Thermal energy storage (TES) tanks help data center cooling plants store chilled water for use when cooling demand, equipment availability, or operating conditions change. But tank volume alone does not determine how much cooling a system can deliver. Thermal stratification, diffuser performance, usable temperature difference, and the surrounding pumping and control systems all affect performance. For engineers specifying custom TES tanks, those thermal requirements must also translate into practical decisions about materials, construction codes, internal components, and fabrication.
IN BRIEF
A TES tank stores cooling; chillers make it. At a 16°F (8.9 K) usable temperature difference, 1,000,000 gal (3,785 m³) of chilled water holds about 11,120 ton-hours (39.1 MWh thermal) in theory.
Stratification sets usable capacity. In U.S. Army research, a 0.9°F (0.5 K) rise below the thermocline on an 18°F (10 K) design difference cost 5% of capacity.
Diffuser limits are geometry-specific. Research ties mixing to inlet Froude and Reynolds numbers, but the diffuser design basis belongs to the system designer.
A full tank is not backup cooling on its own. Stored water can support cooling during a chiller restart only when the pumps, controls, and other equipment required to deliver cooling remain powered.
The code follows the design conditions. Atmospheric tanks are commonly built to AWWA D100 or API 650. Whether a pressurized vessel needs ASME Section VIII construction depends on Code scope, jurisdiction and owner requirements.
On this page: how TES works · stored energy vs. deliverable cooling · stratification and diffusers · TES vs. buffer vs. ride-through · design, materials and codes · plant integration · capacity limits · common mistakes · RFQ checklist · Harris support · FAQs
How a chilled-water TES tank works
A chilled-water TES tank stores cooling capacity that chillers have already produced. In a stratified tank, cold and warm water share one vessel. During charging, chilled water enters near the bottom while an equal flow of warm water leaves near the top. During discharge, cold water is drawn from the bottom to serve the load and warm return water enters at the top. The tank stays full; only the cold share of its volume changes.
Chilled-water storage is sensible storage, so its capacity comes entirely from the water's temperature change. The Department of Energy's Federal Energy Management Program (FEMP) notes in its Federal Technology Alert on cool storage that this takes more volume than ice or eutectic-salt storage. In return, standard water chillers can charge the tank without secondary coolants or significant capacity loss, typically cooling water to 39–44°F (4–7°C). Ice storage requires substantially less volume, but the lower evaporator temperatures needed for ice production can reduce chiller capacity and efficiency, depending on the equipment and operating conditions.
In a data center, storage can shift chiller operation in time, add discharge capacity beyond chiller output, carry load during chiller maintenance or bridge a chiller restart. Each duty sizes and controls the tank differently.
Stored energy versus deliverable cooling
Stored energy follows the sensible-heat relationship:
Q = m × Cp × ΔT
Q is the heat the stored water can absorb (Btu), m is the water's mass (lb), Cp is its specific heat (about 1.0 Btu/lb·°F) and ΔT is the usable difference between supply and return temperatures (°F). Table 1 applies the relationship to an illustrative 1,000,000 gal tank.
| Step | Value | Basis |
|---|---|---|
| Water volume | 1,000,000 gal (3,785 m³) | Assumed for illustration |
| Water mass | 8,340,000 lb (3.78 million kg) | 8.34 lb/gal |
| Usable ΔT | 16°F (8.9 K) | For example, 40°F (4.4°C) supply and 56°F (13.3°C) return |
| Stored energy | 133.4 million Btu | 8,340,000 lb × 1.0 Btu/lb·°F × 16°F |
| Ton-hours | 11,120 ton-hours | 133.4 million Btu ÷ 12,000 Btu per ton-hour |
| Thermal energy | 39.1 MWh thermal | 11,120 ton-hours × 3.517 kWh per ton-hour |
| Discharge rate | 2,000 tons (7.03 MW thermal) | 3,000 gpm (189 L/s) × 16°F ÷ 24 |
| Time to deplete at that rate | About 5.6 hours | 11,120 ton-hours ÷ 2,000 tons |
Table 1. Theoretical stored energy before losses. In SI units, 3,785,000 kg × 4.19 kJ/kg·K × 8.9 K is about 141 GJ, or 39 MWh.
These are theoretical figures, not guaranteed delivered capacity. They assume every gallon swings through the full 16°F, with no mixing, no heat gain and no volume stranded beyond the diffuser openings. The tons figure is instantaneous cooling power, limited separately by the flow the diffusers can pass without mixing, by pump capacity and by the ΔT the load actually returns. Nor is 7 MW of cooling the same as 7 MW of avoided electrical demand; that depends on chiller plant efficiency and on the pumps still running.
