Vertiv (VRT): what it does, how it makes money, and why AI racks need its cooling

The standard $VRT story is that Vertiv is "the AI stock that is not a chip stock": a picks-and-shovels industrial riding the data-center building boom. That story is half-right, and the half it misses is the half that sets the price.

The more accurate frame: the bet is not "more data centers get built." It is "every rack that does get built carries more Vertiv content," because the current generation of AI accelerators runs too hot for air to cool. Vertiv makes the equipment that keeps a data center alive (the cooling, the power conditioning, the racks that bundle them), and the physics of dense AI compute is pushing more of that equipment into every rack. This piece walks through what Vertiv actually sells, how it makes money, where it sits in the AI infrastructure stack, and the concentration risk underneath the order book. Figures are as of 2026-09-11 unless noted.

Why it matters now

In the year to 2026-09-11, VRT rose about 89%. Over the last three months of that window it gave back about 15%, and about 10% in the final month alone. Here is the part that makes the evergreen question worth answering: over that same stretch, backlog kept growing roughly 25% year over year and book-to-bill stayed above 1.0 for an eighth consecutive quarter. Nothing in the disclosures broke while the stock came off. That points to a multiple repricing rather than a business cracking, and it is exactly why the durable question, what does this company actually own, deserves a careful answer instead of a read of the tape.

What does Vertiv do?

Vertiv is the least glamorous name in the AI trade, and that is the point. It does not design chips and it does not own data centers. It makes the physical plant that keeps a data center alive.

Thermal comes first: air cooling, and increasingly liquid cooling, coolant distribution units (CDUs), and rear-door heat exchangers. Power is the second pillar: uninterruptible power supplies (UPS), switchgear, busways, and rack-level power distribution. The third piece is the rack and row systems that bundle thermal and power into a single deployable unit. Roughly three quarters of revenue is product and a quarter is service, split about half Americas, 30% Europe and the Middle East, and 20% Asia Pacific.

The structural bet: content per rack

Here is the whole thesis in one sentence: every incremental gigawatt of AI compute pulls more Vertiv content into each rack. A dense accelerator rack does not simply need more power. It needs a different kind of cooling, because air stops working at the densities this generation already runs. The coolant has to reach the chip.

QA's own thesis has the liquid-cooling attach rate on new racks inflecting from under 10% toward more than half. That is a QA estimate, not a disclosed figure, and it deserves to be held loosely. What is disclosed is the product line built to capture it: Vertiv bought CoolTera in 2024 for direct-to-chip cooling intellectual property, released the CoolPhase Flex coolant distribution unit aimed at megawatt-class rack density, and published a joint reference architecture with $NVDA for liquid-cooled NVL72 racks. The estimate is the upside; the product line is the fact.

How Vertiv makes money

Vertiv sells equipment and then services it, so the model is product revenue plus a recurring service tail. Trailing revenue is about $11.5B (as of 2026-09-11). The service quarter of the business is the stickier part: installed equipment needs maintenance, and that revenue recurs.

The measurable part of the story is the order book:

• Backlog was reported at about $7.9B, up roughly 25% year over year. Against ~$11.5B of trailing revenue, that is roughly eight months of work already signed.
• Book-to-bill above 1.0 for eight consecutive quarters. Above one means orders arrive faster than revenue is recognized, so the backlog builds rather than drains. Eight quarters in a row is the difference between a lumpy win and a trend.

The headline risk sits right under that backlog: concentration. The top four hyperscalers represent more than 40% of orders, and they are not named in the filings. A framework agreement with a customer you cannot see is harder to underwrite than one you can. That is the number to hold onto through the rest of this piece.

Where it sits in the AI infrastructure bubble

Vertiv is a core name in QA's Cooling / DC Infra bubble, the physical plant of AI: cooling, switchgear, and prefabricated data-center components. It maps onto the AI Power theme, the cluster of names that turn electricity into usable, cooled compute capacity.

