Verified liquid-cooling deployment tracker

What you can answer with it

• Count operating capacity without counting announcements twice.
• See which architectures have moved past pilot in a given market.
• Find the projects a supplier can actually point to, and the ones it cannot.

How the rows were built

1. Rows are the deployment records on this site. Each carries a stage taken from the most recent source, and the stage history is on the record.
2. Stage is deliberately limited to four values (operating, deploying, planned, and pilot) because a finer scale would imply knowledge the sources do not contain.
3. Capacity is quoted as the source states it. Where a project reports IT load and facility load differently, the record says which.
4. A project with no source is not published, so it is not in this file.

The full editorial standard, including how sources are ranked and how project stages are labelled, is on the methodology page.

PreviewFirst 12 of 24 rows

Data dictionary13 columns

projectProject

Project or site name as the sources give it.

operatorOperator

The party running the facility, not the tenant.

locationLocation

Stated to the granularity the source supports.

stageStage

Where the project is: announced, planned, pilot, deploying, or operating. Stage is not evidence strength.

announcedAnnounced

Date of the earliest public announcement on file.

capacityCapacity

As stated by the source; IT load and facility load are distinguished on the record.

densityRack density

Per-rack power as published, where published.

computeCompute

Accelerator or server platform named by the source.

architectureArchitecture

Cooling architectures in use, from the record's links.

suppliersNamed suppliers

Only suppliers a source names. Absence is not evidence of non-involvement.

sourcesSources

Count of cited sources on the underlying record.

reviewedReviewed

When a human last read the record this row came from.

recordRecord

Canonical URL of the record, so every row is traceable.

Versions and change log

This file is v4, generated 2026-11-08T10:46:17.330Z from 24 records. The version counts the publishes that changed this file, so it never moves backwards and a citation keeps meaning what it meant. The rows themselves are always regenerated from the records, so a correction reaches the file as soon as it reaches the record.

1. v4Oct. 4, 2026Published by event research-correction-2026-10-04-1791102709240.
2. v3Oct. 4, 2026Published by event research-correction-2026-10-04-1791102066737.
3. v2Oct. 4, 2026Published by event research-correction-2026-10-04-1791101342042.
4. v1Oct. 4, 2026Published by event research-release-plant-cooling-2026-rack-loop-2026-fluids-controls-2026.