No universal sizing rule applies. FEMP notes that discharge temperature rises toward the end of every cycle, keeping useful capacity below the theoretical maximum, so volume follows from an hourly load profile, the operating strategy and the designer's loss allowances.
Stratification: why usable capacity is smaller than tank volume
Stratification works because colder water is denser across the normal chilled-water range, so it settles beneath warm return water. The rule reverses near 39°F (4°C), where water reaches its maximum density. Scale-model tests by Wildin and Truman, published in ASHRAE Transactions, found that inlet water below 39.2°F (4°C) would need a physical barrier to keep the tank stratified.
Between the layers sits the thermocline, a band of steep temperature gradient that rises during charging and falls during discharge. A U.S. Army Construction Engineering Research Laboratories study describes it moving through the tank like a plug. With little mixing, conduction alone thickened it from about 1 ft to 3 ft (0.3 to 1 m) over a six-hour charge. Water inside the thermocline, and water beyond the diffuser openings, never swings through the full ΔT.
The same study shows what mixing costs. Warming the water below the thermocline by 0.9°F (0.5 K), in a tank with an 18°F (10 K) design difference, removes 5% of the cooling capacity. The tank still looks full; the loss appears as a narrower ΔT.
How diffusers keep the layers apart
Diffusers introduce and withdraw water at each end of the tank. They spread flow across a long opening at low velocity, so buoyancy rather than momentum carries entering water across the tank as a thin gravity current. The Army study puts the density difference driving that current at only about 0.1%, which is why modest momentum causes mixing.
Two dimensionless groups, both based on flow per unit length of diffuser opening, describe the inlet. The densimetric Froude number compares inertia with buoyancy, and in the scale-model tests the Army study cites, keeping it at about 2 or less reliably formed a gravity current. The Reynolds number compares inertia with viscous forces and dominates mixing once the thermocline exists. The Army study placed its upper limit for optimal performance at 400 to 600 for the rectangular scale-model tank it tested.
These are experimental findings tied to the geometry and flow conditions tested, and they do not transfer directly between tanks. Penn State researchers tested a 44.5 ft (13.6 m) deep tank of roughly 1.4 million gal (5,300 m³) with radial diffusers. Its half-cycle figure of merit exceeded 90% in controlled constant-flow tests, even though inlet Reynolds numbers ran up to five times literature recommendations. In controlled-flow tests of four full-scale tanks with radial parallel-plate or octagonal slotted-pipe diffusers, less than 4% of theoretical capacity was lost at flows up to design. Geometry, diffuser type and flow all change the answer, so diffuser hydraulics belong to the system designer or a qualified specialist.
Stratification also degrades in operation. Flow above the diffuser design, uneven distribution along a diffuser and out-of-level openings all add mixing. The Army study adds that inlet temperature should not rise during charging or fall during discharge. Fabrication details count as well. Research summarized in the Army study found slot-shaped openings mixed less than rows of circular holes, which create multiple jets, and Wildin and Truman found that flow approaching a radial diffuser should be axially symmetric.
TES tank, buffer tank, or ride-through storage?
A buffer tank adds volume and thermal mass so plant controls stay stable and equipment does not short-cycle; it is not designed to hold a defined inventory of stored cooling. A TES tank is designed for deliberate charge and discharge against a load profile. Ride-through storage holds enough water at the required supply temperature to carry the load until mechanical cooling recovers.
| Characteristic | Chilled-water TES tank | Chilled-water buffer tank | Ride-through storage |
|---|---|---|---|
| Primary purpose | Shift chilled-water production in time; add discharge capacity | Stabilize plant control; prevent short-cycling | Carry the load until mechanical cooling recovers |
| Operating strategy | Scheduled charge and discharge cycles | Always in the loop; no deliberate cycling | Held charged; discharged only during an event |
| Capacity basis | Usable ton-hours over a load profile, after losses | System volume and thermal mass | Load × recovery time, at supply temperature |
| Flow and temperature | Design charge and discharge flows at a defined usable ΔT | Loop flow; temperature swings damped by mixing | Full load flow at or below the required supply temperature |
| Controls | State-of-charge tracking; mode sequencing | Minimal; the plant sees a larger loop volume | Automatic transfer; reserve protected from routine use |
| Stratification | Required; diffusers designed for it | Usually not designed for it | Needed if the outlet must stay near supply temperature |
| Chillers and pumps | Dedicated charge and discharge valving and pumping | Piped into the loop to add volume | Pumps, controls and other delivery equipment kept powered, as the architecture requires |
| Reliability | May be depleted when an outage occurs unless a reserve is held | Adds stability, not stored capacity | Only as reliable as power to the equipment that moves water |
Table 2. Duties overlap in practice; the design basis should state which duties a given tank serves.
