Zero-Water Data Center Cooling, Explained: Microsoft's Newest Region Uses Almost No Water to Cool
Air-cooled chillers instead of evaporative towers is a real engineering decision with a real cost. It moves the problem from the water bill to the power bill.
In this article 5 sections
Quick answer: Zero-water data center cooling replaces evaporative cooling towers with closed-loop, air-cooled chillers, so the facility consumes effectively no water. Microsoft uses this design in its Hyderabad region, launched September 21, 2026. The tradeoff is energy: air-cooled systems generally draw more power, especially in hot climates. You are trading a water number for a power number.
Microsoft formally launched its fourth India cloud region in Hyderabad on Monday, part of a $20.5 billion commitment to Indian digital and AI infrastructure. The region, India South Central, became generally available on August 6, 2026, and runs a three-zone architecture alongside Pune, Chennai, and Mumbai. Coverage of the Hyderabad launch notes the region is positioned as a hub for Asia and the Global South, with India's roughly 2GW of data center capacity expected to reach 6-7GW by 2035.
The launch detail worth reading past the press release is the cooling design. This explainer is for infrastructure engineers, capacity planners, and anyone who has to answer a sustainability questionnaire about their cloud footprint. It covers how zero-water cooling actually works, what the water numbers look like at national scale, what the energy tradeoff costs, and when any of this should change a region decision. Protocol Wire covers this as an operations question, not a marketing one.
How Does Zero-Water Data Center Cooling Work?
A conventional large data center rejects heat by evaporating water. Warm water from the server loop goes to a cooling tower, some of it evaporates, and that phase change carries the heat away. It works extremely well and it consumes water permanently, because the evaporated portion does not come back.
An air-cooled design skips the evaporation step. Heat moves from the server loop to refrigerant to ambient air through a chiller, with the water sealed inside a closed loop that gets topped up rarely rather than consumed continuously. Microsoft describes the Hyderabad facility as using effectively zero water for cooling on that basis.
Google and Meta have run variations on this same idea for years, usually branded around "free cooling" or air-side economization rather than a zero-water label. What makes the Hyderabad announcement notable is scale and region choice: pairing this design with a hot, humid climate is a harder engineering bet than deploying it in a cooler market like Dublin or Des Moines, where ambient air alone already does most of the work for free.
The word "effectively" is accurate rather than evasive. There is still water in the building for humidity control, for sanitation, and for the initial loop charge. What has gone away is the continuous evaporative draw, which is the number that shows up in a watershed impact assessment.
The mechanical piece worth understanding is water usage effectiveness, or WUE — liters of water per kilowatt-hour of IT energy delivered. A typical evaporative-cooled facility runs a WUE somewhere around 1.8 to 2.0 liters per kilowatt-hour, depending on climate and cooling tower design. An air-cooled, closed-loop facility like Hyderabad drives that number toward zero, since the loop water is not being evaporated away. In effect, it trades a metric reported quarterly for one that rarely has to be explained to a regulator.
What the Water Numbers Look Like Nationally
Scale makes this concrete. The 2024 United States Data Center Energy Usage Report from Lawrence Berkeley National Laboratory (LBNL) estimated that US data centers directly consumed roughly 17 billion gallons of water in 2023, primarily through evaporative cooling, and projected that figure could reach about 34 billion gallons by 2028.
The indirect number is larger and gets discussed less. LBNL put water tied to generating the electricity those facilities consume at roughly 211 billion gallons in 2023. That water is spread across power plants rather than concentrated in one watershed, which is why it draws fewer local objections and why it is the harder problem.
Here is the part that connects the two. The same LBNL report found data centers consumed 4.4% of US electricity in 2023, with a projected range of 6.7% to 12% by 2028. A cooling design that cuts direct water use but raises power draw is shifting load from the smaller number to the larger one. For the hardware driving that load, see the chips powering the AI boom.
