Why Use a TES Tank in Data Centers: Uptime Load Shifting and AI Clusters

Data centers use TES tanks because the UPS protects IT power, not cooling, and high-density racks climb out of the ASHRAE envelope within minutes of lost cooling. A charged stratified tank gives 15 to 30 minutes of ride-through, flattens the day/night load curve, and absorbs abrupt AI/GPU ramps. Center Enamel has no published data-center TES delivery; its 44,900 m3 Namibia GFS build is cited as large-vessel capability evidence only. The TES shell itself is a bolted GFS/FBE/stainless or ASME vessel built to AWWA D103 or ASME VIII.

Data centers use TES tanks for uptime, load shifting and AI-cluster stability. A charged tank gives 15 to 30 minutes of ride-through when a chiller trips or a feeder drops - the cooling equivalent of the UPS that protects IT power - bridging the gap before generators stabilize. Hyperscale and colocation campuses use stratified storage to flatten the day/night load curve and cut peak demand, while AI and GPU clusters use it to absorb abrupt thermal ramps without over-building chillers. Free-cooling hours are captured and equipment duty cycles smooth out, extending life. The shell is a large bolted or welded vessel built to AWWA D103 or ASME VIII. Center Enamel has no published TES delivery; its large-bolted-vessel builds are cited as manufacturing-capability evidence, not as data-center projects.

This is the single fact that drives data-center TES. The UPS keeps servers powered, but if chilled water stops in a high-density hall, rack inlet temperatures climb within minutes and the room can leave the ASHRAE recommended envelope before a generator finishes its start sequence. A charged TES tank is the cooling equivalent of the UPS battery - the reserve that bridges that gap.

• UPS protects IT power, not cooling.
• Rack inlet temperature climbs within minutes of lost cooling.
• Tank bridges the chiller-trip to generator-stable gap.
• Cooling has no equivalent of the UPS battery without a tank.

Power failure - often a transfer-switch or generator fault rather than a grid event - is a leading cause of impactful outages. A tank sized for 15 to 30 minutes of full cooling keeps the halls safe through the restart and the gap before standby generation is fully online, exactly where data centers actually fail.

• 15-30 minutes full-cooling ride-through.
• Bridges generator start and transfer.
• Covers the leading cause of outages (transfer/switch faults).
• Keeps racks inside the ASHRAE envelope.

Cooling can be 30 to 40 percent of a data center's electricity. By charging off-peak and discharging through the afternoon peak, a TES tank cuts demand charges and shifts energy to cheaper, cooler hours. For a large facility buying on wholesale power, that arbitrage is concrete.

• Cooling: 30-40 percent of facility electricity.
• Off-peak charge, peak discharge.
• Cuts demand charges and peak draw.
• Arbitrage on wholesale power where available.

AI clusters create abrupt, large cooling ramps that a chiller plant cannot follow instantly. A TES tank absorbs those ramps as a thermal buffer, smoothing the mismatch between GPU load spikes and chiller response, and protecting compressors from short-cycling during volatile demand.

• AI/GPU loads ramp abruptly.
• Tank absorbs ramp as thermal buffer.
• Smooths chiller response to load spikes.
• Protects compressors from short-cycling.

Hyperscale campuses use TES to flatten the day/night curve across many halls; colocation sites use it as shared resilience for tenants; edge and AI sites use it to absorb local ramps. The tank strategy is the same; only the sizing and the tenant model differ.

• Hyperscale: day/night curve flattening.
• Colocation: shared tenant resilience.
• Edge/AI: local ramp absorption.
• Same tank strategy, different sizing.

Honest boundary: Center Enamel has no published thermal-energy-storage or data-center TES delivery in its verified project record. The references above (Namibia 44,900 m3 GFS drinking water, 2022; Cilincing 21,088 m3; Sardinia 30,549 m3) are cited strictly as large-diameter bolted-vessel manufacturing-capability evidence, not as TES or data-center projects. No TES project result, dimension or client is claimed.

Because the UPS protects IT power, not cooling, and high-density racks leave the ASHRAE envelope within minutes of lost chilled water. A tank is the cooling reserve that bridges a chiller trip to generator stability.

Common practice sizes 15 to 30 minutes of full cooling ride-through, enough to cover chiller restart and generator transfer.

Yes, by shifting cooling off-peak and discharging on the afternoon peak, cutting demand charges and moving energy to cheaper, cooler hours.

AI clusters ramp cooling abruptly; the tank absorbs those ramps as a buffer, smoothing chiller response and protecting compressors from short-cycling.

No. Hyperscale uses it for curve flattening, colocation for shared resilience, and edge/AI sites for local ramp absorption - the same strategy at different scales.

Honest boundary: Center Enamel has no published thermal-energy-storage or data-center TES delivery in its verified project record. The references above (Namibia 44,900 m3 GFS drinking water, 2022; Cilincing 21,088 m3; Sardinia 30,549 m3) are cited strictly as large-diameter bolted-vessel manufacturing-capability evidence, not as TES or data-center projects. No TES project result, dimension or client is claimed.

Typically a bolted GFS, FBE or stainless atmospheric tank, or an ASME pressurized vessel for a pressurized loop, built to AWWA D103 or ASME VIII.

The site peak cooling deficit and the required hours of reserve (ride-through plus load shift), converted to a water volume using the supply-to-return temperature difference.

Atmospheric bolted tanks follow AWWA D103 and ISO 28765; pressurized shells follow ASME Section VIII, with NSF/ANSI 61 and WRAS for potable or indirect-potable contact.

Data centers use TES tanks because the UPS protects IT power but not cooling, and high-density racks fail within minutes of lost chilled water. A charged stratified tank gives 15 to 30 minutes of ride-through, flattens the load curve, and absorbs AI/GPU ramps. Center Enamel supplies the bolted GFS, FBE, stainless and ASME vessel families behind the shell and has no published data-center TES delivery; its large-water builds are cited as capability evidence only.