• Oct. 4, 2026

  Research expansion published: 90 records

  • Smardt New York-region data-center chiller replacement description: Status: operating. That status rests on Smardt's report of early startup with stable operation and efficiency. The unnamed New York-area operator needed two AD120 water-cooled chillers. Each unit had three TT350 oil-free magnetic-bearing compressors. The project had a fixed startup date. Smardt says a third-party carrier badly damaged one unit. Smardt then changed factory plans and supply work. It says a replacement arrived within three months. This supports the supplier's response claim. The account still comes from the supplier, and the operator is not named. Public sources omit cooling duty and water temperatures. They also omit refrigerant, measured efficiency, design-load backup, test rules, rack density, computing platform, final startup date, and long-term results. Treat this as a delivery reference, not a heat-duty benchmark. Ask for an operator contact and test data. “Stable efficiency” has no stated rating method. No kilowatts-per-ton value, load point, condenser-water condition, test period, or outside witness is given. Buyers should ask if both chillers passed final tests. They should ask if the damaged unit was rebuilt or replaced. Open startup defects should be disclosed. The three-month period needs a clear start and end date. A reference call should cover packing, shipping controls, spare compressors, electronics, startup staff, alarms, control links, restart tests, service space, and response time. Design comparison needs the exact AD120 selection, certified duty, power, refrigerant, size, hours, availability, and output after loss of a compressor, chiller, or support system. → Status: operating, based on Smardt's report of an early startup with stable operation and efficiency. The unnamed New York-area operator needed two AD120 water-cooled chillers, each with three TT350 oil-free magnetic-bearing compressors. A third-party carrier badly damaged one unit before a fixed startup date. Smardt says it changed factory and supply plans and delivered a replacement within three months. The account demonstrates the supplier's reported response, but the operator is unnamed and has not provided a separate account. Cooling duty, water temperatures, refrigerant, measured efficiency, design-load backup, test rules, rack density, computing platform, final startup date, and long-term results remain undisclosed. Use the project as a delivery reference rather than a heat-duty benchmark. Smardt gives no rating method for “stable efficiency,” kilowatts per ton, load point, condenser-water condition, test period, or outside witness. Reference questions should establish whether both chillers passed final tests, whether the damaged unit was rebuilt or replaced, and whether startup defects remained open. The three-month period also needs defined start and end dates. Ask about packing, shipping controls, spare compressors, electronics, startup staff, alarms, control links, restart tests, service space, and response time. Design comparison requires the exact AD120 selection, certified duty, power, refrigerant, size, operating hours, availability, and output after loss of a compressor, chiller, or support system.
  • BAC 160 MW HPC HXV cooling project description: Status: operating. Baltimore Aircoil Company says the installed HXV system met the customer's water-temperature target. It also says the system supported the information-technology load. The customer had 160 MW of operating power for high-performance computing. It had used open towers with water-cooled chillers. It chose HXV hybrid coolers for dry and evaporative operation without chillers. The supplier case gives estimates and claimed savings. It withholds the operator and location. It gives no outside meter data, unit count, water temperatures, weather file, baseline boundary, startup report, or test period. The record proves a supplier-documented site and design. It does not independently prove power usage effectiveness or water usage effectiveness. It also does not prove energy, water, or cost results. Buyers should ask for the model, measured data after startup, and a customer contact. The 160 MW figure is customer operating power. It is not published tower heat duty. Do not infer unit count or size from it. The case does not say if HXV serves the whole campus, one phase, or part of the load. Qualification needs exact old and new system boundaries. It needs hourly weather, load, and mode hours. Ask for makeup, blowdown, fan energy, pump energy, outlet temperatures, reserve rule, and hot-day output. Ask about water limits, poor water, plume, freezing, fan faults, spray-pump faults, and loss of shared controls. Final test records and one year of utility and service data would be more useful than percentage claims alone. → Status: operating. Baltimore Aircoil Company says the installed HXV system met the customer's water-temperature target and supported the information-technology load. The customer had 160 MW of operating power for high-performance computing. It replaced open towers and water-cooled chillers with HXV hybrid coolers capable of dry and evaporative operation without chillers. The supplier case gives estimates and claimed savings but withholds the operator and location. It provides no outside meter data, unit count, water temperatures, weather file, baseline boundary, startup report, or test period. The case documents a site and design without independently proving power usage effectiveness, water usage effectiveness, energy, water, or cost results. Request the model, measured post-startup data, and a customer contact. The 160 MW figure is customer operating power, not published tower heat duty, and cannot support an inferred unit count or size. The case also leaves unclear whether HXV serves the entire campus, one phase, or part of the load. Qualification needs exact old and new system boundaries, hourly weather, load, and mode hours. Ask for makeup, blowdown, fan and pump energy, outlet temperatures, reserve rule, and hot-day output. Water limits, poor water, plume, freezing, fan faults, spray-pump faults, and loss of shared controls also need review. Final test records and one year of utility and service data would carry more weight than percentage claims alone.
  • Güntner Zhangjiakou immersion-cooling heat rejection description: Status: operating. Güntner says an unnamed large Chinese online seller runs the data center. The Zhangjiakou site has used two-phase immersion cooling since 2017. Four V-shape VARIO dry coolers serve it. hydroBLU adiabatic help is also fitted. Güntner states 1,060 kW of total heat-rejection duty. Vapor condenses inside the immersion chambers. A loop then moves that heat to the outdoor coolers. The supplier says water is used for help only when weather requires it. Güntner claims large energy and cost savings over normal systems. It does not name the operator or give the baseline. It omits the unit schedule, water use, weather adjustment, raw meter data, rack density, computing platform, availability, and outside review. Buyers can use the case to check the design path and scale. They should still request a site contact and yearly fan, pump, and water data. The evidence does not define the 1,060 kW figure. It may be installed duty, design-day duty, or measured heat. Fluid temperatures, glycol, air approach, fan power, and reserve are also unknown. Four units may allow staging. Their backup role is not stated. A similar-site review should ask for model and fan details. It needs the hydroBLU start point, yearly and peak water, pad service, treatment, dry-only output, winter order, alarm history, cleaning, and output after one cooler or shared pump loop fails. Review the middle heat exchanger and pumps too. Outdoor duty alone does not prove chamber condensation, fluid fit, server service, or full cooling availability. → Status: operating. Güntner says an unnamed large Chinese online seller has run two-phase immersion cooling at the Zhangjiakou site since 2017. Four V-shape VARIO dry coolers with hydroBLU adiabatic assistance provide 1,060 kW of stated total heat-rejection duty. Vapor condenses inside the immersion chambers, and a loop moves the heat to the outdoor coolers. According to Güntner, adiabatic water is used only when weather requires it. The supplier claims large energy and cost savings over conventional systems without naming the operator or defining the baseline. Unit schedules, water use, weather adjustment, raw meter data, rack density, computing platform, availability, and outside review are not disclosed. The case is useful for checking the design path and scale, subject to a site contact and yearly fan, pump, and water data. Güntner does not define whether 1,060 kW means installed duty, design-day duty, or measured heat. Fluid temperatures, glycol, air approach, fan power, and reserve also remain unknown. Four units may allow staging, but no backup role is stated. A comparable-site review should cover model and fan details, the hydroBLU start point, yearly and peak water, pad service, treatment, dry-only output, winter sequence, alarm history, cleaning, and output after one cooler or the shared pump loop fails. Include the intermediate heat exchanger and pumps. Outdoor duty alone cannot establish chamber condensation, fluid compatibility, server service, or full cooling availability.
