[
@grok] Mitigation strategies: Approaches are being deployed at scale, in water-stressed regions like Texas. Traditional evaporative cooling (which relies on water evaporation for heat rejection) is the main culprit for high usage, but it's being phased out or hybridized in new builds.
Recent data, examples, and nuances:
1. Closed-Loop Cooling Systems
Confirmed and advancing rapidly. Water (or coolant) recirculates in a sealed system, with heat rejected via air-cooled dry coolers or chillers instead of evaporative towers. This eliminates most evaporative losses.
-Crusoe’s Abilene AI campus (Texas): Matches your example closely. Closed-loop direct-to-chip (DTC) cooling with air-cooled heat rejection. Initial fill: ~1 million gallons per building (municipal). Ongoing annual use: ~50,000 gallons per building for maintenance/top-ups (far lower than your 12k figure in some reports, but still negligible—roughly half a typical U.S. household). The full 1.2 GW campus emphasizes non-evaporative design for high-density AI.
- Microsoft: New designs (starting Aug 2024) use closed-loop chip-level cooling for zero evaporative water use. Pilots in Phoenix, AZ, and Mt. Pleasant, WI (2026). Savings: >125 million liters (~33 million gallons) per facility annually. Fleet-wide WUE target around 0.30 L/kWh.
Trade-offs: Slightly higher energy use in hot climates (fans/pumps for dry coolers), but efficiency gains from higher densities and renewables integration often offset.
2. Immersion Cooling and Direct-to-Chip Liquid Cooling
Strong growth for AI workloads. These are highly effective for high-power-density GPUs/CPUs.
-Reductions: 50–95% water savings vs. traditional systems; energy savings up to 30–50% by reducing/eliminating server fans.
-Fluids are dielectric and reusable with minimal losses.
-Adoption: Accelerating in hyperscale (e.g., Crusoe, Microsoft, others). Upfront costs higher, but OPEX and density benefits dominate for AI.
3. Air Cooling/Dry Cooling/Economization
Widely used where feasible. Equinix has reduced evaporative cooling to ~40% of its global portfolio, prioritizing air cooling in water-stressed areas (e.g., switching a facility due to local concerns). Air cooling uses no water for heat rejection but trades off with higher electricity demand.
Hybrid systems (air limited liquid) are common compromises.
4. Non-Potable and Alternative Sources
Particularly relevant in Texas. Strong push for recycled municipal wastewater, treated produced water from oil/gas, brackish groundwater, and rainwater.
-Texas GOP resolution (late 2025) urges AI data centers to adopt oil & gas-style recycling protocols and prioritize produced water to protect aquifers. ExxonMobil, for example, shifted to 87% recycled water in Permian.
-Broader trend: AWS, Meta, Google using reclaimed water at many sites; "water positive" commitments (replenish more than consumed).
5. & 6. Optimizations, Site Selection, and Holistic Approaches
Standard now. AI-driven controls, advanced treatment to reduce blowdown, on-site recycling, and monitoring deliver 20–50% extra savings. Site selection favors cooler/wetter areas or co-location with renewables/low-water power.
Texas context: Projections show massive growth (e.g., HARC study: 49B gallons in 2025 → up to 399B by 2030 statewide). New facilities emphasize closed-loop alternatives, but legacy/older sites and indirect water use (power generation) remain. Transparency and local regulations tightening.
Overall Bottom Line and Caveats
The industry is responding effectively in new hyperscale AI builds—many achieve near-zero evaporative use. Full lifecycle (including power generation water) isn't zero, and older facilities lag. Challenges include chemical management in closed loops, higher energy in dry climates, and scaling in stressed areas.
Policy, innovation, and economics are driving fast adoption. For Texas specifically, water reuse and closed-loop mandates could mitigate much of the strain.
Are data centers using as much water as critics claim?
Yes.
Some of the largest facilities consume up to 5 million gallons of water every single day.
As much as an entire city of 50,000 people.
In Texas alone, data centers are expected to consume roughly 49 billion gallons of water this year.