One tank can serve several duties only if the design says so. A tank discharged daily for load shifting may be nearly empty when a power event arrives unless the controls protect a reserve.
A full tank does not keep IT equipment cool by itself
Stored chilled water becomes cooling only when something moves it. Under Uptime Institute's Tier Standard, Tier IV is the only Tier that requires Continuous Cooling, and Uptime separately recommends it above 4 kW per rack regardless of Tier. Its guidance on close-coupled cooling shows that the electrical scope depends on the architecture. Where close-coupled units reject heat to a chilled-water system, central-plant thermal storage can meet the mechanical part of that requirement, while fans, pumps, CDUs and in-room chillers or heat exchangers need uninterruptible power. Units that reject heat through refrigerant condensers to outside air need uninterrupted power to every component instead.
The required storage is set by how long mechanical cooling takes to return. One manufacturer's rapid-restart sequence for its air-cooled chillers assumes the chiller controller is on UPS power. It shows compressors starting about a minute after power returns and output passing 80% of the prior load at about 160 seconds; without the feature, the same bulletin reports about 15 minutes to reach 80%. Other machines, generator transfer and a later retransfer to utility power can lengthen or repeat the gap. Liquid cooling tightens it further: as Harris Thermal's article on AI data center liquid cooling explains, cold plates depend on continuous coolant circulation.
Mechanical design, materials, and codes
Tank configuration drives the mechanical design, the code basis and how much work a shop can do.
Atmospheric tank or pressurized vessel
Large stratified TES tanks are generally atmospheric: vented, vertical and cylindrical, and field-erected once they exceed shipping size. They offer the lowest cost per gallon, but the free water surface becomes part of the hydraulic system. Connecting an open tank to a closed loop requires the designer to decide how the loop is held at pressure, for example with a heat exchanger or a pressure-control arrangement. That choice changes nozzles and elevations. Open tanks also admit oxygen and need level, vent and overflow provisions.
A pressurized vessel connects directly at system pressure, but wall thickness and cost per gallon rise with diameter and pressure, which tends to suit smaller buffer and ride-through volumes. In a horizontal vessel, a thermocline of a given thickness also occupies more of the shallow water depth.
Which codes and standards apply
The design conditions and the project specification decide the code:
AWWA D100 sets minimum requirements for new welded carbon steel tanks storing water at atmospheric pressure. Apart from aluminum dome roofs, it does not cover other materials.
API 650 is written for oil storage but extends to other liquid products. It covers vertical, cylindrical, aboveground welded tanks at internal pressures approximating atmospheric, with uniformly supported bottoms and design temperatures up to 200°F (93°C). Annexes cover small internal pressure, seismic loading, stainless steel and complete shop assembly up to 20 ft (6 m) in diameter, each applying only when the purchaser specifies it.
ASME Section VIII, Division 1 governs pressure vessels within its scope. That scope excludes several vessel classes, including low-pressure vessels and, within stated limits, some vessels containing water under pressure. ASME considered removing several of these exclusions for its 2025 edition and deferred the change, so the edition in force matters. A pressurized chilled-water vessel does not automatically require ASME Section VIII construction. Applicable Code scope, jurisdictional rules, and owner specifications determine the requirements; voluntary Code construction and certification must also satisfy the applicable provisions.
Before fixing the design basis, confirm each document's current edition and scope, the jurisdiction's pressure vessel rules and the owner's requirements. Wind, seismic and foundation design follow the governing building code and the standard selected.
Materials, insulation, and condensation
Coated carbon steel with a corrosion allowance is common for water storage, and stainless steel removes most internal coating maintenance at higher material cost. Selection should follow the water, because a closed, treated loop and an open tank exposed to air carry different corrosion and biological risks. FEMP notes that storage water treatment resembles a non-storage system's, applied to a much larger volume.