  • Infosys Hyderabad data-center heat exchangers description: Status: operating. SWEP says Infosys and design lead Schneider Electric built several data-center cooling projects. The work ran from 2016 through 2020. B439 and B649 brazed plate heat exchangers separate the main cooling source from the inner loop. At the Hyderabad campus, several B649 units have 6-inch ports. SWEP reports 350 cubic meters per hour through each unit. It says the units were running without problems when the case appeared. The named operator, design partner, models, loop role, period, and unit flow make this a useful reference. The evidence still comes from the supplier. Public sources omit unit count, heat duty, temperatures, pressure loss, rack density, computing platform, power usage effectiveness, measured savings, service history, and outside test data. Flow is not heat duty without temperatures and fluid data. Do not turn 350 cubic meters per hour into megawatts by assumption. “Without problems” also lacks a time period and uptime measure. There is no alarm log, leak history, fouling trend, or service record. Reference questions should cover exact articles and plate counts. Ask for water chemistry on both sides, approach, pressure loss, pump energy, strainers, filters, cleaning rate, isolation, bypass, spares, and cross-leak detection. Ask how parallel units share flow and what duty remains during cleaning or replacement. Confirm field tests. Separate exchanger savings from results of the wider cooling design. → Status: operating. SWEP says Infosys and design lead Schneider Electric built several data-center cooling projects between 2016 and 2020. B439 and B649 brazed plate heat exchangers separate the main cooling source from the inner loop. At the Hyderabad campus, several B649 units have 6-inch ports and a reported flow of 350 cubic meters per hour through each unit. SWEP says the units were running without problems when the case appeared. The named operator, design partner, models, loop role, period, and unit flow make the installation a useful reference, although the account still comes from the supplier. Unit count, heat duty, temperatures, pressure loss, rack density, computing platform, power usage effectiveness, measured savings, service history, and outside test data remain undisclosed. Flow is not heat duty without temperatures and fluid data, so 350 cubic meters per hour cannot be converted into megawatts by assumption. “Without problems” also lacks a time period and uptime measure. No alarm log, leak history, fouling trend, or service record is available. Reference questions should cover exact articles and plate counts, water chemistry on both sides, approach, pressure loss, pump energy, strainers, filters, cleaning rate, isolation, bypass, spares, and cross-leak detection. Confirm parallel-unit flow balance, duty during cleaning or replacement, and field tests. Keep exchanger savings separate from the results of the wider cooling design.
  • MITA Barcelona data-processing center cooling towers description: Status: operating. This status rests on MITA Group's 2023 case. The case describes a completed supply and says the towers support equipment performance. TORRAVAL Cooling is part of MITA Group. It supplied sixteen CTFP 2436 forced-draft open-circuit towers. The units have laminar fill. They serve a large Barcelona data-processing center for an international information-technology company. The source gives a required load of 20 MW. It says the site can grow to 40 MW and has 5 MW of emergency supply. The source does not define these values as computing, electrical, or heat duty. This record keeps the source wording and does not convert it. Two facing rows fit the towers on a limited roof. The layout also supported sound control. Walkways and ladders were included. Public evidence withholds the operator and installer. It omits water temperatures, flow, measured power, measured water, backup, startup tests, rack density, computing platform, uptime, and an operator account. Buyers should request the approved schedule, test data, yearly records, and a reference contact. The case also states 15 kW for each tower. That figure is not used here as heat duty. Its meaning conflicts with the multi-megawatt need and may be motor power, but the source is silent. Resolve it from the schedule. Qualification needs design wet bulb, water temperatures, cell flow, fan power, drift, evaporation, blowdown, concentration cycles, sound, plume, treatment, hygiene, roof vibration, structural loads, and output after a cell or common pump fails. Confirm whether future growth was installed, reserved, or only planned. → Status: operating, based on MITA Group's 2023 account of a completed supply. TORRAVAL Cooling, part of MITA Group, supplied sixteen CTFP 2436 forced-draft open-circuit towers with laminar fill. They serve a large Barcelona data-processing center for an international information-technology company. MITA gives a required load of 20 MW, possible growth to 40 MW, and 5 MW of emergency supply. It does not identify those values as computing, electrical, or heat duty, so they remain unconverted. Two facing rows fit the towers on a limited roof and supported sound control. Walkways and ladders were included. The operator and installer are unnamed. Water temperatures, flow, measured power and water, backup, startup tests, rack density, computing platform, uptime, and an operator account remain undisclosed. Request the approved schedule, test data, yearly operating records, and a reference contact. MITA also states 15 kW for each tower. That figure is not treated as heat duty because its meaning conflicts with the multi-megawatt requirement and may instead be motor power. Only the schedule can resolve it. Qualification needs design wet bulb, water temperatures, cell flow, fan power, drift, evaporation, blowdown, concentration cycles, sound, plume, treatment, hygiene, roof vibration, structural loads, and output after a cell or common pump fails. Confirm whether future growth was installed, reserved, or only planned.
  • Fugaku Stäubli CGD cooling connections description: Status: operating. Stäubli says Fujitsu selected CGD metal couplings to connect liquid-cooled blades to Fugaku's chassis and to permit tool-free hot swapping with flush-face, non-spill connections. Its current data-center page reports 400 racks, more than 150,000 processors and 384 connection points per rack. RIKEN independently confirms that development was completed and shared use began on 9 March 2021. The sources establish an operating named system and connector role, but they do not publish coolant chemistry, coupling size, pressure drop, leak history or maintenance replacement rate. Buyers should treat Fugaku as evidence of deployment scale, not as proof that the same connector selection fits a different rack or coolant. The Stäubli story was written while Fugaku was being installed, whereas RIKEN provides the firmer operating milestone; neither source is a connector acceptance report. The public record does not identify the number of coupling pairs in active coolant service versus other reported connection points, the pressure and temperature envelope, the guide mechanism's tolerance, or any leakage events over time. Due diligence should ask Stäubli and Fujitsu for the exact CGD configuration, qualification standard, coolant and material stack, hot-swap procedure, inspection interval, spare strategy and field-return data. A prospective buyer should also reproduce blade insertion and removal with the proposed chassis tolerances and verify pressure drop, residual spill and air inclusion after repeated cycling. Fugaku supports confidence that CGD can be engineered into a very large system; it does not establish an independently measured failure rate or lifecycle cost. → Status: operating. Stäubli says Fujitsu selected CGD metal couplings to connect liquid-cooled blades to Fugaku's chassis and to permit tool-free hot swapping with flush-face, non-spill connections. Its current data-center page reports 400 racks, more than 150,000 processors and 384 connection points per rack. RIKEN independently confirms that development was completed and shared use began on 9 March 2021. The sources establish an operating named system and connector role. They do not publish coolant chemistry, coupling size, pressure drop, leak history or maintenance replacement rate. Fugaku demonstrates deployment scale, not that the same connector selection fits a different rack or coolant. The Stäubli story was written while Fugaku was being installed, whereas RIKEN provides the firmer operating milestone. Neither source is a connector acceptance report. The available evidence does not identify the number of coupling pairs in active coolant service versus other reported connection points. It also omits the pressure and temperature envelope, guide-mechanism tolerance and leakage events over time. Ask Stäubli and Fujitsu for the exact CGD configuration, qualification standard, coolant and material stack and hot-swap procedure. Inspection intervals, spare strategy and field-return data are also needed. Reproduce blade insertion and removal with the proposed chassis tolerances. Verify pressure drop, residual spill and air inclusion after repeated cycling. Fugaku shows that CGD was engineered into a very large system; it does not establish an independently measured failure rate or lifecycle cost.
