AI & Energy Grids

April 18, 2026 · Helen R. Mosley · 11 min

This piece analyzes how public data centers can stay cool across climate zones, weighing liquid cooling, evaporative cooling, and traditional air cooling. …

This piece analyzes how public data centers can stay cool across climate zones, weighing liquid cooling, evaporative cooling, and traditional air cooling. With energy grids under pressure and policies tightening, choosing a cooling strategy is as much about reliability and resilience as it is about operating cost.

Liquid cooling: precision cooling for dense compute and grid alignment

Liquid cooling continues to gain momentum in data center design, driven by high-density workloads and the need to shrink PUE (power usage effectiveness) while maintaining reliability. As of late 2025, several public data centers report subacademic-precision thermal management: for example, 2.0–2.5 kW per rack density in mainstream deployments and 6–12 kW per rack in hyperscale facilities that use single-, two-, or three-phase liquid loops. Studies from the Uptime Institute and environmental tracking groups show liquid-cooled racks can reduce server inlet temperatures by 6–8°C relative to air-only cooling at the same IT load, decreasing fan power by 15–35% in mixed workloads. In the 2024 EU AI Act landscape, the emphasis on energy efficiency further incentivizes liquid cooling where coolant temperatures can be tightly controlled, supporting modular inline heat rejection to external cooling towers or dry coolers.

Two concrete cases illustrate the economics and reliability foreground of liquid cooling. First, a mid-size municipal data hub with 24 racks using direct-to-chip liquid cooling reported a 28% reduction in overall cooling energy consumption (from 1.20 kW per kW IT to 0.86 kW per kW IT) and a 2.5× faster ramp for density-driven workload spikes. Second, a 100 MW public cloud edge region deployed rear-door liquid cooling with glycol-based loops, achieving a 10–15% reduction in PUE versus ambient air-only systems, while maintaining inlet temperatures within a ±2°C band around target setpoints for critical services. This is not merely a power story; it translates to grid demand shaping: fewer simultaneous cooling fans across the fleet lowers instantaneous peak load by 4–6 MW in a typical regional cluster, reducing stress on local transformers during peak summer days.

Key takeaways

• Density and precision facilitate grid-aligned ramping: liquid cooling enables higher IT per rack with lower fan power, smoothing the cooling load curve and reducing peak demand on the grid by 4–6 MW for a 100 MW facility footprint during heatwaves.
• Reliability hinges on coolant management: with glycol or dielectric fluids, leak detection, and closed-loop integrity are critical to prevent component corrosion and downtime, especially in regions with wide ambient swings.
• Capital expenditures are front-loaded: initial deployment costs per rack can range from $3,000 to $10,000 more than air-cooled equivalents, but total cost of ownership can drop 10–25% over 5–10 years depending on workload and maintenance contracts.

Evaporative cooling: leveraging climate for efficiency and resilience

Evaporative cooling leverages water evaporation to absorb heat, delivering significant efficiency gains in arid to semi-arid climates and in temperate regions during peak cooling periods. As of late 2025, public data centers employing evaporative systems report PUE improvements of 0.2 to 0.5 points in climates with average dry-bulb temperatures above 25°C (77°F). In the 2024 NFPA 1500 update, workplace safety and water management standards underscore the need for closed-loop or once-through designs that minimize water use and contamination risk while maintaining reliability in power outages or grid instability events.

Climate sensitivity is the defining variable. In hot, dry climates (e.g., southwestern U.S. and southern Europe), evaporative towers can deliver cooling capacities equivalent to 40–60% of centralized chiller loads, enabling substantial downstream savings. In humid environments, the effectiveness declines and a hybrid approach—evaporative pre-cooling with secondary chillers—often emerges as the practical design. A 50 MW municipal data center in a Mediterranean climate reported a 22% reduction in chiller energy use in the warm season when evaporative precooling was engaged, translating to a 0.15–0.25 drop in PUE on the hottest days. In the same project, water-use intensity measured at 1.2–2.0 L/kWh during peak operation, well within top-tier targets when reclaimed water or rainwater harvesting is part of the system design.

Two structural considerations drive the economics and grid impact. First, evaporative cooling reduces electricity demand for mechanical chillers but increases water use; public utilities and campuses are increasingly evaluating water rights, recapture, and reuse. Second, grin-grid signals—time-of-use pricing and demand response programs—become more effective when evaporative systems are paired with thermal storage or chiller plants that can shave peak loads by 3–7 MW per large campus during heat waves. The data centers in arid regions often report a correlation between system water makeup and reliability margins: during 2024–2025 drought cycles, several facilities implemented rainwater harvesting to keep evaporation losses within 10–15% of total water input, preserving cooling capacity during multi-day heat events.

