The Real Water Cost of AI: Grid-Linked Data Center Use
As AI workloads surge, most water risk sits offsite—at the power plants feeding data centers. Leaders must price water into AI capacity, location, and energy strategy.
Executive Summary
The largest driver of AI data center water use often sits offsite at the power plants that generate electricity. As AI workloads scale, this indirect footprint becomes a material operational and reputational risk. Enterprises should expand water accounting, integrate water into siting and procurement, and modernize cooling and power strategies. Early movers will secure capacity, reduce volatility, and earn stakeholder trust.
- ▸The biggest water risk from AI data centers often lies in the electricity supply, not the facility.
- ▸Integrate water into AI siting, procurement, and design decisions now; do not wait for standards.
- ▸Modernize cooling and prioritize reclaimed water to cut direct draw and enhance resilience.
- ▸Procure energy with lower water intensity and consider time-matched clean electricity.
- ▸Expand disclosures to include indirect water and assign clear executive accountability.
Context
AI infrastructure is accelerating faster than most corporate resource models. While data center operators often track and report onsite water for cooling, the larger and less visible driver is the water used to generate the electricity that powers those facilities. In many regions, thermal power plants require significant water for cooling, and that indirect draw can exceed the water used within the data center itself. As AI clusters scale and power densities rise, the true water footprint of enterprise compute is increasingly tied to the grid mix, not just facility taps.
What’s changing
Attention is shifting from narrow metrics to system-level accounting. Traditional measures like Water Usage Effectiveness (WUE) focus on onsite consumption. Stakeholders are now pressing for a fuller picture that captures offsite water tied to electricity supply. Utilities in arid geographies face mounting constraints; some municipalities are tightening approvals for new data centers or imposing conditions around water sourcing. Enterprises are also renegotiating cloud and colocation contracts, seeking transparency on both direct and indirect water use associated with AI workloads.
Why this matters for the enterprise
- Risk concentration: AI-ready campuses cluster in select regions with available power, fiber, and land. Many of these same areas face water stress, exposing projects to permitting delays, curtailments, and reputational risk.
- Cost of resilience: Securing redundant non-potable supplies, upgrading to closed-loop or liquid cooling, and hedging with cleaner power can raise near-term capex. Delaying action risks higher operating costs and stranded capacity when constraints tighten.
- Investor and stakeholder scrutiny: Capital markets increasingly reward credible, transparent resource strategies. Reporting that omits offsite water tied to electricity looks incomplete as expectations evolve.
Immediate actions for leaders (next 90 days)
- Establish a total-water baseline: Combine known onsite water metrics with the water intensity of your electricity supply by location. Use utility disclosures and recognized lifecycle sources to estimate water per kWh where precise data is unavailable; document assumptions.
- Integrate water into site selection: Add water scarcity, grid water intensity, and municipal reuse infrastructure to your AI siting scorecards alongside power price, latency, and tax incentives.
- Update procurement language: For cloud, colo, and power agreements, request disclosures on facility WUE, the share of reclaimed water, and the water intensity of the power mix. Include improvement pathways and periodic reviews.
- Convene a cross-functional council: Bring CIO, COO, CFO, sustainability, and procurement together to embed water constraints into capacity planning and portfolio governance.
Technology levers to reduce water dependence
- Cooling modernization: Transition from open evaporative systems to closed-loop configurations, raise inlet temperatures where safe for equipment, and evaluate direct-to-chip liquid cooling to reduce evaporative needs.
- Non-potable and reuse: Prioritize reclaimed wastewater and industrial reuse where available, reducing draw on municipal potable systems.
- Grid mix optimization: Procure energy with a higher share of wind and solar, which generally have lower water intensity than thermal generation. Explore 24/7 clean energy strategies and demand shifting to periods with higher renewable penetration.
- Efficiency by design: Improve airflow management, hot-aisle containment, and workload placement to decrease overall cooling demand per unit of compute.
Procurement and contracting upgrades
- Power agreements: Embed water-aware criteria in energy procurement. Where feasible, align power purchase agreements with resources that reduce both carbon and water intensity.
- Cloud and colo SLAs: Request reporting on water metrics by region and workload class, including the share of reclaimed water and qualitative descriptions of local water risk. Favor providers with concrete roadmaps to reduce both onsite and grid-linked water dependence.
