Researchers have proposed a zeolite-based “thermal battery” system that could cut the electricity data centers spend on cooling by up to 86%, according to a preprint published on ChemRxiv. The finding arrives as the U.S. Department of Energy channels tens of millions of dollars into next-generation cooling solutions and as artificial intelligence workloads push power demand to new highs. If the concept scales beyond the lab, it could reshape how operators manage one of the fastest-growing drains on the national grid.
Why Cooling Dominates Data Center Power Bills
Servers generate enormous heat, and removing that heat is expensive. Cooling can represent up to roughly 40% of a data center’s total energy usage, according to the U.S. Department of Energy. That share has grown alongside the rise of GPU-dense AI training clusters, which pack far more thermal output per rack than traditional compute hardware. For a large campus drawing 100 megawatts, the cooling load alone can rival the electricity consumption of a small city, especially in hot climates where mechanical chillers must run nearly year-round to maintain narrow temperature and humidity bands.
Data centers already account for a notable share of total U.S. electricity use, and every new hyperscale facility deepens the strain on local grids during peak afternoon hours when air conditioning demand is highest. Utilities in northern Virginia, central Texas, and the Pacific Northwest have flagged interconnection queues stretching years into the future, partly because cooling infrastructure inflates each facility’s nameplate power request well beyond what the servers themselves need. Any technology that materially shrinks that gap would free grid capacity for additional compute or for other customers waiting in line, while also reducing the need for costly grid upgrades that can delay new capacity by several years.
How Zeolite Thermal Batteries Achieve an 86% Reduction
The core idea, detailed in a ChemRxiv preprint (DOI: 10.26434/chemrxiv-2026-28wv2), relies on zeolite minerals that absorb and release large amounts of thermal energy through a reversible adsorption process. Zeolites can be “charged” at a remote location using waste heat or cheap off-peak electricity, then transported to a data center where they absorb heat from the facility’s cooling loop without consuming on-site power for compressors or chillers. The researchers modeled this cycle and found it could reduce cooling-related electricity consumption by up to 86%, a figure that accounts for the full round-trip energy cost of the process and assumes modern heat exchangers and control systems to optimize the charging and discharging cycles.
A key practical question is whether trucking tons of zeolite back and forth erases the efficiency gains. According to an NYU engineering analysis published in early March 2026, the system’s energy accounting includes the fuel required to haul material using modern electric trucks. Even with that transportation penalty factored in, the net savings remain substantial. The concept is still at the preprint stage and has not yet undergone formal peer review, so the 86% figure should be treated as a modeled projection rather than a field-tested result. Still, the idea highlights an alternative path: instead of building ever-larger on-site cooling plants, operators could import stored “cold” the way they import fuel, potentially siting the charging facilities where surplus renewable power or industrial waste heat would otherwise go unused.
Federal Funding and Competing Thermal Storage Approaches
Washington is already betting that thermal storage will play a role in taming data center energy growth. The DOE announced $40 million specifically for more efficient data center cooling, channeled through competitive awards that encourage both incremental efficiency gains and more radical redesigns of how facilities manage heat. Program details and technical reports are cataloged through resources such as the DOE scientific information portal, which aggregates research outputs, and the agency’s infrastructure funding exchange, which tracks large-scale deployment and demonstration projects across the country. Together, these platforms show a pipeline that stretches from basic materials science to full-scale pilot plants attached to live data centers.
The DOE’s advanced research arm also plays a central role. The ARPA-E program office has supported several cooling-focused research tracks, including its COOLERCHIPS initiative, which targets chip-level and facility-level thermal management breakthroughs. While zeolite-based thermal batteries are not yet listed among ARPA-E’s flagship projects, the same logic underpins much of its portfolio: shifting when and where cooling energy is consumed so that data centers draw less power during the most expensive and carbon-intensive hours. In that context, the zeolite proposal joins a broader ecosystem of ideas, from advanced immersion cooling to district-scale heat reuse, all competing for a share of federal dollars and industry attention.
Underground and On-Site Thermal Storage Alternatives
Zeolite batteries are not the only thermal storage architecture under investigation. The National Renewable Energy Laboratory is studying a concept called Cold Underground Thermal Energy Storage, which buries chilled water or ice slurry beneath a facility and draws on it during peak demand periods, as described in NREL’s research overview. In that design, large insulated tanks or engineered aquifers are cooled when electricity is abundant and inexpensive, then discharged hours or days later to offset chiller operation. Because the storage is stationary and co-located with the data center, there is no need to move material by truck, but the system requires substantial excavation, permitting, and integration with local groundwater or soil conditions.
Underground systems avoid the logistics of a zeolite approach but come with their own constraints. They are best suited to campuses with access to land and favorable geology, and they may be difficult to retrofit at dense urban sites where data centers share space with other critical infrastructure. The two concepts address the same bottleneck from different angles: shifting cooling energy use away from the hours when the grid is most constrained, either by storing cold underground or by importing it on wheels. In practice, operators could eventually combine approaches, using on-site storage for daily load shifting and transportable media like zeolites to handle seasonal peaks or temporary capacity shortfalls during grid upgrades.
Lessons from Ice-Based Thermal Storage in Buildings
While zeolite batteries remain a laboratory concept, a related technology is already generating measurable savings in commercial buildings. Ice-based thermal storage systems freeze water overnight when electricity is cheaper, then melt it during the day to cool a building. One hospital that installed such a system saw its energy costs drop by $278,000 in the first year of operation, according to the Associated Press, with the system projected to save nearly $4 million over its lifetime. Those savings arise not only from reduced total kilowatt-hours but also from lower demand charges, since the hospital can cap its peak power draw even during heat waves.
Those numbers come from a single healthcare facility, not a data center, so direct comparisons require caution. Hospitals have variable occupancy and seasonal cooling loads, while data centers run at near-constant thermal output around the clock. That constant baseline actually makes thermal storage more predictable to size and schedule for a data center, because operators can model hourly thermal demand and align charging windows with off-peak grid conditions or on-site solar production. The hospital experience nonetheless offers a proof of principle: when cooling is shifted in time rather than reduced outright, the financial and grid benefits can be substantial, especially in markets where utilities heavily penalize peak usage.
What Needs to Happen Before Zeolite Batteries Reach Data Centers
Translating the zeolite concept from preprint to production will require progress on several fronts. Materials scientists must validate that candidate zeolites can withstand thousands of charge–discharge cycles without losing capacity or breaking down mechanically, and engineers will have to design containers and heat exchangers that can move large amounts of heat in and out of the mineral beds quickly. Safety and reliability standards will also matter: data center operators are notoriously risk-averse, and any new cooling technology must match or exceed the uptime and fault tolerance of conventional chillers before it will see wide adoption. Demonstration projects attached to smaller edge facilities or research campuses could provide the performance data needed to build confidence.
Economics may prove just as decisive as physics. The modeled 86% reduction in cooling electricity is compelling, but investors will look closely at capital costs, trucking logistics, and integration with existing mechanical rooms. If zeolite systems can be deployed as modular add-ons that complement, rather than replace, conventional chillers, they may find an early niche helping operators navigate grid constraints or meet corporate decarbonization targets without waiting for new transmission lines. In the longer term, as DOE-backed research on thermal storage, underground cooling, and chip-level heat management matures, data center cooling could shift from a fixed overhead into a flexible, dispatchable resource, one that can absorb surplus renewable energy, relieve stressed substations, and make room on the grid for the next wave of AI compute.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