The useful way to place Vertiv is by layer. It is not the chip (NVDA). It is not the power plant either: independent power producers like Vistra, grid and generation equipment like GE Vernova, and on-site power like Bloom Energy sit upstream, making the electricity. Vertiv sits at the building and rack level, conditioning that power and removing the heat. And it is distinct from die-level power delivery, the 48V-to-point-of-load problem right at the chip, which is Vicor's layer. Building-level cooling and power (Vertiv) and die-level power delivery (Vicor) are two different bottlenecks on the same AI rack.

Its peers are revealing. Empirically, among VRT's tightest correlations in the QA universe (252 trading days, market beta stripped out) are Eaton ($ETN, 0.74), nVent ($NVT, 0.70), Comfort Systems ($FIX, 0.68), Advanced Energy ($AEIS, 0.68), Caterpillar ($CAT, 0.67), and Quanta Services ($PWR, 0.63). That is mostly the electrical and industrial complex rather than the semiconductor one. Vertiv trades like what it is: an electrical-equipment maker whose demand curve happens to be set by AI capex. The full, daily-refreshed peer table is on /stocks/vrt. Its credible competitors at scale are a short list: Schneider Electric, Eaton, and ABB.

The numbers

Two of those rows deserve a caveat. The trailing P/E overstates what you pay for next year's earnings on a name whose order book is growing faster than its booked revenue; QA's own bear note still puts the forward multiple above 40x, which is an estimate, not a filing. Either way the description holds: an electrical-equipment maker priced like a semiconductor company. And the Street mean target is an estimate, an average of analyst opinions, not a promise or a QA view; it is the first thing cut when a hyperscaler pushes an order a quarter to the right.

The bull case

• The order book is measurable, not narrative: backlog up ~25% year over year and book-to-bill above 1.0 for eight straight quarters.
• Rising content per rack. A dense liquid-cooled accelerator rack carries materially more cooling and power equipment than the air-cooled generation it replaces.
• A real product line aimed at the transition: CoolTera direct-to-chip IP, the CoolPhase Flex CDU, and a joint NVDA reference design for liquid-cooled NVL72 racks.
• A growing service tail that recurs and sticks, on top of a product base with few credible competitors at scale.

The bear case

• Concentration. The top four hyperscalers are more than 40% of orders, and they are not named. That is hard to underwrite.
• The multiple on a cyclical line. AI capex does not have to fall for the damage, it only has to slow; a stock priced above 40x forward earnings (QA estimate) does not reprice gently.
• Commoditization. Liquid cooling is hard today and gets less hard every year, and server builders such as $SMCI and the ODMs have every reason to pull the cooling loop in-house.
• Tariffs on China-sourced components land directly on the cost line.
• Heavy insider selling through 2025 and into 2026, the kind of signal that shows in the price before it shows in a filing.

None of these breaks the business in a given quarter. Together they explain how a double-digit drawdown can happen with no bad news in the disclosures at all.

How to access

Vertiv trades on the NYSE as VRT, a clean US listing, so direct ownership needs nothing exotic. To trade it from a US-retail account alongside the rest of the AI-infrastructure names, see /stack/ibkr. The live ETF holdings breakdown, showing which funds carry VRT and at what weight, is on /stocks/vrt.

Bubble-correlation shifts and rule-based alerts on $VRT, the kind that fire when a name crosses a curated level or its bubble correlation breaks, are part of /pro.

What to watch

• Book-to-bill in the next print. It is the single cleanest read on whether the order book keeps building, more telling than any price target.
• Backlog trajectory: whether the ~25% year-over-year growth holds, and how much of it converts to recognized revenue.
• Hard evidence on liquid-cooling attach, the variable QA's thesis leans on but that is not yet a disclosed figure.
• Hyperscaler capex guides. Vertiv's demand is set by a spending cycle it does not control; a slowdown there reaches the order book fast.
• The curated $192 level (marked 2026-04-25, well below the recent ~$257 tape) is best read as an observable structural reference, not a trade level or a target. The other observable marker is roughly $298, where the stock traded three months before 2026-09-11.

Live data on this ticker: /stocks/vrt. Price, ETF holdings, bubble correlation, curated levels, bot positions.

Bubble context: /bubbles/cooling-dc-infra. The cluster this name belongs to and how it's moving.

QuantAbundance is educational research. Nothing here is investment advice. See /disclosures.