External insulation needs a continuous vapor retarder. A chilled-water shell runs below the ambient dew point, so any breach condenses water on the steel. Inspection guidance flags insulated carbon steel between 10°F and 350°F (–12°C and 177°C) for corrosion under insulation. Nozzles, supports, ladder clips and anchor chairs are the usual breach points.
Fabrication details that drive cost and schedule
Most fabrication complexity sits in the internals and penetrations:
Diffuser headers, risers and supports, with elevation and levelness tolerances on the openings
Inlet and outlet nozzles sized for diffuser flow and reinforced for piping loads
A vertical array of temperature connections to track the thermocline and state of charge
Manways for inspection and coating work, with gasket selection suited to cold, wet service
Vents sized for fill and draw rates, plus overflow, drains and vacuum protection where applicable
Hydrostatic load, thermal movement of internal piping and anchorage complete the mechanical design. Shipment planning usually decides between shop fabrication and field erection, because the route often sets the equipment envelope. Hydrostatic testing, examination records, material test reports and coating inspection belong in the purchase order.
How the tank fits a data center cooling plant
A chilled-water TES tank sits on the facility side of the cooling system. Depending on the hydraulic design, a TES tank may be integrated in parallel with the chillers or through another arrangement that allows controlled charging and discharging. It does not normally connect to a server liquid-cooling loop: coolant distribution units (CDUs) separate the technology cooling system from the facility water system, and the tank serves their facility side along with air-side equipment.
Storage temperature should be set against the supply temperature equipment actually needs. A colder store widens ΔT and raises capacity per gallon, but FEMP notes that chillers charging storage run at lower evaporator temperatures, which raises energy use unless cooler night conditions offset it.
The building management system tracks state of charge from the sensor array, sequences valves and pumps between modes, and holds any ride-through reserve that routine operation must not spend. Typical modes are overnight charging, discharge through tariff or load peaks, carrying load during chiller maintenance, and bridging a chiller restart.
Why a constant load changes the value case
FEMP's cool storage guidance lists conditions that favor storage, including large daily swings in energy charges, high or ratcheted demand charges, and an average cooling load well below the peak. A data center with a nearly constant IT load has a small peak-to-average ratio, so the classic case for smaller chillers plus storage is weaker.
The value case then rests on site-specific variables: time-varying prices, demand charges or demand-response programs in the tariff; added capacity during expansion or maintenance; and resilience. Night charging can also help chiller efficiency, because FEMP cites dry-bulb temperatures about 20°F (11 K) lower and wet-bulb about 5°F (2.8 K) lower at night across most of the U.S. Storage also adds pumping energy and standby heat gain, so the economics belong in a project-specific study, not a general payback figure.
What reduces usable capacity in operation
Most of the gap between nominal and usable capacity traces to a short list of mechanisms, and most are set during specification.
| Issue | Effect on usable capacity | Where it is controlled |
|---|---|---|
| Inlet mixing and short-circuiting | Warms the cold layer and thickens the thermocline | Diffuser design, flow limits, acceptance testing |
| Thermocline growth and dead volume | Water that never swings through the full ΔT | Tank height, diffuser elevations, cycle timing |
| Flow above design | Higher inlet Froude and Reynolds numbers, more mixing | Pump and valve limits, control sequences |
| Low return temperature | Smaller ΔT, so each gallon stores less | Load-side valve control, ΔT monitoring |
| Heat gain | Stored water warms through the shell, roof, bottom and piping | Insulation, vapor retarder, thermal breaks at supports |
| Sensor placement | Misreported state of charge drives wrong control decisions | Sensors through the full height, periodic calibration |
| Pumping and pressure drop | Separate charge and discharge flows add pump energy | Diffuser and piping sizing; a wider ΔT reduces flow |
| Water quality | Corrosion, fouling and biological growth | Treatment program, sampling points, coatings |
| Access and commissioning | Internals cannot be inspected; limits never verified | Manways and internal access; tests at design flow |
Table 3. Mechanisms that separate usable from nominal capacity.
Common mistakes in TES tank specification
Specifying gallons instead of usable ton-hours. Capacity should be stated at a supply temperature and a ΔT the load will reliably return.
Treating the tank as backup cooling. Unless the pumps, controls and other equipment that deliver cooling stay powered, the stored water stays in the tank.
Leaving diffuser responsibility unassigned. The designer owns hydraulics and performance; the fabricator builds the geometry to drawings and tolerances.
Assuming the code from the label. A pressurized vessel is not automatically a Code vessel, and an atmospheric tank still needs a defined standard.