  • DataBank ATL1 ColdLogik rear-door cooling description: Status: operating. USystems and DataBank describe ColdLogik rear-door heat exchangers at ATL1 in Georgia Tech's CODA development. The published design point is 50 kW per enclosure with 73°F warm water, with a stated path to 100 kW per rack after limited infrastructure changes. DataBank's current facility page confirms ATL1 remains an operating site hosting Georgia Tech's High-Performance Computing Center and advertises liquid-cooled cabinet capacity. The vendor article's savings claims are not backed by a disclosed measurement protocol, and the installed door count, actual IT load and current water conditions are not public. The record therefore supports the architecture and operating site, not the full claimed efficiency comparison. DataBank's current page lists 7.1 MW of facility critical IT load, but that figure must not be attributed to the ColdLogik subsystem; the cooling source does not state how many racks or megawatts use rear doors. Likewise, the 100 kW statement describes potential after infrastructure changes rather than demonstrated operating rack density. Procurement teams should request current rack counts, door models, supply and return temperatures, water flow, fan power, maintenance history and measured seasonal cooling energy. They should also ask how ATL1 maintains temperatures while a door is open or isolated, how leaks are detected, and whether server-fan settings changed. An operator reference should separate original design targets from present operation and identify any control, hinge, hose or water-quality lessons learned since commissioning. → Status: operating. USystems and DataBank describe ColdLogik rear-door heat exchangers at ATL1 in Georgia Tech's CODA development. The published design point is 50 kW per enclosure with 73°F warm water, with a stated path to 100 kW per rack after limited infrastructure changes. DataBank's current facility page confirms ATL1 remains an operating site hosting Georgia Tech's High-Performance Computing Center and advertises liquid-cooled cabinet capacity. The vendor article's savings claims are not backed by a disclosed measurement protocol. The installed door count, actual IT load and current water conditions are also unavailable. The sources support the architecture and operating site, not the full claimed efficiency comparison. DataBank's current page lists 7.1 MW of facility critical IT load, but that figure must not be attributed to the ColdLogik subsystem. The cooling source does not state how many racks or megawatts use rear doors. Likewise, the 100 kW statement describes potential after infrastructure changes rather than demonstrated operating rack density. Request current rack counts, door models, supply and return temperatures, water flow and fan power. Maintenance history and measured seasonal cooling energy are also needed. Ask how ATL1 maintains temperatures while a door is open or isolated, how leaks are detected and whether server-fan settings changed. An operator reference should separate original design targets from present operation and identify any control, hinge, hose or water-quality lessons learned since commissioning.
  • West Cambridge Data Centre ColdLogik cooling description: Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later raised cabinet density from 30 to 44 kW. It also reports that data-hall capacity rose from 900 kW to 1.2 MW. Those figures come from the cooling supplier, not an independent acceptance test. The university does confirm the live site and its ongoing research-computing role. Public sources do not give the current door count or inlet-water temperature. They also omit yearly cooling energy, backup design and whether each hall uses the same system. Buyers should use the reference to seek operator contacts and measured seasonal data. The university history confirms the move and continued use of the site. It does not link the later capacity gain to a given ColdLogik model. The 1.2 MW figure must remain a supplier-reported hall value. It is not a measured IT load or proof that all cabinets run at 44 kW. An operator interview should confirm installed models and counts. It should cover normal and peak rack loads, water temperatures and fan energy. Ask about leaks, service work and changes made after start-up. Buyers should also ask if the growth required plant, pipe or control changes beyond the rear doors. They should learn how cooling stays live during door service. Annual trend data needs a clear electrical boundary. Without it, the reference cannot predict power usage effectiveness (PUE) or operating cost. → Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later raised cabinet density from 30 to 44 kW. It also reports that data-hall capacity rose from 900 kW to 1.2 MW. Those figures come from the cooling supplier, not an independent acceptance test. The university does confirm the live site and its ongoing research-computing role. Available sources do not give the current door count or inlet-water temperature. They also omit yearly cooling energy, backup design and whether each hall uses the same system. Use the reference to seek operator contacts and measured seasonal data. The university history confirms the move and continued use of the site. It does not link the later capacity gain to a given ColdLogik model. The 1.2 MW figure must remain a supplier-reported hall value. It is not a measured IT load or proof that all cabinets run at 44 kW. An operator interview should confirm installed models and counts. It should cover normal and peak rack loads, water temperatures and fan energy. Ask about leaks, service work and changes made after start-up. Confirm whether the growth required plant, pipe or control changes beyond the rear doors. Determine how cooling stays live during door service. Annual trend data needs a clear electrical boundary. Without it, the reference cannot predict power usage effectiveness (PUE) or operating cost.
  • NorthC Eindhoven MIXIT chip-cooling loop description: Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes; a customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The public record does not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. The June 2022 date in this record represents the reported mid-2022 delivery period; the cited sources do not publish a precise commissioning day. MIXIT's exact configuration, valve size, sensors, controlled variable and relationship to pumps or a coolant distribution unit (CDU) are also undisclosed. Procurement teams should ask NorthC or Hamer for a loop diagram, control sequence, temperature and flow trends, alarm history, maintenance interventions and behavior during server or network changes. They should distinguish heat made available to the campus ring from heat actually reused by a customer and quantify backup rejection when campus demand is absent. Evidence that would strengthen the record includes a named server platform, installed kilowatts, measured pump and cooling energy, water temperatures, uptime and acceptance criteria. Until then, capacity and rack density correctly remain not publicly disclosed. → Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes. A customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The sources do not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. The June 2022 date reflects the reported mid-2022 delivery period; the sources give no precise commissioning day. MIXIT's exact configuration, valve size, sensors and controlled variable are also undisclosed. Its relationship to pumps or a coolant distribution unit (CDU) is unknown. Ask NorthC or Hamer for a loop diagram, control sequence and temperature and flow trends. Alarm history, maintenance interventions and behavior during server or network changes are also needed. Distinguish heat made available to the campus ring from heat actually reused by a customer. Quantify backup rejection when campus demand is absent. A named server platform, installed kilowatts, measured pump and cooling energy, water temperatures, uptime and acceptance criteria would add useful evidence. Until then, capacity and rack density remain not publicly disclosed.
  • Digital Realty HKG10 Armstrong chilled-water plant rooms description: Status: operating. Armstrong reported in September 2016 that it supplied off-site-manufactured, integrated chilled-water plant rooms for the second phase of HKG10, then a Digital Realty and CenturyLink joint venture. The packages were assembled and tested at Armstrong's Halesowen factory and used variable-primary distribution. Digital Realty's current HKG10 page confirms the facility is operating at 33 Chun Choi Street with N+1 cooling, although it does not identify the plant supplier. Public sources do not disclose pump models, plant tonnage, measured efficiency, water use, commissioning results or how much of today's facility remains served by the 2016 packages. The record supports a real delivered plant reference, with performance claims still supplier reported. Digital Realty's current N+1 statement describes the facility today but does not establish the redundancy arrangement of each 2016 package or prove that the original plant remains unchanged. Armstrong's article describes factory integration and testing without publishing the test schedule, witnessed results, controls architecture or post-commissioning energy data. A buyer evaluating packaged plant should request the original and current equipment schedules, package boundaries, design water temperatures, pump curves, staging logic, failure tests, transport splits and site reconnection work. Reference questions should cover factory-versus-field