Caterpillar

Caterpillar makes the large yellow machines you see digging foundations, building roads and moving earth in mines: excavators, bulldozers, loaders, haul trucks the size of a house. It also makes the engines and turbines that power them, and, increasingly, power other things: oil and gas compressors, ships, locomotives, and the generators that keep hospitals and data centres running when the grid fails. A ten-year-old could follow it. Someone needs to move a mountain or keep the lights on, and Caterpillar sells them the machine, finances it, and then sells parts and service for twenty years.

It is the largest company of its kind in the world. In the twelve months to September it had $77.8bn of sales and revenues and earned $11.8bn. It sells almost everything through about 150 independent dealers in 190 countries. Those dealers are the most valuable thing it owns, and we will come back to them.

The essential fact about Caterpillar is that it is cyclical. When builders, miners and oil companies are confident, they buy machines; when they are not, they stop, and they run the old ones for longer. In 2009 Caterpillar's revenue fell 37% and its earnings per share 75%. From 2012 to 2016 revenue fell 42% and the company briefly lost money. Any honest valuation has to start there.

Ten years ago Caterpillar was a cyclical company at the bottom of a mining bust. Since then it has compounded at more than 22% a year, through better margins, a large buyback and, this year, a burst of enthusiasm about artificial intelligence. Part of that record was earned and will last. Part of it was the market paying a price for this year's earnings that it has never paid before. Sorting out which is which is the job of this report.

The pattern in the table is worth stating plainly. Caterpillar's worst moments have come straight after its most confident ones: the mining purchase in 2011 was made at the top of a commodity boom that ended within two years. In each case the business survived and paid its dividend, which is a great deal more than most cyclical companies can say. But in each case an investor who paid the boom-time price for boom-time earnings waited years to get the money back. After 2008 the stock regained its high in late 2010. After 2012 it took until August 2017.

This year has a different shape. The share price fell by half before anything bad happened to the business. That is unusual, and it is the central question of this report: was the fall the market correcting an excessive price, or the market seeing a turn in the cycle earlier than the company?

The money is made in three places. The first is the sale of the machine, which is lumpy, competitive and cyclical. The second is parts and service. A mining truck runs for decades and needs tyres, engines rebuilt, hydraulics replaced. Caterpillar does not disclose the margin on this business, but it is higher and steadier than on new machines, and it was about $24bn of revenue in 2025. The third is financing: Cat Financial lends to customers, at a modest profit and with modest losses, and in doing so helps sell the machines.

The company runs to a simple rule it set itself in 2017: an adjusted operating margin of 15–25% that should rise as sales rise across a band of $60–100bn. At this year's sales, management has indicated it is near the bottom of that range once tariff refunds are excluded. That is an important point in Caterpillar's favour. Margins of 20% would normally look like a peak, but by the company's own yardstick they are not, mainly because tariffs are costing about 3% of sales.

Return on capital tells you the business is good. Over ten years the return on invested capital has averaged about 14%, and on the machinery business alone it is far higher, because the consolidated figure includes Cat Financial's loan book. Free cash flow has been about 125% of net income over ten years. Like Deere, whose X-Ray we published in August, Caterpillar earns much more on its capital than a typical heavy manufacturer. Unlike Deere's farmers, its customers are spread across construction, mining, energy and now power, which smooths the cycle a little.

Caterpillar reported its third quarter on 29 October. On most measures it was the best quarter in its history. The shares fell 9.25% that day, the largest one-day fall in a year, and another 3.8% the next.

So why did it fall? Four reasons were given at the time, and all four are fair.

First, the guide did not rise. Caterpillar kept its forecast of sales growth in the "mid-to-high teens" for 2026, when the market had hoped for about 18%, and said the fourth quarter would show only "moderate" growth. Analysts noted that this implied revenue barely up from the third quarter.

Second, the beat was lower in quality than it looked. About 21 cents of earnings per share came from refunds of tariffs that the Supreme Court struck down in February. Without them, adjusted EPS was slightly below forecasts, not above them.

Third, the dealers are destocking. Caterpillar now expects dealers to cut their inventory of construction machines by about $1.6bn in the fourth quarter, against $600m a year earlier. In August it had said "over $1.0bn". The company says this is seasonal and that dealer inventory will still end the year higher than last, ahead of strong sales in 2027. That may be right. It is also exactly the kind of number that moves first at the top of a cycle.

Fourth, the orders are being booked further out. The backlog rose by $6bn in the quarter, mostly in Power & Energy, to a record $78.1bn. But 42% of it now ships after twelve months, against 38% at the end of 2025. Large engines are sold into 2029 and turbines into 2030. A long order book is good news. It is also less certain news: an order for 2029 is a promise from a customer whose own plans, for a data centre in Texas, say, may change.