Fixing the diameter before checking the route. Shipping clearances and site access can force field erection late in design.
RFQ checklist for a custom TES tank
A TES tank inquiry needs two kinds of information, usually from different people: what a fabricator needs to price and build the tank, and what defines system performance.
Needed to price fabrication
Project location, delivery site, route clearances, crane access and laydown space
Atmospheric or pressurized construction, and vertical or horizontal orientation
Nominal volume, preferred diameter and height, and any dimensional envelope
Design pressure and vacuum, design temperature, governing code or standard, and whether a U-stamp is required
Materials, corrosion allowance, water chemistry, and coating or lining systems
Nozzle schedule with sizes, ratings, orientations, elevations and piping loads
Internals to be fabricated, with drawings: diffuser headers, risers, supports and sensor connections
Insulation and vapor retarder scope, and the attachments it requires
Foundation interface, anchorage, and wind and seismic design data
Examination, testing and documentation requirements, with any witness points
Needed to define thermal performance
Usable capacity in ton-hours or MWh thermal, at a stated supply temperature
Design supply and return temperatures, and the minimum ΔT the load will reliably return
Charge and discharge flow rates, operating modes and duty cycles, including any ride-through reserve
The diffuser design basis and its owner: engineer of record, specialist or another party
Any performance guarantee, with its test method and acceptance criteria
The first list can be priced from drawings and specifications. The second is system design; naming its owner keeps a tank from being bought on gallons but judged on ton-hours.
How Harris Thermal supports thermal energy storage projects
Harris Thermal fabricates the vessel or tank portion of data center cooling systems to project-specific volume, design pressure, temperature and connection requirements. Pressure vessels and storage tanks are built to ASME Section VIII, Division 1 where the service requires it, or otherwise to the project specification, in carbon steel, stainless steel or higher alloys selected for the water chemistry. Nozzle orientation, supports and internals are built around the system the tank must fit. Coatings and linings are applied by qualified third-party applicators that Harris coordinates, and examination, testing and documentation follow the purchase specification. Harris also designs and fabricates custom shell-and-tube heat exchangers for facility-side heat transfer duties.
Storage capacity, operating strategy, diffuser hydraulics and stratification performance remain with the project's system designer. Pumps, valves, instrumentation, controls, chillers and CDUs are supplied by others, and Harris does not perform field erection. Sending design conditions, the nozzle schedule, internals drawings and site constraints early lets Harris confirm what can be shop-built and quote it against the project's design basis.
FAQs about thermal energy storage tanks for data centers
How much cooling can a chilled-water TES tank store?
Stored energy equals water mass × specific heat × usable ΔT. At a 16°F (8.9 K) usable difference, 1,000,000 gal (3,785 m³) holds about 11,120 ton-hours, or 39.1 MWh thermal, in theory. Delivered capacity is lower because of mixing, thermocline and dead volume, heat gain and the ΔT the load actually returns.
Why is cold water stored at the bottom of a TES tank?
Across the normal chilled-water range, colder water is denser, so it settles beneath warmer return water with a thermocline between them. Water is densest near 39°F (4°C), and colder water becomes lighter. Storage much below about 39°F therefore cannot rely on natural stratification alone.
What is the difference between a TES tank and a chilled-water buffer tank?
A buffer tank adds volume and thermal mass to steady plant control and prevent short-cycling; it is not designed to deliver a defined inventory of stored cooling. A TES tank is designed for deliberate charging and discharging against a load profile, with diffusers that maintain stratification.
Can a thermal energy storage tank keep a data center cool during a power outage?
Only as part of a system. Stored water can bridge the minutes until chillers restart and reload, but it must still be pumped, and the controls and other equipment that deliver cooling must stay powered. Which equipment needs uninterruptible power depends on the cooling architecture.
Does a pressurized chilled-water storage vessel need an ASME U-stamp?
Not automatically. ASME Section VIII, Division 1 excludes some vessel classes from its scope, including certain vessels containing water under pressure within stated limits. Jurisdictional rules or owner specifications may still require Code construction, and a vessel built and certified voluntarily must meet the applicable Division 1 provisions. Confirm the edition in force and local requirements first.
What causes a TES tank to lose usable capacity?
The main causes are inlet mixing, thermocline growth, flow above design, return water colder than design, heat gain, and sensors that misreport state of charge. In U.S. Army research, warming the cold layer by 0.9°F (0.5 K) on an 18°F (10 K) design difference cost 5% of capacity.