Data-centre cooling: how to test water-saving and heat-reuse claims
Ask for the complete heat path, an annual water balance and an hourly heat-demand match before accepting a cooling claim. A closed coolant loop does not tell you how heat leaves the site. Low water intensity does not prove low total water use. And recoverable heat has value only if a customer can use it at the required temperature and time. These distinctions belong in the concept brief, while plant space and service routes can still change.
Part 2 of 4: What the data-centre build-out is teaching construction.
Follow the heat to its final rejection point, not just around the coolant loop
Microsoft’s June 2026 explanation distinguishes evaporative cooling towers from its newer direct-to-chip designs. The distinction matters: a server-side loop can remain closed while a separate part of a cooling system evaporates water outdoors. Follow every heat exchanger to the final heat sink.
In December 2024, Microsoft described a new design using recirculated coolant and no evaporation for cooling. It also identified the limits of that claim:
- Scope: administrative uses such as kitchens and toilets still require water.
- Comparison: its predicted water saving uses a fleet-based baseline. It is not a measured result for every new facility.
- Trade-off: Microsoft expected a small increase in annual energy use relative to its evaporative designs. This does not establish the penalty for another climate or system.
- Status: the announcement described pilots and future openings. A design announcement is not an operating water meter.
Makao’s recommendation is to assess heat rejection separately from server cooling. Request the design load, local weather assumptions, permitted operating temperatures, part-load controls and performance during a plant outage. These determine whether the proposed arrangement can meet the brief without an unreported fallback water demand.
Microsoft reported a fleet-average water-use intensity of 0.27 L/kWh in 2025. A lower ratio can coexist with a higher annual volume if IT energy use grows. Compare both total litres and the ratio, using the same period and boundary. State whether the water quantity means withdrawal or consumption; those terms are not interchangeable.
Worked check: lower intensity, higher annual water use
Hypothetical example, not Microsoft or CDC data: 0.30 L/kWh × 100 million kWh of annual IT energy = 30 million litres (30 ML). At 0.20 L/kWh × 200 million kWh, the result is 40 ML. Intensity improves while total volume rises. This comparison assumes an unchanged water-accounting boundary and the same annual period; it says nothing about local water stress or the supply source.
The Australian check: what does the site water balance include?
CDC’s Laverton page describes cooling water recirculated in a sealed loop after the initial fill. It also identifies other water uses and rainwater collection. This is useful design information, but a complete water brief still needs volumes, supply sources and operating conditions.
Ask for separate estimates of the initial fill, routine make-up, maintenance drain-and-refill, any evaporative or adiabatic operation, and amenities or irrigation. Identify potable, recycled and rainwater inputs. Count each external input once; circulating water inside a loop is not a new withdrawal every time it passes the server.
In Finland, the heat network came before the waste heat
Fortum’s May 2026 account offers a useful sequencing lesson. It reported two Finnish heat plants operating before the expected arrival of waste heat from neighbouring Microsoft data centres:
- Operating at the time: heat production from ambient-air heat pumps and electric boilers, supported by thermal storage.
- Planned next: data-centre heat recovery introduced in phases from 2027, aligned with construction and commissioning.
- Projected result: about 40% of the local network’s annual district-heating demand once fully implemented. This was a forecast, not measured recovery.
The transferable lesson is the need for an available heat customer and delivery system. An Australian project without a nearby compatible heat load cannot assume the Finnish outcome. Establish demand, temperature, pipe route, connection cost, backup arrangements and a commercial agreement before booking a heat-reuse benefit.
The concept-stage review: ask for these six outputs
Output | What makes it decision-ready |
|---|---|
A complete heat-path diagram | Show chip or room → coolant or air → heat exchanger/chiller → final rejection or heat customer. Identify each water-using component. |
A water balance | Show initial fill separately from annual operation. List supply sources, evaporation, discharge and replenishment without double counting internal circulation. |
Peak and annual performance | Use the same IT load and climate assumptions for both options. Include high-temperature operation and the agreed resilience case. |
A whole-system energy comparison | Include chillers, pumps, fans and any heat-upgrading equipment. Report kWh and peak kW; do not infer either from water savings. |
A heat-use match | Match available heat with customer demand by time and temperature. Include transfer losses, heat-pump electricity, downtime and backup heat. |
A verification plan | Name meters, measurement boundaries, responsible operator, comparison conditions and review date. Separate estimates from measured results. |
Makao’s decision rule: retain a viable cooling solution even when the heat customer is unavailable. For heat reuse, assess the complete combined system against the same cooling and heating services delivered separately. Count the extra electricity, distribution losses and displaced heat supply. Record who can accept the design, what remains conditional and what evidence is needed before procurement.
For concept coordination, turn the chosen system into a spatial reservation matrix covering plant, electrical supply, access and heat rejection.
If a water-saving option changes electrical demand, revisit the grid-capacity decision before freezing the brief.
Sources
Dated project examples, linked below. The review outputs are Makao’s practical synthesis; they are not measured savings or a universal design specification.
Questions or corrections: contact Makao.
