Thermal Batteries for AI Data Centers

How Next-Generation Cooling Could Cut Electricity Use by Up to 86%

**Prof. Aecio D’Silva, Ph.D.

Keywords: thermal batteries, AI data center cooling, data center energy efficiency, zeolite thermal battery, aquifer thermal energy storage, phase change materials, thermal energy storage, liquid cooling, sustainable data centers, cooling electricity reduction

Executive Summary

AI data centers are entering a new thermal-design regime in which rack-level heat flux, coolant supply temperature, chiller lift, and grid-interconnection constraints increasingly determine compute scalability. As accelerator-dense racks approach and exceed 100 kW, thermal management must be treated as a coupled energy-storage, heat-rejection, and workload-orchestration problem rather than a conventional HVAC load. Thermal batteries—thermal energy storage systems that charge by storing sensible, latent, or sorption potential and discharge by absorbing server heat—can shift cooling load from peak grid hours, reduce compressor runtime, improve power usage effectiveness (PUE), and provide short- to long-duration thermal ride-through. Recent zeolite-based sorption models suggest cooling-electricity reductions of up to 86% for the data center cooling subsystem under specified benchmark assumptions, while aquifer thermal energy storage (ATES), borehole thermal energy storage (BTES), ice storage, chilled-water tanks, and phase-change materials (PCMs) offer different tradeoffs in storage duration, round-trip efficiency, water use, site constraints, and dispatchability.

• Lower cooling electricity use: Thermal storage can reduce chiller runtime and, in emerging zeolite systems, may cut cooling power consumption dramatically.
• Reduced peak demand: Stored cooling can be discharged during high-load periods, helping operators avoid expensive peak electricity charges.
• Improved grid flexibility: Data centers can shift cooling loads to hours when renewable power is cheaper, cleaner, or more abundant.
• Greater resilience: Thermal buffers give operators more time and flexibility during workload spikes, grid constraints, or cooling system stress.
• Better sustainability profile: By lowering electricity demand and supporting renewable integration, thermal batteries can help reduce the carbon intensity of AI infrastructure.

Why AI Data Center Cooling Has Become a Grid-Level Challenge

The rapid growth of AI training and inference has changed the boundary conditions for data center thermal engineering. Compared with conventional cloud halls, AI clusters concentrate higher heat loads into smaller footprints, increase coolant-flow requirements, and reduce the margin for transient thermal excursions. At the rack level, 100 kW-class deployments require careful management of heat transfer from GPUs, accelerators, memory, power delivery components, and coolant distribution units (CDUs). At the campus level, the cooling plant must coordinate chillers, dry coolers, cooling towers, economizers, pumps, heat exchangers, thermal storage, and controls against variable IT load and utility demand charges.

Cooling energy can represent a significant fraction of total facility electricity use, with actual performance governed by climate, supply-water temperature, approach temperatures, fan and pump power, containment effectiveness, and the coefficient of performance (COP) of mechanical refrigeration. The engineering objective is not simply to reduce chiller nameplate capacity; it is to reduce annual kWh, peak kW, water withdrawal and consumption, and thermal risk while maintaining allowable inlet conditions for IT equipment. This makes metrics such as PUE, water usage effectiveness (WUE), cooling load factor, thermal storage capacity, discharge duration, and control-response time central to AI campus design.

What Are Thermal Batteries?

A thermal battery stores exergy in a thermal form and releases it through a controlled heat-transfer process. In data center applications, this may mean storing chilled water, ice, solid-liquid latent heat, subsurface cold, or chemically bound sorption potential. The storage system is charged when grid electricity is inexpensive, renewable generation is abundant, ambient wet-bulb temperature is favorable, or waste heat is available. It is discharged when IT load peaks, electricity prices spike, grid capacity is constrained, or the mechanical cooling plant needs support.

Unlike electrochemical storage, which stores electrical energy and later converts it back to electricity, thermal storage bypasses unnecessary conversion steps when the end use is cooling. This can improve system economics because the stored output directly offsets compressor work, pump and fan energy, or heat-rejection load. The design challenge is to match storage temperature, discharge rate, pressure drop, heat-exchanger approach temperature, and control logic to the thermal profile of air-cooled, direct-to-chip, rear-door heat exchanger, or immersion-cooled IT systems.

Three Thermal Battery Technologies Scaling for AI Data Centers

1. Aquifer and Borehole Thermal Energy Storage

Aquifer thermal energy storage (ATES) and borehole thermal energy storage (BTES) use the subsurface as a seasonal or multi-week thermal reservoir. In an ATES configuration, operators typically inject and extract groundwater through warm and cold wells, using the aquifer’s heat capacity and stable underground temperature to store cooling potential. BTES systems use vertical borehole heat exchangers embedded in the ground; they avoid direct groundwater extraction but usually require larger bore fields and careful thermal-response modeling.

For hyperscale AI campuses, subsurface storage is most attractive when the site has favorable hydrogeology, predictable seasonal cooling demand, and enough land for wells, piping, and heat-exchange infrastructure. Engineering feasibility depends on aquifer permeability, allowable groundwater temperature change, well spacing, thermal breakthrough risk, water chemistry, permitting, and long-term monitoring. When properly designed, ATES or BTES can reduce peak chiller load, extend economizer operation, and lower evaporative water losses, but they require early integration into site selection and civil design.

2. Zeolite and Solid-State Sorption Thermal Batteries

Zeolite thermal batteries operate through a reversible adsorption-desorption cycle. Zeolites are microporous aluminosilicate materials with high internal surface area and strong affinity for water vapor. During charging, low- to medium-temperature heat—potentially below about 200 °C from industrial waste streams—drives desorption and dries the zeolite. During discharge, water is reintroduced or evaporated using heat rejected from the data center; adsorption into the dry zeolite provides an effective heat sink and can reduce the need for vapor-compression cooling.

