The AI Infrastructure Premium: When Ordinary Industrial Businesses Reprice

Three transactions in the liquid cooling business, worth about $19 billion combined, were signed or completed within eleven days of each other this past March. The multiples attached to them were not industrial multiples. For owners of manufacturing, contracting, and component businesses, the interesting question is not why the buyers moved. It is what, exactly, they were paying for, and whether the same logic reaches companies further down the chain.

Nineteen Billion Dollars, Eleven Days

Eaton completed its acquisition of Boyd Thermal for $9.5 billion, at roughly 22.5 times the target's estimated 2026 adjusted EBITDA. Ecolab agreed to buy CoolIT Systems from KKR for $4.75 billion, which analysts placed at approximately 29 times next-twelve-month adjusted EBITDA and 24 times 2027 estimates. Trane Technologies acquired LiquidStack in the same window. Separately, TDK agreed to pay around $400 million for Fabric8Labs, a maker of cooling components.

For context, EBITDA (earnings before interest, taxes, depreciation, and amortization) is the operating cash flow measure most private company purchase prices are quoted against. Middle market transactions across all sectors averaged 9.8 times EV/EBITDA in 2025. Lower middle market manufacturing businesses typically trade between 5 and 7 times. Buyers in these cooling deals paid three to five times those benchmarks.

The contrast within industrials is just as sharp. Deal volume across the broader industrials sector fell by double digits, while thermal management and liquid cooling activity held up. The capital did not leave industrials. It concentrated inside a narrow band of it.

What the Premium Is Actually Paying For

The instinct is to assume buyers paid for technology. That explains part of it, but not most of it.

What Eaton and Ecolab bought was position in a supply chain with a hard physical constraint. Hyperscale data center operators need cooling and power capacity delivered on schedule, and the number of suppliers who can manufacture at that volume, to that specification, on that timeline, is small. Ecolab's purchase of CoolIT brought qualified production lines and, more importantly, existing procurement relationships with hyperscalers who had already tested and approved that hardware. Qualification takes years. Buying a qualified supplier is faster than becoming one.

That is the shape of the premium. It is paid for verified capacity and customer qualification, not for the elegance of the product. A buyer facing a demand curve it cannot serve will pay a great deal for certainty of supply, and very little for a promising design without a production record behind it.

The second element is integration. Eaton described the Boyd Thermal transaction as building a grid-to-chip solution, meaning it could sell power distribution and thermal management as one system rather than as parts. Hyperscalers increasingly want integrated systems from fewer vendors. Every acquirer in this wave was buying its way toward being one of those fewer vendors.

Inside the Chain Versus Near It

This is where owners of adjacent businesses need to be careful, because the distinction determines whether the premium applies to them at all.

A company sits inside the chain when its revenue is contracted or repeatedly ordered by parties building or operating this infrastructure, and when replacing it would cost the customer time. A company sits near the chain when it serves the same broad end market opportunistically, on a project-by-project basis, without qualification, contract, or switching cost protecting the relationship.

The distance between those two positions is worth several turns of EBITDA. Buyers know how to test it, and the test is not subtle. They will ask which customers are contracted and for how long. They will ask what portion of revenue is repeat versus one-time project work. They will ask whether the company is an approved vendor on a customer's qualified list, and what it took to get there. They will ask what happens to the relationship if a competitor bids ten percent lower.

Owners frequently overestimate their position here. Serving a data center construction project is not the same as being embedded in a data center supply chain. The first is a customer. The second is a moat.

How the Demand Reaches the Middle Market

The demand is real, and it does travel downward. Comfort Systems USA, which installs mechanical and electrical systems inside data centers, reported backlog of $12.45 billion at March 31, 2026, against $6.89 billion a year earlier, an increase of about 81 percent. Revenue for 2025 reached $9.10 billion, up from $7.03 billion in 2024, with technology projects, primarily data centers, accounting for 45 percent of the total. Public markets have priced that growth aggressively, at roughly 40 times forward earnings against a sector median near half that.

Backlog of that size does not stay with one contractor. It flows outward to electrical subcontractors, switchgear and busway fabricators, sheet metal shops, controls integrators, testing and commissioning firms, pipe fabricators, and specialty trades with the licensing and workforce to staff large projects. Many of those are privately held middle market businesses whose owners have watched their order books grow without fully connecting the growth to a repricing of their company.

The connection is worth making, but with discipline. Growth driven by a small number of large projects is not the same as growth in a durable franchise, and buyers price the difference.

The Durability Question

Every buyer paying a premium for AI infrastructure exposure is underwriting a view about how long the capital expenditure cycle lasts. Sellers should expect that view to show up as a discount on anything that looks like a spike.

Practically, this means a buyer will separate the business into a base and an increment. The base is what the company earned before this demand arrived and would earn if it went away. The increment is the recent surge. Base earnings tend to receive a normal multiple. The increment gets scrutinized, discounted, or shifted into an earnout, which is a portion of the purchase price paid later and only if performance targets are met.

The way to shorten that argument is with evidence that the increment is structural rather than cyclical: multi-year contracts, qualification status that would take a competitor years to match, capacity a customer helped fund, or engineering relationships that put the business into the design of a project rather than the bidding of one. Absent that evidence, a buyer will treat the surge as borrowed earnings and price it that way.

There is also a straightforward concentration risk. A contractor deriving most of its growth from two or three hyperscale programs holds meaningful exposure to decisions made by a very small number of capital allocators. That is not a reason to avoid the work. It is a reason to be able to show what the business looks like without it.

What to Watch

Three developments will determine whether this premium broadens or narrows over the next several quarters. First, whether hyperscaler capital expenditure guidance holds through the 2027 planning cycle, since the entire premium rests on demand visibility. Second, whether strategic acquirers keep buying capacity or shift to building it, which would reduce the scarcity value of qualified suppliers. Third, whether private equity moves down-market into the subcontractor and component tier, which is where a repricing would actually reach middle market owners.