defects, commissioning duration, spare parts, controls updates and service response in Hong Kong. Measured annual chiller, pump and heat-rejection energy under local weather would be needed before using this project as an efficiency benchmark; no such dataset appears in the cited public sources. → Status: operating. Armstrong reported in September 2016 that it supplied off-site-manufactured, integrated chilled-water plant rooms for the second phase of HKG10, then a Digital Realty and CenturyLink joint venture. The packages were assembled and tested at Armstrong's Halesowen factory and used variable-primary distribution. Digital Realty's current HKG10 page confirms the facility is operating at 33 Chun Choi Street with N+1 cooling, although it does not identify the plant supplier. Available sources do not disclose pump models, plant tonnage, measured efficiency, water use or commissioning results. They also do not show how much of today's facility remains served by the 2016 packages. The evidence supports a delivered plant reference, while performance claims remain supplier reported. Digital Realty's current N+1 statement describes the facility today. It does not establish the redundancy arrangement of each 2016 package or prove that the original plant remains unchanged. Armstrong's article describes factory integration and testing without publishing the test schedule, witnessed results, controls architecture or post-commissioning energy data. Request the original and current equipment schedules, package boundaries, design water temperatures and pump curves. Staging logic, failure tests, transport splits and site reconnection work also need review. Reference questions should cover factory-versus-field defects, commissioning duration, spare parts, controls updates and service response in Hong Kong. Measured annual chiller, pump and heat-rejection energy under local weather would be needed before using this project as an efficiency benchmark. No such dataset appears in the cited sources.
  • Castrol–Hypertec Pangbourne immersion lab description: Status: pilot. On 20 July 2023, Castrol said its Pangbourne immersion systems were installed and working. The research setup joins Castrol fluids with Hypertec server skills. It also uses Submer SmartPod and MicroPod tanks. The partners planned to test fluids, servers, and the full cooling setup. This is a real site with named partners. It is a supplier lab, not a production data center. The release gives no information-technology load or tank count. It omits server models and the fluid used in each test. It also omits water temperatures, heat-rejection gear, test steps, measured efficiency, uptime, and customer acceptance. The site proves joint research, not capacity or savings. Buyers should request the full bill of materials. They need test length, duty, material checks, fluid results, fault tests, and later design changes. Long-run data would strengthen the record. So would an operator report or a commercial site with the same mix. The release does not say which Castrol fluid was in each tank. It does not say if tests used sale batches. It also does not tie Submer's later DC 20 approval to this lab. Those links cannot be assumed. Buyers should ask for fluid values before and after each test. They need the server list, time in fluid, heat profile, pump and heat-exchanger data, service work, spills, dirt events, and final inspection. A commercial project also needs fire rules, worker steps, fluid stock, lift and drain plans, spare capacity, and a warranty owner. No later public update confirms current use or setup. A fresh site reference or dated report is needed before calling Pangbourne active. → Status: pilot. On 20 July 2023, Castrol said its Pangbourne immersion systems were installed and working. The setup combines Castrol fluids with Hypertec server expertise. It also uses Submer SmartPod and MicroPod tanks. The partners planned to test fluids, servers, and the full cooling system. Pangbourne is a named supplier lab, not a production data center. The release gives no information-technology load or tank count. It omits server models and the fluid used in each test. Water temperatures, heat-rejection equipment, test steps, measured efficiency, uptime, and customer acceptance are also unknown. The site supports joint testing, not a capacity or savings claim. Reference checks need the bill of materials and test duration. They also need the duty cycle, material inspections, fluid results, failure tests, and later design changes. Long-run data, an operator report, or a commercial site with the same qualified mix would offer stronger proof. The release does not state which Castrol fluid was in each tank. It does not say whether tests used production batches. It also does not link Submer's later DC 20 approval to Pangbourne. Those links cannot be assumed. Test records should show fluid properties before and after each run. They should list servers, components, time in fluid, heat load, pump and heat-exchanger data, service work, spills, contamination, and final inspections. Commercial qualification also needs fire rules, worker steps, fluid stock, lift and drain plans, spare capacity, and a warranty owner. No later update confirms current use or setup. Calling Pangbourne active requires a dated report or current site reference.
  • Valvoline–Iceotope 18-month fluid validation description: Status: pilot. A Valvoline case from December 2025 covers an 18-month test of HTF-DE1 fluid. The test used an Iceotope MicroDC system. It had HPE DL380 servers and NVIDIA A40 graphics processors. The stated load was 5.2 kW. Valvoline says viscosity stayed steady. It says dielectric strength stayed above the Open Compute Project floor. A final inspection found no rust or material harm. The named gear, long test, and later inspection add weight. Limits remain. The site is not named. A 5.2 kW test does not represent a dense full rack. Raw values and uncertainty are not public. The fluid supplier wrote the report. The test does not prove fit with other servers, plastics, cables, drives, or heat ranges. Buyers need the full test plan. They should request fluid results from start to finish, sample history, inspection limits, faults, service events, and written Iceotope approval for the exact system. A repeat test or named user would add confidence. The source does not say if 5.2 kW is average, peak, or nameplate load. It does not say how often the servers ran. Inlet and outlet temperatures are unknown. Filter use and top-ups are also unknown. Individual electrical and viscosity readings are not public. Nor are lab methods, uncertainty, or detailed inspection records. Buyers should confirm that HPE and NVIDIA warranties stayed valid. They should ask which MicroDC versions Iceotope approves. A production test must use the planned materials and heat range. It should include real service work, sealed baseline samples, preset pass limits, and a named owner for action after a bad sample. → Status: pilot. Valvoline's December 2025 case covers an 18-month test of HTF-DE1 fluid in an Iceotope MicroDC system with HPE DL380 servers and NVIDIA A40 graphics processors. The stated load was 5.2 kW. Valvoline reports stable viscosity, dielectric strength above the Open Compute Project minimum, and no corrosion or material damage during final inspection. The named hardware, duration, and inspection support compatibility for that setup. The site is not disclosed, and a 5.2 kW test does not represent a full high-density rack. Raw measurements and uncertainty are unavailable, and the fluid supplier wrote the report. Compatibility with other servers, plastics, cables, storage devices, and operating temperatures remains unproven. Qualification requires the full protocol, starting and ending fluid analyses, sampling history, inspection criteria, excursions, maintenance events, and written Iceotope approval for the exact production system. Independent replication or a named commercial operator would provide stronger evidence. The source does not identify 5.2 kW as average, maximum, or nameplate load. It also omits server duty cycle, inlet and outlet temperatures, filtration, top-ups, individual dielectric and viscosity readings, laboratory methods, uncertainty, and detailed forensic records. HPE and NVIDIA warranty status needs confirmation, as does the set of approved MicroDC revisions. A production pilot should reproduce the intended materials and temperature range, include representative service events, retain sealed baseline samples, set pass and fail limits before testing, and assign authority for action after an out-of-limit sample.
