How a Road Builder and an Auto Parts Maker Became AI Infrastructure: Sterling and Modine

The call a hyperscaler makes before it ever calls a chip company

Most data center conversations start with the chip. The actual sequence runs the other way.

Long before any silicon shows up, somebody has to move millions of cubic yards of earth across a 200 to 300 acre site, grade it dead flat, and bury the power and water underneath it. Nothing gets built until that is done.

The company taking that call is Sterling Infrastructure.

A few years ago, neither of these was a growth story

Sterling was a low-margin road builder that was losing money. Modine made auto parts. Neither belonged in a conversation about compounding. Pull them up on a screener back then and you would have seen two industrials grinding along near the bottom of a cycle.

Then AI data centers turned their existing capabilities into bottleneck resources almost overnight.

Turning point one: Sterling doubles its margin

Sterling's edge is not technology, it is org structure. It self-performs the entire scope of work under one roof. Because the job is not carved up across a chain of subcontractors, it never stalls waiting on the next crew to show up.

For a hyperscaler, that difference decides the contract. Every day a data center sits unfinished costs far more than a few percentage points on the bid ever would. So they do not hire the cheapest option, they hire the one that does not stop. That is real pricing power in an industry that almost never has any.

The numbers show the shift clearly:

• Data center site work is now roughly 70% of revenue and grew 174% year over year last quarter
• Operating margin climbed from under 8% to almost 17%
• Earnings per share are up more than 110% in two years, with guidance for them to nearly double again this year
• Backlog of $5.15 billion against $2.49 billion of revenue last year
• PEG near 0.88

Those last two lines are the whole argument. More than two full years of work is locked in before the year even starts, and you are paying almost no premium for it.

Turning point two: Modine, and the moment air stopped working

A normal server rack used to draw 5 to 15 kilowatts, which air could handle comfortably. An AI rack now pulls 50 to over 130.

At that density, air is not a design choice anymore — it is physically insufficient. The entire industry is being forced onto liquid cooling, piped directly to the chip.

Under its Airedale brand, Modine sells the whole stack: the chillers, the coolant distribution units, and the controls. Not one component of it. That matters because a hyperscaler can source an entire cooling system from a single vendor, which is exactly the kind of supplier you reserve capacity with years ahead of need.

• One customer committed to more than $4 billion of cooling gear from 2027 through 2029, with about $165 million paid up front
• The data center business grew 73% last year to roughly a third of the whole company
• Management is targeting close to $2 billion by 2028
• Adjusted earnings grew 24% while the legacy auto parts arm gets spun off to leave a pure play
• The stock trades a touch rich against its own fair value, but the PEG is 0.6

I compared the competitive setup in cooling more directly in Modine versus Carrier on AI cooling.

The one signal both companies share

I put these two in the same piece not because the businesses are similar, but because the evidence is.

Money paid in advance.

Modine has already collected $165 million against 2027 deliveries. Sterling starts its year with more than two years of revenue already contracted. That is not a demand forecast, it is a commitment — the customer did not offer an opinion, they wired funds.

In this cycle, that distinction is the one I weight most heavily. A total addressable market on a slide and a prepayment sitting on the balance sheet are not the same class of evidence.

What I will be checking from here

This is not a one-sided story. Four things I want to see each quarter:

Burn rate on the backlog. If conversion from backlog to revenue slows, the present value of that backlog falls with it.

Margin durability. Sterling's 17% operating margin depends on jobs not stalling. Labor and materials bottlenecks would hit that line before any other.

Customer concentration. A $4 billion commitment from one customer is a strength and a fragility at once. If that customer's capex plan changes, the picture changes with it.

Completing the spin-off. Modine only earns a pure-play multiple once the auto parts business is actually gone. Until then, some discount is fair.

Even with all that, I think this is the least crowded seat in the AI trade. These names get dragged down alongside the chip stocks in every sell-off, while their actual order books are booked out for years. The same logic runs through data center power picks and shovels.