The Infrastructure AI Cannot Be Built Without

Synopsis

Approximately 80% of Gen Z university students are "concerned" or "very concerned" about climate change, with many extending this anxiety to the environmental toll of AI. Headlines warn of data centres drinking millions of litres of water daily, but water is the symptom, not the disease. The real problem is simply physics.

A single NVIDIA Blackwell GPU rack consumes over 100 kilowatts of power. The next generation pushes toward 240kW. At those densities, air simply cannot transfer heat out of a box fast enough. This is why liquid cooling is not optional, but rather a mandatory infrastructure for every AI data centre being built today.

Liquid cooling also solves the very problem that started this conversation. Unlike the evaporative cooling towers that give air-cooled data centres their water appetite, direct-to-chip liquid cooling systems are largely closed-loop. The coolant circulates, transfers heat, and returns. Some estimates suggest advanced liquid cooling reduces water consumption by up to 99% compared to traditional methods. The "AI water crisis" narrative is real, but the solution is already being built. 

Vertiv designs and manufactures the power systems, cooling units, and thermal management infrastructure that keeps data centres running across 130 countries and roughly 34,000 employees. With FY2025 revenue of $10.2 billion and a $15 billion backlog, Vertiv is an infrastructure company supplying something the world has already decided it cannot do without.

Catalysts

The liquid cooling market is already growing at roughly 20-30% annually as hyperscalers like Microsoft, Google, and Amazon race to build AI infrastructure. Vertiv's Q4 2025 organic orders surged 252% year-over-year, and its $15 billion backlog is equivalent to roughly 1.5 years of trailing revenue.

Critically, Vertiv hasn't just benefited from this tailwind passively; they've been co-engineering thermal solutions directly alongside NVIDIA for next-generation GPU architectures. When your product is effectively designed into the chip roadmap before the chip ships, that's a structural moat that competitors cannot easily cross.

Second, the December 2025 acquisition of PurgeRite expands Vertiv into liquid cooling services, introducing the recurring, high-margin end of the business, making their business model stickier and less cyclical, while further strengthening their competitive moat. 

Lastly, Vertiv’s strong free cash flow generation ($1.89B in 2025) and low net leverage (~0.5x) provide financial flexibility to scale capacity and pursue further acquisitions, a strong indication of future growth potential. 

Assumptions

Revenue in 5 years: Vertiv posted $10.2 billion in revenue for FY2025 and guided for 27% organic growth in 2026 alone. The liquid cooling market is still in early innings, most legacy data centres haven't been retrofitted yet, and every new AI facility being built requires Vertiv's infrastructure from day one. Being conservative, I expect growth to stay at 20-25% over the next 2-3 years, then moderate to about 15% as the market matures. That puts revenue somewhere around $22 billion by 2030, roughly 2.2x from today.

Earnings in 5 years: Margins are already expanding, as full-year adjusted operating margins reached 20.4%, with Q4 exiting at 23.2%. This reflects an accelerating operating leverage driven by the mix shift toward higher-margin liquid cooling products and services. As the PurgeRite services business scales and manufacturing productivity improves, I expect margins to push toward 26-28% by 2030. Combined with the revenue growth above, that implies adjusted operating profit in the range of $6-7 billion, roughly 3x today's level.

Risks

The biggest risk is the one hiding in plain sight: valuation. At roughly 28x EV/adjusted operating profit today, Vertiv is priced for continued perfection. Any stumble: NVIDIA pausing capex, a tariff shock, an earnings miss, all get punished hard at this multiple. This isn't a cheap stock, and likely never will be.

The second risk is customer concentration. A meaningful portion of Vertiv's demand flows through 4-5 hyperscalers. If Microsoft or Google slow their data centre buildout, as briefly observed after DeepSeek's efficiency revelations in 2025, then order momentum can shift quickly. The backlog provides a buffer, but it's cancellable.

Finally, competition is intensifying. Schneider Electric acquired Motivair in late 2024 specifically to enter the liquid cooling market. Eaton and Delta Electronics are also investing. Vertiv's head start and co-engineering relationships are real moats, but they're not impenetrable.

Valuation

In 3 years (2028), assuming ~22% revenue growth and margins around 25%, revenue of roughly $18 billion and adjusted operating profit of ~$4.5 billion feels achievable. A 22x multiple, a modest de-rating from today as growth normalises, implies an EV of approximately $99 billion, versus today's ~$58 billion. Meaningful upside if execution holds.

In 5 years (2030), using the $22-25 billion revenue and $6-7 billion profit figures above, and applying a more mature 18x multiple as the business shifts from hypergrowth to compounding, EV lands in the $110 billion range. That's roughly 90% upside from today's market cap of ~$57.6 billion, or a 14% annualised return.

Conclusion

Vertiv is not a bet on AI. Rather, it's a bet on the infrastructure AI cannot exist without. As GPU density climbs and liquid cooling shifts from emerging technology to industry standard, Vertiv's co-engineering relationships, expanding services business, and $15 billion backlog position it as the defining ‘picks-and-shovels’ play of the data centre era. The stock will never look cheap, but for a business supplying something the world has already decided it cannot do without, that's exactly what you'd expect.