  • NREL ESIF warm-water cooling loop description: Status: operating. Xylem says more than 55 Bell & Gossett products went into the Energy Systems Integration Facility. The site is at the National Renewable Energy Laboratory in Golden, Colorado. Named pump families include e-1510, Series 90, Series 80, and Series 60. The data center uses warm water close to the computer parts. It sends heat through an energy-recovery loop. Useful heat warms labs and offices. NREL says the water system was installed in 2012. It also describes an order for heat reuse, dry rejection, and cooling towers. The sources prove named product families at a live site. They do not show which pump serves each loop. Current model counts, flow, head, yearly pump power, and service history are unknown. Whole-site efficiency cannot be assigned only to Xylem. The computers have also changed over time. Buyers should study controls, low-load use, standby design, water care, heat demand, and service. They should not copy the old pump schedule. A fair comparison needs current drawings, trends, and service records. The source gives only the year 2012. It gives no exact day. The structured date 2012-01-01 is a format placeholder, not a claimed start date. NREL's current description includes later heat-rejection additions. It may not match the first design. A site review should split the technology, energy-recovery, hot-water, thermosyphon, and tower loops. It should then map Xylem gear and controls to each loop. Useful proof includes pump curves, yearly duty points, standby tests, seal and bearing history, water records, measured pump power, times when no one used the heat, and current supply and return temperatures. → Status: operating. Xylem says more than 55 Bell & Gossett products were selected for the Energy Systems Integration Facility at the National Renewable Energy Laboratory in Golden, Colorado. Named pump families include e-1510, Series 90, Series 80, and Series 60. The data center uses component-level warm-water cooling and transfers heat through an energy-recovery loop for reuse in laboratories and offices. NREL says the water-based system was installed in 2012 and describes a hierarchy of heat reuse, dry rejection, and cooling towers. The sources establish named equipment families at an operating facility but do not identify which pump serves each loop. Current model quantities, selected flow and head, annual pump energy, and maintenance history are unknown. Whole-site efficiency cannot be attributed solely to Xylem, and the computing platform has changed over time. A reference review should examine controls, low-load operation, redundancy, water chemistry, heat-reuse availability, and maintenance rather than copy the original pump schedule. Current drawings, trend data, and service records are needed for a like-for-like comparison. The sources provide only the installation year, so 2012-01-01 is a structured-date placeholder rather than a claimed commissioning date. NREL's current cooling description includes later heat-rejection additions and may not match the original design. Diligence should separate the technology, energy-recovery, process-hot-water, thermosyphon, and tower loops, then map Xylem equipment and controls to each. Relevant evidence includes pump curves, annual duty points, standby tests, seal and bearing history, water-treatment records, measured pump energy, periods without a heat customer, and current supply and return temperatures.
  • Novva Colorado Springs cooling-controls retrofit description: Status: operating. Siemens and Novva Data Centers say Desigo PXC controllers replaced the old controls in Colorado Springs. Holbrook Service used a flat control design. It covers four chillers, two chiller plants, towers, pumps, and fans. If one chiller fails, the other gear can respond. Siemens says the work helped Novva move away from water cooling. It ended a need for weekly service. Siemens reports no controller-related plant downtime for a year and a half. It also reports yearly savings above two million kilowatt-hours and $176,000. This is a named site and completed project. Siemens is the source. No outside audit or raw trends are public. The savings include new controls, new wires, new code, and a wider cooling change. The source omits information-technology capacity and controller count. It also omits exact PXC models, baseline dates, weather and load adjustments, project cost, and final water use. Buyers need acceptance tests, a fault plan, trend data, network design, service records, and the savings method. Siemens lists completion in 2023. The savings release is dated 22 July 2024. The structured date uses the public release date, not a claimed start date. The phrase water-free cooling needs care. The same source names chillers and cooling towers but gives no final water balance or system drawing. A reference call should ask how the live change was staged. It should ask which controls stayed local and what faults were tested. It should also test controller and network standby design. Reported zero downtime may not cover power, mechanical, or sensor events. Raw interval data and a baseline adjusted for load and weather are needed to prove the savings. → Status: operating. Siemens and Novva Data Centers say Desigo PXC controllers replaced the previous cooling controls at Novva's Colorado Springs facility. Holbrook Service arranged the controllers in a flat architecture across four chillers, two chiller plants, cooling towers, pumps, and fans so remaining equipment could respond to a chiller fault. Siemens says the retrofit supported Novva's move away from water cooling, ended prior weekly service needs, produced no controller-related central-plant downtime for a year and a half, and saved more than two million kilowatt-hours and $176,000 annually. The operator, site, equipment family, and completed project are named, but Siemens supplies the evidence. No independent audit or raw trends are available. The reported outcome includes controller replacement, rewiring, reprogrammed sequences, and a broader cooling-strategy change. Information-technology capacity, controller count, exact PXC models, baseline dates, weather and load normalization, capital cost, and final water use are not disclosed. Reference diligence needs acceptance tests, a fault matrix, trend data, network design, maintenance records, and the savings method. Siemens lists project completion in 2023, while the savings release is dated 22 July 2024; the structured date reflects the release, not commissioning. The phrase water-free cooling also needs reconciliation with references to chillers and cooling towers because no final water balance or equipment schematic is provided. A reference call should cover live cutover staging, local autonomous functions, injected faults, controller and network redundancy tests, and whether zero downtime excludes power, mechanical, or sensor events. Raw interval data and a load- and weather-normalized baseline are needed to verify the savings.
  • Illinois data-center cooling-water treatment retrofit description: Status: operating, based on a Veolia case dated 23 July 2025. Veolia says an Illinois data center had low tower cycles and high blowdown. Its old treatment program was alkaline. Veolia added a TrueSense Ready-Set-Go controller. The controller watched pH and fed sulfuric acid. Remote checks and alarms were also used. Veolia says tower cycles doubled. It reports a 50% cut in cooling-water demand. It also reports yearly savings of 12 million gallons and $150,000. The operator and city are not named. Site capacity, tower count, tower type, baseline period, final cycles, source-water chemistry, acid dose, test method, and outside review are unknown. These gaps prevent a fair adjustment for information-technology load, weather, or water quality. This is a supplier case, not an audited result. Acid feed can harm people and equipment if controls fail. Buyers need the water model and a review of wet materials. They need setpoints, interlocks, containment, alarms, trends, the savings math, and an operator contact. The release date is not the install or start date. Project timing and proven run time remain unknown. The word doubled is incomplete without start and end cycle values. It also needs seasonal water data, blowdown meter records, cooling load, and weather. A reference check should cover acid storage and transfer, secondary containment, material fit, probe checks, dose proof, pH interlocks, failed sensors, stuck pumps, and alarm escalation. Operator input, water bills, sewer bills, lab results, rust and deposit trends, and savings adjusted for rejected heat would make the claims much stronger. → Status: operating, based on a Veolia case dated 23 July 2025. Veolia says an Illinois data center had low cooling-tower cycles of concentration and high blowdown. Its old treatment program was alkaline. Veolia added a TrueSense Ready-Set-Go controller to measure pH and feed sulfuric acid. Remote checks and alarms were also used. Veolia reports that tower cycles doubled and cooling-water demand fell by 50%. It also reports annual savings of 12 million gallons and $150,000. The operator and city are not named. Site capacity, tower count and type, baseline period, final cycles, source-water chemistry, acid dose, test method, and outside review are unknown. Those gaps prevent a fair adjustment for information-technology load, weather, or water quality. The result is a supplier case, not an audited benchmark. Acid feed creates safety and corrosion risks if controls fail. Reference checks need the water model and a wetted-materials review. They also need setpoints, interlocks, containment, alarms, trends, savings calculations, and an operator contact. The release date is not the install or start date. Project timing and proven run time remain unknown. The word doubled lacks starting and final cycle values. Seasonal source-water data, blowdown-meter records, cooling load, and weather are also missing. A site call should cover acid storage and transfer, secondary containment, material compatibility, probe calibration, proof of dosing, pH interlocks, failed sensors, stuck pumps, and alarm escalation. Operator confirmation, water and sewer bills, laboratory results, corrosion and deposition trends, and savings adjusted for rejected heat are needed to support the claims.
• Oct. 4, 2026

