Liquid cooling’s primary barrier is installation cost, as high as $5 million per megawatt for immersion systems, more than double the evaporative equivalent
Decision Focus
An industry analysis published by Data Center Knowledge examines why evaporative cooling persists across new and legacy data center builds despite the availability of alternatives. On the surface, the story reads as a facilities question. The operational signal for energy heads is sharper: every cooling technology decision is simultaneously an energy procurement decision. Cooling mix determines how many megawatts you need from the grid, how you model demand across your portfolio, and what your capex exposure looks like as the industry moves toward a liquid cooling transition. If that shift accelerates materially, it rewrites energy demand forecasts and the assumptions underpinning your current PPA book.
90-Second Brief
Now, the source article documents the economic and operational logic keeping evaporative cooling dominant: it consumes roughly 1,000 liters of water per megawatt-hour of IT load, translates to approximately $0.80 per megawatt in water costs under US municipal pricing, and uses less electricity than mechanical refrigeration alternatives. Liquid cooling’s primary barrier is installation cost, as high as $5 million per megawatt for immersion systems, more than double the evaporative equivalent. What the article names only in passing is the energy dimension: grid power scarcity is explicitly cited as one of the reasons facilities stay on evaporative systems. That makes the cooling choice downstream of your power procurement position, not separate from it.
What Is Really Happening?
Evaporative cooling persists not simply because it is cheap, but because it is cheap in a market where electricity is increasingly expensive and constrained. The source explicitly frames electricity generation capacity as a major barrier to data center expansion, noting that some facilities are now considering bypassing the grid entirely to operate their own generation. In that context, adopting mechanical refrigeration—which substitutes water for electricity—is not an infrastructure upgrade; it is an energy demand increase at a moment when adding load to your interconnection queue creates years of delay. Operators are rationally trading water volume for grid headroom. The feedback loop runs both directions: your power strategy is constraining your cooling options, and your cooling options are in turn shaping how much grid capacity you actually need.
Liquid cooling breaks this tradeoff only partly. Direct-to-chip and immersion systems carry minimal operating costs in both electricity and water, which should attract energy-constrained operators. But the installation cost barrier—at the multi-million-dollar-per-megawatt level for immersion—means the transition is more likely to happen on new greenfield builds than through retrofit of existing facilities, where conversion costs compound further.
Why It Matters for Global Heads of Data Center Energy
Three implications deserve direct attention in your portfolio planning cycle. First, your current power demand forecasts may rest on a cooling assumption that will not hold across the portfolio over the medium term. As liquid cooling adoption grows at new sites, those facilities will carry different MW demand profiles than the legacy fleet. A PPA sized against today’s cooling mix may be over- or under-hedged against a materially different power draw at the site level, particularly at high-density AI compute campuses where cooling load correlates tightly with GPU utilization.
Second, the capital exposure on cooling transition is not in your energy budget—it is in your infrastructure budget—but the decision lands in your operating model. At up to $5 million per megawatt for immersion, a 100 MW campus carries a ceiling of $500 million in cooling capex before a single watt of IT load is factored in. That figure reshapes site economics and may influence which interconnection queue positions are worth holding. Greenfield sites where liquid cooling is designed in from day one carry a fundamentally different cost structure than retrofitted assets.
Third, water risk is a geographic exposure that is becoming increasingly material to siting decisions. The source notes that water shortages vary in severity across regions. In markets where water access is tightening—parts of the US Southwest, Southern Europe, and water-stressed regions of Asia—a cooling technology mix that consumes millions of gallons per day becomes a regulatory and reputational liability that intersects directly with your site sustainability reporting.
Forward View
If liquid cooling adoption accelerates at new builds, two dynamics follow. Power demand per square foot of data center space may decline modestly at new facilities if immersion’s low electricity draw offsets thermal load—but that depends heavily on AI workload density continuing to rise in parallel, which may neutralize the efficiency gain. More concretely, any portfolio with a large legacy evaporative fleet will face a widening performance and cost gap against new competitors building liquid-cooled from the ground up.
In geographies where water regulation tightens ahead of grid improvements, operators may face pressure to adopt electricity-hungry mechanical refrigeration as a water conservation measure precisely when power availability is most constrained. That is the worst-case scenario for an energy head managing both a water compliance obligation and a stretched interconnection timeline.
What Is Still Uncertain
Several material gaps limit the precision of this analysis. Forecasts for liquid cooling’s share of new installations are attributed in the source to unnamed experts without a cited methodology, sample base, or geography; applying any specific adoption figure as a portfolio planning assumption carries meaningful risk. Water pricing used in the source ($0.003 per gallon under EPA estimates) reflects US municipal averages and will not transfer directly to international markets, where water cost and scarcity vary significantly. The $5 million per megawatt immersion figure is a ceiling estimate without a documented floor, and retrofit costs for existing facilities are described only qualitatively. No regulatory threshold—water withdrawal limits, carbon intensity standards, or grid injection rules—is cited as a confirmed forcing function for accelerating the transition.
One Question for Your Team
Does your power procurement model for the next PPA cycle explicitly account for a cooling technology shift at new sites—and if not, what demand assumption is it built on?
Sources
- Datacenterknowledge — Why Data Centers Still Rely on Evaporative Cooling (Link)
