Caterpillar's full-year 2026 tariff guide came in at $2.2-2.4 billion, down from a prior $2.6 billion estimate, with Q2 tariff costs forecast around $700 million — a 75% jump from a year ago. Then management said the line that matters: the company's two-percent pricing action is "explicitly not intended to cover the tariff expense." Volume growth and AI data center demand will cover it. [1]

That distinction is what this paper missed yesterday when the framing was "$710 million of tariffs absorbed by pricing." The $710 million of unfavorable manufacturing costs is real, and the $426 million of new price realization is real, but management has now said on the record that the price action and the tariff bill are not the same conversation. The pricing is a normal-course adjustment; the tariff offset comes from somewhere else. The somewhere else is power-generation revenue, which grew 41% to $2.82 billion in Q1, with most of that growth tied to data-center demand. [2]

The order backlog corroborates the framing. Caterpillar exited Q1 with a record $63 billion order book, up 79% year-over-year. The construction segment grew 38% on revenue and the power-and-energy segment grew 22%. Management told the call it expects to operate at the midpoint of its adjusted operating-margin target range — a guide that would not be reachable if pricing were doing the tariff work, because the resulting margin compression would cap operating leverage. The AI data-center power-generation product line is the variable absorbing the tariff hit, and CAT is now scaling reciprocating-engine capacity nearly threefold by 2030 to keep up. [3]

The Apr 30 framing was that Caterpillar is an AI-infrastructure proxy. Today's clarification — pricing is not the tariff offset — sharpens that frame. CAT is no longer a domestic-construction story or a global-mining story; it is the company that builds the engines that power the data centers that run the AI models that justify Meta's $145 billion capex and Microsoft's $190 billion capex. The tariff bill is a friction cost on that conversion, not a structural problem. The buy-side cheered the print: CAT touched a record high Wednesday and held the gain into Thursday, even as the tariff guide implied $700 million of Q2 cost pressure. [4]

The framing matters for how to read the next four industrial earnings prints. If pricing is not the tariff offset, then the test for any other industrial — Eaton, GE Vernova, Cummins — is whether their AI-data-center exposure carries the cost. CAT is the cleanest case because power-generation engines are a discrete reportable segment and the data-center end-market is named explicitly in the earnings deck. For diversified industrials whose AI exposure is buried in mixed end-markets, the pricing-versus-volume question will have to be inferred. CAT just told the market how to do that inference: ignore the price line, watch the data-center volume.

The macro question this opens is whether the AI capex cycle continues to absorb tariff costs across the industrial complex. Caterpillar's near-tripling of large-engine capacity by 2030 implies a forward order book that the buy-side believes is real demand. If Meta, Microsoft, Alphabet, and Amazon collectively pause or cut capex at any point in the next eighteen months, the volume cushion goes away and the tariff bill becomes a margin problem. The bull case is that AI capex is a structural cycle and CAT is positioned correctly. The bear case is that AI capex is a 2024-2027 spike funded by Meta's $25 billion bond raise and Apple's memory-cost-bound guide, and the cushion is only as durable as the funding chain.

Caterpillar has been around long enough to know the difference between a cycle and a regime change. Today's guide treats AI data center demand as a regime — a multi-year capacity expansion underwriting a multi-year tariff absorption. If that read is wrong, the tariff bill stops being a footnote and starts being a story. For now, the market is willing to take CAT at its word.

-- THEO KAPLAN, San Francisco