AI Chips Need Liquid Cooling. That's Why Vertiv's Stock Rally Can Continue.

VRT is becoming a “picks-and-shovels on top of picks-and-shovels” exposure: the trade is no longer just new data-center builds, but the higher thermal content per megawatt as rack density rises. That matters because revenue per project can expand faster than headline MW capacity, and the service/maintenance layer should start to matter more than the initial install over the next 12-24 months. The market is probably still underpricing how much of the value chain shifts from one-time capex to sticky, specification-driven aftermarket revenue.

The bigger second-order winner is NVDA, but not because cooling is a direct revenue line; it is because thermal management reduces throttling and failure risk, which raises realized utilization and ROI for every deployed GPU cluster. More broadly, ETN and other electrical infrastructure names should capture an earlier portion of the wallet share than pure cooling vendors if power delivery remains the gating item, while smaller data-center operators with weaker balance sheets can see project IRRs get squeezed by higher all-in infrastructure costs. If financing costs and interconnect delays rise, some “planned” capacity simply won’t convert.

The near-term risk is that the stock is already discounting perfection. The first falsifier is not demand, but margin/booking deceleration: if backlog growth or gross margin inflects down on the next print, multiple compression can hit hard despite solid industry end demand. Over 1-3 months, the key catalyst is conversion of announced capacity into energized capacity; over 6-18 months, the key risk is that power availability, not cooling, becomes the bottleneck and caps the annuity narrative.