Every conversation about AI infrastructure eventually turns to power. But there is a second resource constraint quietly shaping where and how data centers get built: Water.
| 4.2 – 6.6 billion m³ AI’s projected global annual water consumption by 2027 — four to six times the annual water usage of a country like Denmark. |
Writing a single 100-word email with AI consumes roughly 519 milliliters of water —about the same as a standard plastic water bottle. This water is used indirectly by the massive data centers required to cool servers and generate the electricity that powers your prompt.
Most of that demand traces back to cooling. Modern hyperscale facilities now run at power capacities of roughly 100–1,000 MW, and the heat these loads generate requires continuous cooling — with widely used methods like evaporative cooling depending on substantial freshwater withdrawals. As a result, water has become a key design constraint for AI data centers, alongside power, land, and connectivity, especially in regions facing scarcity — and Water Usage Effectiveness (WUE) is emerging as a critical metric for evaluating cooling strategies.
This is where civil engineering has moved from a supporting role to a first-order design discipline. Site selection, water sourcing, and stormwater strategy now happen alongside — not after — the mechanical and electrical planning. A few approaches are leading that shift:
| Reclaimed and municipal wastewater as a primary cooling source. Instead of drawing on potable supplies, engineers are increasingly routing treated wastewater into cooling systems — lowering environmental impact and easing regulatory hurdles. | |
| Zero liquid discharge (ZLD) treatment systems. In water-stressed regions, advanced treatment technologies enable high-reuse systems that significantly cut freshwater withdrawals by recovering and recirculating nearly all process water. | |
| Data center–wastewater treatment plant (WWTP) symbiosis. Emerging research models are pairing data center siting decisions directly with nearby treatment plant capacity — turning site selection itself into a water strategy. | |
| Utility-partnered conservation programs beyond the fence line. One recent initiative in the Brazos River Watershed is expected to conserve roughly 26 million gallons of water over five years through a verifiable, long-term measurement program. |
Designing Out Water Waste
Before a facility ever pulls from an alternative source, good civil and MEP design squeezes waste out of the water it already uses:
| Closed-loop cooling with blowdown treatment and reuse. Rather than sending cooling tower blowdown to discharge, engineers are routing it through constructed wetlands or on-site treatment and recirculating it — cutting waste while providing habitat and stormwater co-benefits. | |
| Real-time leak detection and predictive maintenance. Instrumenting valves, pumps, and cooling loops lets operations teams catch failures early — before a slow leak becomes a major water loss — while also extending equipment life. | |
| Right-sizing cooling technology to climate and load. Pairing adiabatic, direct-to-chip, and immersion liquid cooling with the local climate — rather than defaulting to evaporative cooling everywhere — can meaningfully cut water use, since one large evaporative-cooled facility alone can consume up to 1.8 billion gallons a year. | |
| Water quality engineering that extends reuse cycles. Treatment systems designed to maintain the right hardness and chemistry let the same water cycle through cooling loops more times before it needs replacing, reducing overall withdrawal. |
Designing the Building to Stay Cool Naturally
The least expensive gallon of water is the one a facility never needs. Passive design — using the building envelope and site orientation to manage heat before mechanical systems ever engage — is becoming a core civil and architectural strategy:
| Building orientation and massing to minimize solar heat gain on the walls and roof areas most exposed to direct sun, reducing the cooling load mechanical systems have to offset. | |
| Thermal mass, insulation, and reflective or low-e building envelopes that slow heat transfer into the facility, paired with fixed shading on server-hall facades where daylight isn’t needed. | |
| Natural and free-air ventilation strategies, taking advantage of local climate and diurnal temperature swings so mechanical cooling — and the water that often comes with it — only kicks in when outside conditions require it. | |
| Green and vegetated roofing, which adds a natural insulation layer, reduces the urban heat island effect around the site, and can double as a landscape feature that supports biodiversity. | |
Capturing What Falls On-Site: Rainwater Harvesting
Rainwater harvesting has moved from a nice-to-have to a standard line item in data center site design, particularly where evaporative or hybrid cooling is still part of the strategy:
| Roof and site capture systems collect runoff for filtration and storage, supplementing — or in some regions substantially offsetting — potable water drawn from municipal supply for cooling tower makeup and humidification. | |
| Sizing tied to actual demand. Systems are engineered around cooling load, roof and site catchment area, and local rainfall patterns, so storage capacity is matched to real seasonal water needs rather than built as an afterthought. | |
| Treatment matched to end use. The level of filtration required depends on the collection surface, contaminant load, and whether the system feeds a single-pass or recirculating cooling loop — engineering decisions made early in site design, not bolted on later. | |
| Stormwater and rainwater harvesting as a single system. Because civil engineers are already designing stormwater management for permitting and flood control, integrating harvesting into that same infrastructure is often a low-marginal-cost way to add a genuine water source. |
Leading operators are already showing what’s possible at scale: some hyperscale data centers now source the large majority of their water from recycled, reused, or non-potable sources in their most water-stressed regions, and new-generation designs are piloting zero-water evaporative cooling entirely — relying on outside air until temperatures cross a threshold that makes water use necessary at all.
None of this happens without groundwork most people never see. Civil engineering is foundational to data center construction — enabling access to power and water, compliant stormwater systems, and resilient, buildable land long before a single server is installed. Site assessment, permitting, and hydrology studies determine whether a proposed facility is even viable in a given watershed.
| The industry still has a visibility problem. In many regions, data center water use is aggregated within utility reports rather than disclosed at the facility level, and no centralized public database tracks withdrawals — leaving limited insight into how quickly this demand is affecting local water stress. Better design can only go so far without better data. |
THE TAKEAWAY
Water-conscious design is not a sustainability add-on anymore — it’s becoming a permitting requirement, a community relations necessity, and increasingly a competitive advantage for developers who can prove their facilities are good neighbours to the watersheds they sit in. Civil engineers sit at the center of that shift.