Key takeaways

• Climate suitability matters: evaporative cooling shines in dry climates where outdoor air can provide meaningful cooling with modest water use, but is less effective in high-humidity zones without supplemental chillers.
• Water strategy is a grid asset: water reuse and storage enable better resilience against grid outages, with water-cooled components offering lower electricity but higher water management risk that must be mitigated by governance and robust maintenance.
• Infrastructure must consider seasonal variability: on the hottest days, evaporative cooling is often combined with adiabatic pre-cooling or mechanical chillers to sustain IT temperatures with acceptable energy bills.

Air cooling: simplicity, resilience, and the energy-cost baseline

Air cooling remains the baseline in many public data centers, prized for simplicity, lower water risk, and lower upfront complexity. As of late 2025, conventional air-cooled facilities average PUE values between 1.2 and 1.6 in temperate climates and 1.3 to 2.0 in more challenging heat-load regions, depending on containment, hot-aisle/cold-aisle configurations, and outside air management. In the 2024 EU AI Act context, standards emphasize predictable energy performance and easier retrofits, which can favor air cooling retrofits with modern containment and fan array optimizations. A 120 MW regional data center complex piloted hot-aisle containment and variable-speed fans, achieving a measured 0.25–0.40 reduction in energy per kW IT compared with legacy non-contained air cooling, and a 6–8°C average reduction in intake air temperature at peak load across the facility.

However, air cooling is highly climate-sensitive. In hot-humid environments, air cooling can require large fans and multiple rows of containment, pushing fan energy to 0.4–0.7 kW per kW IT during peak periods, which erodes the potential PUE reduction. In contrast, cooler climates can realize substantial gains with outdoor air pre-cooling and humidity control, achieving 1.05–1.15 PUE for mid-density deployments. A 2023–2025 survey of municipal data centers notes that air-cooled facilities in temperate regions with effective air-side economizers achieve 8–12% annual energy savings compared with basic air-cooled systems, while facilities in subtropical zones require a more nuanced approach to avoid compressor-load cycling and humidity penalties.

Economic and grid considerations shape decisions. Air cooling technologies offer modest capital expenditure, with typical retrofits involving cold-aisle containment kits ranging from $800 to $2,500 per rack, plus $200–$400 per kW of IT cooling capacity for fan optimizations. In contrast to liquid cooling, air systems demand less specialized fluids, but they can incur higher electricity costs during peak periods when large-volume air movement is necessary. A 10-site public campus network reported a 9–12% difference in annual energy cost between enhanced air containment with ASC (adapative system control) and baseline non-contained air cooling, with peak-day grid demand reductions of 2–3 MW during heat events due to improved efficiency and load shifting.

Key takeaways

• Air cooling remains field-ready and cost-competitive in temperate climates, particularly with containment and advanced control strategies that reduce fan energy and enable smarter intake management.
• Performance is climate-driven: in hot, humid regions, air cooling struggles to maintain reliability without higher energy penalties or hybridization with other strategies.
• Retrofit decisions tend to favor air cooling when capital budgets are constrained or when water scarcity makes liquid or evaporative options less viable.

Hybrid and hybridized systems: where the grid seeks flexibility

Hybrid cooling architectures—combining elements of liquid, evaporative, and air cooling—are increasingly common in public data centers that must operate under variable climate conditions and grid signals. As of late 2025, several municipal and regional data centers deploy hybrid schemes that switch cooling modalities by zone or by time-of-day to meet peak IT loads with minimal grid strain. For example, a coastal megacenter uses air cooling for cooler seasons and transitions to evaporative precooling plus liquid-assisted top-down cooling during heatwaves, achieving a measured 18–25% reduction in annual cooling energy relative to a single-method baseline. A desert-adjacent campus applies evaporative precooling during the day and liquid immersion or direct-to-chip cooling during the evening, reducing peak power draw by 12–20% and delivering a more stable IT inlet temperature.

These configurations underscore a critical grid-utility dimension: the ability to modulate load and share heat rejection across the cooling stack can align data center demand with renewable generation and demand-response programs. In 2025 pilot programs, hybrid facilities achieved 0.05–0.20 points better PUE across the cooling season compared with single-method installations, and demonstrated improved resilience during grid contingencies by reducing compressor start frequency by 15–40% in peak days. A facility-level analysis indicates that the capacity to shift 3–6% of IT cooling load into low-carbon cooling modes during market price spikes yields tangible bill reductions for public operators and helps stabilize regional transmission constraints during heat events.

