Zero-Water Cooling for Data Centers: What Is Ready Now

Closed-loop cooling is moving from pilot to standard design, but the tradeoffs are real.

Microsoft says its next-generation datacenter design, launched in August 2024, uses chip-level cooling and avoids more than 125 million liters of water per datacenter each year. Its FY24 average WUE was 0.30 L/kWh, down 39% from 2021 and 80% from its first-generation owned datacenters. That is not a slogan. It is a design shift that changes AI Data Center Development, Data Center Site Selection and the way teams think about water risk.

What zero-water cooling actually means.

Zero-water cooling is usually a closed-loop liquid system.

Microsoft says the system is filled during construction and then circulates water between servers and chillers without requiring a fresh water supply for cooling. That is different from evaporative systems, which consume water, and different again from air-cooled systems, which shift more of the burden to ambient conditions.

The important part is not the branding. It is the physical logic. Power Analysis for Data Centers has to include cooling architecture now, because rack density, heat load and mechanical design all move together.

Why developers care.

Water is no longer a side issue.

It is a siting constraint, a permitting issue and a community issue. Microsoft says the new design avoids more than 125 million liters per datacenter per year, and it is piloting the system in Phoenix and Mt. Pleasant in 2026. That matters in arid markets and in places where local opposition can turn water use into a political fight.

A site that looks weak on water can still work if the cooling package is closed-loop and the power path is credible. That is why AI Site Selection and AI Due Diligence need to read cooling, utility and entitlement together. Water is part of the underwriting, not just the ESG report.

The tradeoff is power.

Microsoft is also clear about the downside. Replacing evaporative systems with mechanical cooling can increase power use.

The company says the resulting annual energy increase is nominal across the global fleet, but nominal is still a cost. For developers, that means zero-water cooling does not remove the infrastructure problem. It moves some of the burden from water to electricity and mechanical equipment.

That is why teams still need to model PUE, transformer capacity, utility service timing and the cost of higher-density systems. A site with perfect water optics can still be a bad deal if the power path is slow or too expensive.

What is ready now, and what is still early.

Ready now: direct-to-chip liquid cooling, closed-loop systems, hybrid air-and-liquid designs and reclaimed water strategies.

Still early: large-scale immersion cooling across mainstream portfolios, retrofits for legacy buildings that were not designed for AI density and universal standards across vendors. The market is moving, but not every portfolio can move at the same speed.

That distinction matters. Some of the portfolio can absorb the new cooling stack now. Some needs a longer retrofit cycle. The development team should know the difference before it locks in a site.

The siting implication is simple.

Zero-water cooling does not fix power scarcity. It just lets water stop being the first rejection reason.

In water-stressed metros, that changes the community conversation and can widen the site funnel. In power-constrained metros, it changes almost nothing unless the project also has a credible utility strategy. The best developers will model both at once, not sequentially.

FAQ

Q: Does zero-water cooling really mean zero water?
A: No. It usually means no water for cooling after the closed-loop system is filled. Facilities still use water for restrooms, kitchens and other administrative needs.

Q: What is the main advantage for data center developers?
A: It reduces exposure to water constraints, especially in arid markets or places with strong community sensitivity to water use. That can make some sites easier to permit and underwrite.

Q: What is the main tradeoff?
A: Higher power demand and more mechanical complexity. Lower water use does not mean lower infrastructure cost.

Q: Where does AI help in this decision?
A: AI can compare cooling options, model WUE and PUE tradeoffs and test whether a site survives with different water and power assumptions. Humans still decide whether the capex and delivery timeline make sense.