The technical appeal is that the charged material can retain sorption potential without the standby thermal losses associated with conventional sensible-heat storage. System design must still account for adsorption kinetics, mass-transfer resistance, vapor management, condenser integration, water recovery, containerized transport, material cycling stability, and logistics between the waste-heat source and the data center. NYU Tandon modeling indicates that a zeolite-based configuration could reduce cooling electricity by up to 86% for the data center cooling load under the modeled benchmark, with system-level savings dependent on transport energy, local waste-heat availability, and water impacts.

3. Phase-Change Materials and Ice Storage

Phase-change materials, ice storage, chilled-water tanks, and ice slurries provide shorter-duration storage by exploiting latent heat or sensible cooling capacity. Ice storage charges at night or during low-price periods by freezing water and discharges by melting ice to absorb heat from the chilled-water loop. PCMs are selected by melting temperature, latent heat of fusion, thermal conductivity, encapsulation method, cycling durability, flammability profile, and compatibility with heat exchangers and coolant loops.

These systems are comparatively modular and can be integrated with existing chilled-water plants, CDUs, rear-door heat exchangers, and direct-to-chip liquid-cooling loops. Their main value is intraday peak shaving and transient-load buffering rather than seasonal storage. Key design parameters include storage ton-hours, discharge temperature, heat-exchanger approach, pump parasitic load, freeze-thaw cycling efficiency, and controls that coordinate the storage state of charge with IT workload forecasts and utility price signals.

Why 2026 Is a Breakout Year for Thermal Energy Storage

Thermal batteries are moving from niche engineering concept to strategic infrastructure because AI campuses are increasingly designed as integrated cyber-physical energy systems. Compute scheduling, liquid-cooling architecture, on-site generation, battery energy storage, thermal storage, heat rejection, and grid interconnection must be optimized together. The practical objective is a dispatchable thermal plant that can respond to IT load, weather, electricity price, water availability, and grid signals in near real time.

Three engineering trends are accelerating adoption. First, variable renewable generation creates low-marginal-cost charging windows that can be converted into stored cooling. Second, direct-to-chip and other liquid-cooling systems raise coolant return temperatures, improving the usefulness of waste heat and reducing chiller lift. Third, utilities increasingly value controllable demand response, making cooling load shifting financially relevant through demand-charge reduction, capacity deferral, and improved grid flexibility.

Where the First Large-Scale Pilots Are Likely to Matter Most

The most technically meaningful pilots are likely to appear where high AI load growth intersects with constrained grid capacity, favorable thermal-storage resources, and clear economic signals. ATES and BTES pilots require subsurface characterization, thermal plume modeling, permitting, and monitoring infrastructure. Zeolite pilots require nearby waste-heat sources, material-handling logistics, vapor-cycle integration, and validation against a conventional compression-chiller baseline. PCM and ice-storage pilots are easier to retrofit but must prove controls integration and lifecycle economics at high utilization.

These systems will not eliminate chillers or heat rejection equipment. Instead, they can reduce equivalent full-load chiller hours, lower coincident peak demand, increase free-cooling utilization, and create a thermal buffer between AI workload excursions and grid or plant constraints. The most successful deployments will likely combine predictive controls, digital twins, weather forecasts, utility tariffs, and IT workload scheduling so that thermal storage is dispatched as an operational asset rather than treated as passive infrastructure.

Conclusion

Thermal batteries are not merely an efficiency upgrade; they are a design pathway for decoupling AI compute growth from peak cooling electricity demand. Zeolite sorption systems, ATES, BTES, ice storage, chilled-water tanks, and PCMs each occupy a different point in the design space for temperature level, storage duration, deployment complexity, water impact, and control flexibility. Their greatest value emerges when power, cooling, heat reuse, storage, and workload management are engineered as one integrated system. Operators that adopt this systems-level approach can reduce cooling kWh, lower peak kW, improve resilience, and increase the siting flexibility of large AI campuses while maintaining tighter thermal control over increasingly power-dense IT equipment.

References

1. D’Silva, A. (2026). Sodium BESS: Scalable Energy for AI Factories. Moura Enterprises Labs. US. https://algaeforbiofuels.com/sodium-bess-scalable-energy-for-ai-factories/ 
2. D’Silva, A. (2026). Artificial Intelligence and Total Excellence Management. Moura Enterprises Labs. US. https://algaeforbiofuels.com/artificial-intelligence-and-total-excellence-management/
3. NYU Tandon School of Engineering. “New Method of Data Center Cooling Could Dramatically Decrease Electricity Use.” March 2026.
4. Tech Xplore. “Zeolite ‘thermal batteries’ could cut data center cooling power up to 86%.” March 2026.
5. University of Illinois Urbana-Champaign News Bureau. “Team looking to tap underground ‘thermal batteries’ to cool AI data centers and save water.” June 2026.
6. National Renewable Energy Laboratory. “Reducing Data Center Peak Cooling Demand and Energy Costs With Underground Thermal Energy Storage.” January 2025.
7. NSF Public Access Repository. “Thermal Time Shifting: Decreasing Data Center Cooling Costs with Phase-Change Materials.”

In Belo Jardim, student of the Grupo Escolar Bento Américo and Prof. Donino Gymnasium, and student of the teachers: Dulce Ramos, Alba Leite, Dona Conceição Moura, Dona Olindina Mergulhão, Estefânia Moura Bezerra, and Maria Luiza