  Research expansion published: 90 records

  • Smardt New York-region data-center chiller replacement description: Status: operating, based on Smardt's report of early commissioning with stable operation and efficiency. The unnamed New York-region operator required two replacement AD120 water-cooled chillers, each configured with three TT350 oil-free magnetic-bearing compressors, while preserving a fixed commissioning schedule. Smardt says one unit was damaged in third-party transit and that it reorganized production and sourcing to deliver a replacement in three months; this is useful evidence of project response, but the account remains supplier-authored and the operator is not named. Public material does not disclose cooling capacity, water temperatures, refrigerant, measured efficiency, redundancy at design load, acceptance criteria, rack density, compute platform, final commissioning date, or long-term operating results, so buyers should treat it as a delivery reference and request an operator contact and acceptance data. The record is strongest as evidence of factory and supply-chain response after transport damage, not as a thermal-performance benchmark. “Stable efficiency” is not accompanied by a rating method, measured kilowatts per ton, load point, condenser-water condition, observation period, or independent witness. Due diligence should ask whether both chillers reached final acceptance, whether the damaged machine was rebuilt or replaced, what commissioning defects remained, and whether the three-month statement ran from authorization to factory shipment or site operation. A reference call should also cover packaging and transport controls, spare compressors and electronics, startup staffing, alarm and controls integration, restart tests, maintenance access, and service response. For design comparison, request the exact AD120 selection, certified duty, electrical demand, refrigerant, dimensions, operating hours, availability, and capacity after one compressor, one chiller, or a supporting plant component is unavailable. → Status: operating. That status rests on Smardt's report of early startup with stable operation and efficiency. The unnamed New York-area operator needed two AD120 water-cooled chillers. Each unit had three TT350 oil-free magnetic-bearing compressors. The project had a fixed startup date. Smardt says a third-party carrier badly damaged one unit. Smardt then changed factory plans and supply work. It says a replacement arrived within three months. This supports the supplier's response claim. The account still comes from the supplier, and the operator is not named. Public sources omit cooling duty and water temperatures. They also omit refrigerant, measured efficiency, design-load backup, test rules, rack density, computing platform, final startup date, and long-term results. Treat this as a delivery reference, not a heat-duty benchmark. Ask for an operator contact and test data. “Stable efficiency” has no stated rating method. No kilowatts-per-ton value, load point, condenser-water condition, test period, or outside witness is given. Buyers should ask if both chillers passed final tests. They should ask if the damaged unit was rebuilt or replaced. Open startup defects should be disclosed. The three-month period needs a clear start and end date. A reference call should cover packing, shipping controls, spare compressors, electronics, startup staff, alarms, control links, restart tests, service space, and response time. Design comparison needs the exact AD120 selection, certified duty, power, refrigerant, size, hours, availability, and output after loss of a compressor, chiller, or support system.
  • BAC 160 MW HPC HXV cooling project description: Status: operating, based on BAC's statement that the installed HXV system delivered water temperatures meeting the customer's specification and supported the information-technology load. The company says the 160 MW operating-power high-performance-computing customer previously used open cooling towers with water-cooled chillers and selected HXV hybrid coolers to combine dry and evaporative operation without chillers. BAC's case study reports comparative estimates and savings, but the operator and location are withheld and no independent meter data, equipment count, water temperatures, weather file, baseline boundary, commissioning report, or observation period is published. The record therefore establishes a real supplier-documented deployment and architecture, not independently verified PUE, WUE, energy, water, or cost performance; procurement teams should request the underlying model, measured post-commissioning data, and a customer reference. The stated 160 MW is customer operating power, not a published tower thermal duty, and should not be used to infer equipment quantity or unit size. The case also does not say whether the HXV system serves the entire campus, one phase, or a defined subset of load. Qualification should request the exact baseline and proposed system boundaries, hourly weather and load assumptions, dry, evaporative, and combined mode hours, makeup and blowdown, fan and pump energy, leaving-water temperatures, reserve criterion, and hot-day capacity. Operators should ask how the plant responds to water restrictions, poor water quality, plume conditions, freezing weather, a fan or spray-pump failure, and loss of common controls. Acceptance records and at least one year of utility and maintenance data would be more decision-useful than supplier percentage claims alone. → Status: operating. Baltimore Aircoil Company says the installed HXV system met the customer's water-temperature target. It also says the system supported the information-technology load. The customer had 160 MW of operating power for high-performance computing. It had used open towers with water-cooled chillers. It chose HXV hybrid coolers for dry and evaporative operation without chillers. The supplier case gives estimates and claimed savings. It withholds the operator and location. It gives no outside meter data, unit count, water temperatures, weather file, baseline boundary, startup report, or test period. The record proves a supplier-documented site and design. It does not independently prove power usage effectiveness or water usage effectiveness. It also does not prove energy, water, or cost results. Buyers should ask for the model, measured data after startup, and a customer contact. The 160 MW figure is customer operating power. It is not published tower heat duty. Do not infer unit count or size from it. The case does not say if HXV serves the whole campus, one phase, or part of the load. Qualification needs exact old and new system boundaries. It needs hourly weather, load, and mode hours. Ask for makeup, blowdown, fan energy, pump energy, outlet temperatures, reserve rule, and hot-day output. Ask about water limits, poor water, plume, freezing, fan faults, spray-pump faults, and loss of shared controls. Final test records and one year of utility and service data would be more useful than percentage claims alone.
  • Güntner Zhangjiakou immersion-cooling heat rejection description: Status: operating. Güntner says a data center run by an unnamed large Chinese e-commerce company has used two-phase immersion cooling in Zhangjiakou since 2017, with four V-shape VARIO dry coolers and hydroBLU adiabatic assistance providing 1,060 kW aggregate heat-rejection capacity. The architecture condenses vapor inside the immersion chambers and transfers that heat to the outdoor coolers; water is used for adiabatic assistance only when ambient conditions require it, according to the supplier. Güntner also reports large energy and cost savings versus conventional systems, but does not name the operator, publish the baseline, equipment schedule, water use, weather normalization, raw metering, rack density, compute platform, availability, or independent verification. Buyers can use the case to validate architecture and scale, but should request site contacts and measured annual fan, pump, and water data. The evidence does not establish whether 1,060 kW is installed nameplate, design-day duty, or measured rejected heat, nor does it disclose the fluid temperatures, glycol, approach to ambient, fan power, or reserve margin. Four units may provide staging, but their N, N+1, or other resilience role is not stated. A comparable-project review should ask for the selected model and fan configuration, hydroBLU activation threshold, annual and peak water use, pad maintenance and treatment, dry-only output, winter sequence, alarm history, coil cleaning, and capacity after one cooler or common pump loop is lost. Immersion-system due diligence also needs the intermediate exchanger and pump boundary, because outdoor cooler performance alone does not validate chamber condensation, dielectric-fluid compatibility, server service procedures, or end-to-end cooling availability. → Status: operating. Güntner says an unnamed large Chinese online seller runs the data center. The Zhangjiakou site has used two-phase immersion cooling since 2017. Four V-shape VARIO dry coolers serve it. hydroBLU adiabatic help is also fitted. Güntner states 1,060 kW of total heat-rejection duty. Vapor condenses inside the immersion chambers. A loop then moves that heat to the outdoor coolers. The supplier says water is used for help only when weather requires it. Güntner claims large energy and cost savings over normal systems. It does not name the operator or give the baseline. It omits the unit schedule, water use, weather adjustment, raw meter data, rack density, computing platform, availability, and outside review. Buyers can use the case to check the design path and scale. They should still request a site contact and yearly fan, pump, and water data. The evidence does not define the 1,060 kW figure. It may be installed duty, design-day duty, or measured heat. Fluid temperatures, glycol, air approach, fan power, and reserve are also unknown. Four units may allow staging. Their backup role is not stated. A similar-site review should ask for model and fan details. It needs the hydroBLU start point, yearly and peak water, pad service, treatment, dry-only output, winter order, alarm history, cleaning, and output after one cooler or shared pump loop fails. Review the middle heat exchanger and pumps too. Outdoor duty alone does not prove chamber condensation, fluid fit, server service, or full cooling availability.