There will be no outlook for 2027 until late January. Until then, the market has to guess, and in a falling market it guesses low.

A bulldozer is not a mystery. Komatsu, Volvo, Hitachi, Deere and a growing list of Chinese makers build good ones, often cheaper. So where is Caterpillar's advantage? Not mainly in the machine. It is in three things around the machine.

The dealer network is the moat Buffett would recognise. It cannot be bought and cannot be built quickly: a competitor would need decades and tens of billions of dollars to put a parts warehouse and a trained mechanic within reach of every mine and building site Caterpillar serves. A mining company with a hundred-million-dollar fleet cares less about the sticker price of a truck than about how quickly it is back in service when it breaks. On 31 October Caterpillar closed the purchase of one of its own dealers, John Fabick Tractor, for about $1.24bn plus holdbacks, to support an ownership transition. It is a rare move, and a reminder of how much the company values that network.

The weakness is that a moat around service does not stop a price war in new machines. In 2025 Caterpillar gave up about $1.1bn of price, much of it in construction: it discounted to hold its share. In 2026 price has turned positive again. That is what a narrow-to-wide moat looks like in a cyclical business: strong enough to survive the bad years, not strong enough to dictate the price in them.

On the trend, we would say stable, with the installed base widening. The growing share of revenue from services and long-lived power equipment makes Caterpillar steadier than it was in 2009 or 2015. It does not make it a non-cyclical business.

Here is the year in one picture. At the end of 2025 Caterpillar sold for about 30 times its trailing earnings. By 30 June, after the first half's rally, it was 46 times. That is a price you pay for a business whose earnings will grow fast for a long time, and in 2026 the market decided Caterpillar was one, because its engines and turbines keep data centres running. Since then the order book has risen by a further $6bn and trailing earnings per share by another 10%. The multiple has fallen to 22 times. Nothing has gone wrong at the company. The market has simply stopped paying an AI price for it.

The most striking fact in this story is how small the data-centre business still is. Power Generation, which includes the gensets and turbines going into data centres but also hospitals, factories and remote sites, was about 15–16% of sales. Construction and mining are well over half. Yet the stock rose and fell as if Caterpillar were a data-centre company. On 5 November it fell 6.4% on the same day Vertiv fell 14%, with no news of its own.

And the data-centre business itself is not slowing, at least not yet. Dealer-reported sales of power-generation equipment to end users grew 75% in the third quarter, the fastest of the year. GE Vernova, Cummins, Wärtsilä and Bloom Energy all reported strong data-centre demand in October, and Rolls-Royce raised its power-generation growth target in July. What changed was not demand today but confidence about 2028: whether states will keep permitting large data centres, whether AI companies keep spending at this pace, and whether the power ends up coming from turbines and engines or from the grid.

Our reading. The fall is mostly the removal of a premium that should never have been paid. At 46 times earnings, Caterpillar was priced as if the data-centre boom were permanent and the rest of the business irrelevant. At 22 times, it is priced like a very good industrial company at a record year. That is a much more reasonable price. It is not yet a cheap one, because the thing that matters for a cyclical company is not this year's earnings but a normal year's, and on a normal year 22 times becomes about 24.

In machines, Caterpillar is the leader and Komatsu the long-standing number two, with Chinese makers growing fast at home and in emerging markets. We could not find a market-share figure for Caterpillar from a source we trust, so we will not quote one. What the numbers do show is that Caterpillar's sales in Asia/Pacific are growing far more slowly than elsewhere, 2% in construction this quarter, and that its construction business had to cut prices in 2025 to hold its position.

In power, Caterpillar is one of several strong suppliers in a market where demand currently exceeds what anyone can build. That is a comfortable position while it lasts. When capacity catches up, every one of these companies will be competing for orders with new factories to fill, and Caterpillar is adding capacity too. Investors who have watched mining-equipment cycles will recognise the pattern.

Management. Joe Creed is a Caterpillar lifer who built the power business that the market fell in love with, and has run the company well: margins up, backlog at a record, the capacity expansion backed by orders before the money is spent. Management has also been careful in what it says: the 10-Q speaks of "moderate" growth and a larger destock, not of a permanent boom. We have no integrity concerns.