The honest framing: zero-water cooling is not free sustainability. It is a decision that direct, local, visible water consumption is the impact worth eliminating first, accepting a higher energy figure to do it. In a water-stressed region that is a defensible call. In a coal-heavy grid it is a harder argument.
Put a rough dollar figure on it. A 20-megawatt facility running at a 1.4 power usage effectiveness ratio and switching to air-cooled at 1.55 is adding roughly 3 megawatts of draw. At an industrial rate near $0.08 per kilowatt-hour, that is close to $2.1 million a year in additional power cost, before accounting for peak-demand charges. That is the number a capacity planner should be modeling, not the marketing line about water saved.
The Energy Tradeoff Nobody Puts in the Headline
Evaporative cooling is thermodynamically efficient, which is exactly why the industry standardized on it. Take it away and the compressors work harder, particularly on hot days when the ambient air is close to the temperature you are trying to reject heat into.
In practice that shows up as a higher power usage effectiveness ratio, sometimes meaningfully higher in a climate like Hyderabad's. Operators offset it with rear-door heat exchangers, higher chilled-water setpoints, and liquid cooling at the rack, but the basic physics does not go away. Direct-to-chip liquid cooling helps most here, because it lets the loop run warmer and shrinks how much work the chiller has to do in the first place.
Facility retrofits tend to surprise operators in one place. It is not the design; it is the summer peak. A cooling architecture that looks fine on annual averages can produce a demand charge in August that nobody budgeted for, and those charges are set by your worst 15 minutes. Our piece on what running AI workloads actually costs once the invoice arrives covers how those infrastructure line items behave in a real bill.
There is also a climate ceiling most vendor pitches skip over. Air-cooled chillers lose efficiency fast once ambient temperatures push past roughly 95°F, since the temperature gap they have to work against shrinks. Hyderabad's peak summer highs regularly clear that mark, so the facility's worst-case power draw, not its annual average, is what should show up in a real capacity plan. A region that looks efficient on a mild March day can behave very differently in June.
When This Should Change Your Region Choice
For most teams, honestly, it should not. Latency, data residency, service availability, and price still decide region selection, and cooling architecture is several rungs below all of them.
A common question when evaluating an India region for the first time is "does the water story matter for us." Almost always the honest answer is no — unless one of the three cases below applies, cooling design is an operations detail the provider should handle without your input.
There are three cases where it genuinely matters, and they are narrow.
- You report water consumption. If your organization publishes environmental disclosures that include supply chain water, region choice changes your reported number without changing your code.
- You operate in a water-stressed area under public scrutiny. Local permitting and community opposition are real project risks, and they attach to water more readily than to electricity. The Environmental Law Institute's January 2026 fact sheet on data centers and water lays out how those siting fights typically unfold.
- You are signing a long-term capacity commitment. Water availability is becoming a siting constraint, which means it is becoming an availability constraint on the capacity you contracted for.
That third one is where I would spend the attention. Availability risk from a resource constraint behaves a lot like availability risk from a regional failure, and most disaster recovery plans handle neither well — a point our breakdown of the AWS Bahrain outage and regional redundancy works through in detail.
There is a fourth case worth naming even though it rarely makes the shortlist: procurement leverage. A region built around zero-water cooling is usually also a newer facility with a lower baseline PUE architecture overall, since operators rarely bolt this design onto old infrastructure. If you are already comparing two otherwise similar regions on price and latency, the cooling architecture can be a reasonable tie-breaker, not because the water number matters directly, but because it is a proxy for how recently the facility was built.
What to Take From This
Zero-water cooling is a real engineering choice with a real cost, not a marketing line. Microsoft picked it for Hyderabad because in that location, at that scale, eliminating direct water consumption was worth an energy penalty. That is a reasonable trade and it deserves to be described as a trade.
If you are planning capacity in the next year, ask your provider two specific questions: what is this region's cooling architecture, and what is its annualized power usage effectiveness. Both answers should be available, and a vendor that will not give you either has told you something. Protocol Wire covers platform shifts, chips, and product launches for the engineers and buyers who have to live with them after the announcement.