- Campus partnerships: Co-invest with developers and municipalities in shared reclaimed water and district cooling assets to spread cost and improve resilience.
Reporting and governance
- Expand disclosures: Supplement existing sustainability reports with narrative and methodology for estimating indirect (offsite) water use tied to electricity, similar in spirit to how organizations address indirect emissions.
- Decision rights and incentives: Assign accountability for water-integrated capacity planning, and tie executive incentives to resource efficiency targets that include both carbon and water.
- Scenario planning: Stress-test AI growth plans against drought scenarios, permitting delays, and fuel mix shifts, and map triggers for capital reallocation or workload rebalancing.
Risk landscape
- Regulatory: Some jurisdictions are tightening water-related requirements for new data center capacity. Enterprises without credible plans face slower approvals and additional compliance costs.
- Operational: Droughts, heat waves, and grid constraints can drive curtailments, forcing workload migration or throttling AI training schedules.
- Reputational: Community pushback is rising where industrial water use conflicts with residential and agricultural needs. Transparent engagement and local benefits matter.
Questions for the board
- How are we quantifying total water per unit of AI compute across our portfolio, including offsite power generation?
- What siting, procurement, and design choices will reduce our water risk without compromising performance or time-to-market?
- How will we govern trade-offs among cost, latency, carbon, and water as AI demand grows?
Executive Perspective
AI is redefining infrastructure economics, and water is now a first-class variable in the compute equation. Treating indirect water use as a line item—alongside carbon and cost—will separate leaders from laggards as capacity tightens in key markets. Executives should not wait for universal standards to act. Establish a pragmatic methodology, disclose assumptions, and iterate. The organizations that align AI growth with water-smart energy and cooling portfolios will build more durable, scalable, and investable platforms.
What This Means for Organizations
Operationally, CIOs and facility leaders must co-own a water-aware capacity plan. This requires new telemetry and modeling across locations to capture both onsite water demand and the water intensity of each grid. Procurement must rebalance SLAs and PPAs to reflect water criteria without eroding performance or resilience. Structurally, cross-functional governance is essential. Create a joint forum where finance, operations, sustainability, and product leaders adjudicate trade-offs between compute growth, latency, cost, carbon, and water. Incentives and KPIs should reward resource efficiency per unit of AI output, not just absolute capacity added.
Strategic Impact
Enterprises that internalize water risk will gain a structural cost advantage in constrained markets, improve siting optionality, and preserve reputational capital. This discipline will also diversify energy portfolios toward sources that tend to be both lower carbon and lower water intensity. Strategically, clarity on total resource costs enables better make-versus-buy decisions for AI workloads, smarter selection of cloud regions, and more resilient product roadmaps that do not hinge on fragile local conditions.
Operational Implications
Expect shifts in facility design toward liquid and closed-loop cooling, broader use of reclaimed water, and tighter airflow and workload orchestration to reduce cooling loads. These choices can change maintenance regimes, vendor rosters, and spare parts inventories. On the energy side, procurement will lean into contracts that pair carbon and water performance, including time-matched clean energy, while operations invest in demand flexibility to align compute with lower-water-intensity generation windows.
Future Outlook
Disclosure expectations will expand beyond onsite consumption to encompass indirect water tied to electricity. This will spur better data, methodologies, and third-party assurance, enabling more consistent benchmarking across regions and providers. Technologically, AI hardware efficiency and advanced cooling will continue to improve, but siting near robust non-potable water infrastructure and cleaner grids will be the decisive edge. Organizations that build flexibility into workload placement and energy sourcing will be best positioned to navigate environmental volatility.
- • Access to AI capacity may hinge on water constraints, influencing time-to-market and regional strategy.
- • Capital allocation must reflect water-adjusted total cost of compute across sites and providers.
- • Transparent water governance can improve stakeholder confidence and lower perceived risk.
- • AI workload placement should consider the water intensity of the grid, not just latency and cost.
- • Training schedules can be tuned to periods with higher renewable generation to reduce water-linked risk.
- • Cooling choices for high-density AI clusters will shape both performance and water dependence.
This analysis was inspired by reporting from AI Data Centers Use Far More Water Than Most Tech Giants Report. All analysis, commentary, and strategic perspective is original work by Geraldine Vilato.