The buyback. Caterpillar's policy is to return substantially all of its machinery free cash flow to shareholders, and it does. In 2025 it returned about 84%. In the first nine months of 2026 it spent $7.7bn in cash on buybacks and $2.1bn on dividends, a little more than the $9.5bn of free cash flow. The problem is the price. A $4.5bn accelerated buyback in the first quarter was struck when the shares were around $650–750; the third-quarter purchases averaged $832.64, with July at $889.39. Today's price is $558.63. Unlike Fair Isaac, whose buyback we criticised this week, Caterpillar did not borrow to buy: the machinery balance sheet is almost debt-free on a net basis. But a buyback done by formula, at whatever the price, is a way of transferring value to the shareholders who sell. Buffett's rule is that a company should buy its shares only when they are below intrinsic value, conservatively estimated. On our figures, most of 2026's buying was well above it.

Capital spending. Caterpillar is investing about $3.3bn this year, against depreciation of about $2.3bn, to double large-engine capacity and more than double gas-turbine capacity by 2030. Management says the expansion is backed by firm orders, with cash payback by the end of the decade. It is the right decision if the demand lasts. It is also the decision every equipment maker makes near the top of a boom, and the reason booms end in surplus capacity.

Shares are diluted weighted averages (the last is the count at 30 September). EPS is GAAP; the 2016 loss and swings in some years include pension mark-to-market charges and restructuring.

The table shows two things at once. The first is the cycle: operating margin of 3% in 2016 and 20% in 2024, earnings per share from a loss to $22. The second is the improvement: each low point is higher than the last, and the 2025 dip, when tariffs arrived and construction prices fell, still left the company earning about $19 a share. The share count fell by a fifth, which lifts every per-share figure. This is a better business than it was ten years ago. It is still the same kind of business.

This is the comparison that makes us cautious. In both earlier falls of this size, the share price fell because revenue and earnings did, by 37–42% and by 75% or more. This year the price fell by the same amount while revenue and earnings rose. One reading is that the market overpaid in June and has now corrected, and that is mostly what we believe. The other reading is that the market, as it sometimes does, has seen the next downturn before the company has. We cannot rule that out. A $1.6bn destock, orders booked further out, and a political turn against data centres are the kind of early signs that come before a cyclical slowdown, and also the kind that come to nothing.

The dividend is one of the safest in American industry: thirty-two consecutive annual increases, quarterly payments since 1933, never cut through two severe downturns, and a third of free cash flow. It is not, however, a reason to own the stock at this price. At 1.2% it is a token next to the buyback, and the buyback is the part that shrinks when the cycle turns.

Why a normal year? Buffett's advice on cyclical companies, and Peter Lynch's, is the same: they look cheapest when earnings are at a peak and most expensive at the bottom, because a low multiple of peak earnings is not a low price. So we value Caterpillar on what it earns across a cycle. Its sales have been $65–68bn in 2023–25 and should be about $78bn in 2026, helped by dealer restocking and the data-centre boom. We take about $72bn as a normal level for the next few years, an 18% operating margin (mid-range by the company's own targets at that size), tax and other items, and about 450m shares after continued buybacks. That gives about $23 a share. It is a cautious number, below this year's $27 but well above the $19 of 2025, which carried a tariff hit and a price war.

The model. We then let normal earnings grow, put a multiple on 2031 earnings, discount five years at 10% and add the dividends received on the way.

The central value moves to about $345 at an exit multiple of 15× and $425 at 19×; to about $400 at a 9% discount rate and $370 at 11%.

What is priced in? At $558.63, with a 17× exit multiple, Caterpillar needs to earn about $50 a share in 2031. That is growth of about 13% a year from this year's expected $27, or 17% a year from a normal $23. At a 19× exit it needs about $45. In other words, the price today is close to our high case: the analysts' path, in which the data-centre boom, infrastructure spending and a mining upswing all continue for five more years without a pause. That may happen. Caterpillar has never managed it before.

Compared with our Deere X-Ray. In August we called Deere "the bottom is in, and largely paid for": a cyclical at the low point of its cycle, on depressed earnings, at a price that already assumed recovery. Caterpillar today is the mirror image: a cyclical at a record, on peak or near-peak earnings, at a price that assumes the record continues. Same kind of business, opposite point in the cycle, opposite conclusion.

Why $300? It is about 22% below our central value, about 13 times normal earnings and a dividend yield of about 2.2%. For a business we rate 3 out of 5 on what can be known, in a cyclical industry, the usual rule would ask for 30% or more; $300 is a compromise in Caterpillar's favour, because its balance sheet and dividend record mean it will survive any downturn intact. Caterpillar last traded there around the start of 2024. It would take a recession, or the end of the data-centre boom, to get there again. Those are exactly the moments when a cyclical franchise is worth buying.