes an $814 stock much lower fast.

Data-center power is real and durable (right tail). If turbine and genset demand from AI data centers is a multi-year buildout and not a spike, Power & Energy becomes a structural growth engine that earns its premium. The 10-Q already calls turbine backlog healthy with solid order activity. Should 2028 consensus of $38.72 prove low, the multiple holds and the peak-cycle worry was misplaced.

Closing thoughts

The order book and Power & Energy backlog over the next few quarters decide whether the market is right to price CAT as a compounder or wrong to pay peak-multiple for peak-cyclical earnings. The stock is not expensive because the market is wrong about the earnings, which are real and just set a record. It is expensive because the market has decided the data-center power story converts Caterpillar from a cyclical into a compounder, and that conversion is not yet proven. Order rates and Power & Energy backlog over the next few quarters resolve it. The fatter tail is the downside, not because the business is weak but because you are paying peak earnings times a peak multiple, and both move the same way when the cycle turns. What is at risk if orders roll is a compression of both at once. What the upside is worth is a durable growth multiple on a business the market still half-prices as cyclical.

The bet is still that the world keeps buying the yellow machines and now the generators that power AI, and that last quarter's margins are the floor rather than the ceiling. What breaks it is the order book. If book-to-bill slips below one while the stock still carries 26 times EBITDA, you own peak earnings at a peak price. Watch the incremental margin and the order rate together. The day they diverge is the day this stops working.