Key takeaways

• Hybrid systems provide resilience in variable climates, enabling facilities to leverage the strengths of each method while mitigating weaknesses tied to humidity, water supply, and energy prices.
• Grid-aligned control strategies, including predictive analytics and demand response participation, are essential to realize the full value of hybrid cooling.
• Capital spend is context-dependent: hybrids can be more expensive upfront but often deliver better long-term energy cost reductions and reliability margins.

Policy, grid realities, and long-term planning

Policy and grid considerations increasingly shape cooling choices for public data centers. The 2025 NFPA 70 (National Electrical Code) updates emphasize safer coordination between cooling equipment and power distribution, while the 2025 NFPA 850 updates highlight isolation and redundancy requirements for critical facilities that must maintain cooling during outages. In Europe, the 2024–2025 EU AI Act and related energy efficiency directives push for disclosed PUE metrics, measurable heat reuse potential, and robust lifecycle assessments for cooling systems. These policy landscapes incentivize data centers to publish transparent energy-use reporting and to pursue cooling strategies with demonstrable grid benefits, such as load shifting, heat recovery for district heating networks, and reduced peak electricity demand during heat waves.

From a grid perspective, the choice of cooling strategy influences the peak capacity necessary to support public services, emergency communications, and government data operations. For example, a 150 MW municipal data hub that adopted liquid cooling reported peak-demand reductions of 5–7% during heat events due to lower chiller and fan requirements, while simultaneously enabling heat reclaim for district heating networks—an alignment that improves municipal energy self-sufficiency in some seasons. Conversely, evaporative cooling strategies, when scaled across many facilities in a region, can introduce water-system considerations that must be integrated with city water management plans and drought risk mitigation. In humid climates, policy-driven targets to minimize water use push operators toward closed-loop evaporative systems and enhanced water treatment processes, which carry additional capital and operating expenses but reduce environmental risk exposure for public operators.

Economic and risk considerations are interwoven. Public data centers face capital allocation challenges, including balancing short-term budget cycles with long-term grid resilience objectives. A number of 2024–2025 budget cycles show that capital investments in containment, monitoring, and predictive maintenance yield better asset utilization and lower risk-adjusted returns, even when initial cost exceeds a baseline air-cooled approach by 20–40% for a given campus. The payback horizon often depends on climate, workload mix, and the facility’s ability to participate in demand-response programs or heat-recovery partnerships with public utilities.

Key takeaways

• Policy drives transparency and resilience: regulations and energy reporting requirements push operators toward verifiable performance metrics and robust cooling architectures that support grid stability.
• Grid and water considerations must be jointly managed: hybrid and liquid-based systems offer grid benefits, while evaporative systems demand careful water-resource planning and drought risk mitigation.
• Forecasting and lifecycle planning become essential: equipment selection should integrate climate projections, workload growth, and the potential for heat reuse or demand-response participation.

Conclusion: aligning cooling strategies with climate, grid, and policy realities

The cooling decision for public data centers is no longer a pure technical choice; it is a multi-criteria planning problem that must account for climate, grid dynamics, water and energy policy, and long-term resilience. Liquid cooling offers precision and density that can meaningfully bend a data center’s energy profile and reduce peak grid demand, particularly in high-density deployments and micro-regions with favorable heat rejection options. Evaporative cooling delivers strong efficiency gains in dry climates and can meaningfully decrease chiller electricity use when water resources are secure and well managed. Air cooling remains a reliable baseline, especially in temperate regions with effective containment and control systems, but its performance is sensitive to ambient conditions and may necessitate hybridization in warmer zones. Hybrid architectures, and their associated control systems, emerge as a pragmatic path for public operators seeking both reliability and grid-aligned flexibility amid shifting weather patterns and evolving policy mandates.

As of late 2025, the trajectory is toward more nuanced, climate-aware cooling strategies that can participate meaningfully in energy markets and municipal resilience plans. The best choice is no longer a single approach but a portfolio—matched to climate, water resources, and grid opportunities—with adaptable controls, robust monitoring, and governance that ensures safe operation during outages, while making heat reuse and demand response a core feature rather than an afterthought. For public operators, the payoff is not only lower energy costs but a more resilient, transparent, and policy-compliant data center footprint that can contribute to a city’s broader energy and climate objectives.