  • Infosys Hyderabad data-center heat exchangers description: Status: operating. SWEP reports that Infosys and project design lead Schneider Electric developed several data-center cooling projects between 2016 and 2020 using B439 and B649 brazed plate heat exchangers to separate primary cooling sources from the secondary loop. At the Hyderabad campus, multiple B649 units have 6-inch ports and a reported flow of 350 m³/h each; SWEP says the units were running without problems when the case was published. The named operator, design partner, product models, hydraulic role, period, and per-unit flow make this a useful deployment reference. Evidence remains supplier-authored, however, and public material does not disclose unit count, thermal megawatts, temperatures, pressure drop, rack density, compute platform, PUE, measured savings, maintenance history, or independent acceptance data. Flow is not thermal capacity without inlet and outlet temperatures and fluid properties, so the reported 350 m³/h must not be converted into megawatts from assumptions. The phrase “without any problems” also lacks an operating period, availability definition, alarm log, leakage history, fouling trend, or maintenance record. Buyer reference questions should cover exact article numbers and plate counts, water chemistry on both sides, approach temperature, pressure loss, pump energy, strainer and filtration practice, cleaning frequency, isolation and bypass, spare units, and the method for detecting internal cross-contamination. The design team should ask how parallel exchangers are balanced, what capacity remains during cleaning or replacement, whether performance was field verified, and which measured savings Infosys attributed specifically to the exchangers rather than to the wider cooling design. → Status: operating. SWEP says Infosys and design lead Schneider Electric built several data-center cooling projects. The work ran from 2016 through 2020. B439 and B649 brazed plate heat exchangers separate the main cooling source from the inner loop. At the Hyderabad campus, several B649 units have 6-inch ports. SWEP reports 350 cubic meters per hour through each unit. It says the units were running without problems when the case appeared. The named operator, design partner, models, loop role, period, and unit flow make this a useful reference. The evidence still comes from the supplier. Public sources omit unit count, heat duty, temperatures, pressure loss, rack density, computing platform, power usage effectiveness, measured savings, service history, and outside test data. Flow is not heat duty without temperatures and fluid data. Do not turn 350 cubic meters per hour into megawatts by assumption. “Without problems” also lacks a time period and uptime measure. There is no alarm log, leak history, fouling trend, or service record. Reference questions should cover exact articles and plate counts. Ask for water chemistry on both sides, approach, pressure loss, pump energy, strainers, filters, cleaning rate, isolation, bypass, spares, and cross-leak detection. Ask how parallel units share flow and what duty remains during cleaning or replacement. Confirm field tests. Separate exchanger savings from results of the wider cooling design.
  • MITA Barcelona data-processing center cooling towers description: Status: operating, based on MITA Group's 2023 case study describing the completed supply and the towers' role in maintaining equipment performance. TORRAVAL Cooling, a MITA Group company, supplied sixteen CTFP 2436 forced-draft open-circuit towers with laminar fill for a large Barcelona data-processing center operated for an international information-technology company. The source states a 20 MW required load expandable to 40 MW plus 5 MW of emergency supply; it does not explain whether those values are information-technology, electrical, or thermal capacity, so this record preserves the source wording rather than converting them into cooling duty. The rooftop towers were arranged in two facing rows to use limited space and support acoustic control, with maintenance walkways and ladders included. Public evidence does not name the operator or installer, disclose tower water temperatures and flow, measured energy or water use, redundancy, commissioning tests, rack density, compute platform, availability, or an operator-authored account, so buyers should request the approved schedule, acceptance data, annual operating records, and a reference contact. The case's separate 15 kW statement for each tower is not treated as heat-rejection capacity here because its meaning is unclear against the cited multi-megawatt requirement; it may describe installed motor power, but the source does not say. That ambiguity should be resolved from the equipment schedule rather than inferred. Qualification should request design wet bulb, hot- and cold-water temperatures, cell flow, fan power, drift, evaporation and blowdown, cycles of concentration, sound measurements, plume review, chemical treatment, hygiene plan, rooftop vibration and structural reactions, and capacity with one cell or common pump unavailable. It should also confirm whether the expansion allowance was physically installed, reserved in the layout, or only a future design intention. → Status: operating. This status rests on MITA Group's 2023 case. The case describes a completed supply and says the towers support equipment performance. TORRAVAL Cooling is part of MITA Group. It supplied sixteen CTFP 2436 forced-draft open-circuit towers. The units have laminar fill. They serve a large Barcelona data-processing center for an international information-technology company. The source gives a required load of 20 MW. It says the site can grow to 40 MW and has 5 MW of emergency supply. The source does not define these values as computing, electrical, or heat duty. This record keeps the source wording and does not convert it. Two facing rows fit the towers on a limited roof. The layout also supported sound control. Walkways and ladders were included. Public evidence withholds the operator and installer. It omits water temperatures, flow, measured power, measured water, backup, startup tests, rack density, computing platform, uptime, and an operator account. Buyers should request the approved schedule, test data, yearly records, and a reference contact. The case also states 15 kW for each tower. That figure is not used here as heat duty. Its meaning conflicts with the multi-megawatt need and may be motor power, but the source is silent. Resolve it from the schedule. Qualification needs design wet bulb, water temperatures, cell flow, fan power, drift, evaporation, blowdown, concentration cycles, sound, plume, treatment, hygiene, roof vibration, structural loads, and output after a cell or common pump fails. Confirm whether future growth was installed, reserved, or only planned.
  • West Cambridge Data Centre ColdLogik cooling description: Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later increased cabinet density from 30 to 44 kW and raised data-hall capacity from 900 kW to 1.2 MW. Those figures come from the cooling supplier rather than an independently published acceptance test, but the university confirms the operating facility and continued research-computing role. Public sources do not disclose the current door count, inlet-water temperature, annual cooling energy, redundancy arrangement or whether every hall uses the same design. Procurement teams should use the reference to ask for operator contacts and measured seasonal data. The university history confirms relocation and continued use of the facility but does not independently attribute the later capacity increase to a particular ColdLogik model. The 1.2 MW figure should therefore remain a supplier-reported hall value, not a measured IT load or proof that all cabinets operate at 44 kW. A reference interview should establish the installed models and quantities, typical and peak rack loads, water temperatures, fan energy, leak events, maintenance burden and changes made after commissioning. Buyers should also ask whether capacity growth required plant, piping or controls changes beyond the rear doors, and how cooling continuity is maintained during door service. Annual trend data with a defined electrical boundary would be needed before using the reference to predict PUE or operating cost. → Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later raised cabinet density from 30 to 44 kW. It also reports that data-hall capacity rose from 900 kW to 1.2 MW. Those figures come from the cooling supplier, not an independent acceptance test. The university does confirm the live site and its ongoing research-computing role. Public sources do not give the current door count or inlet-water temperature. They also omit yearly cooling energy, backup design and whether each hall uses the same system. Buyers should use the reference to seek operator contacts and measured seasonal data. The university history confirms the move and continued use of the site. It does not link the later capacity gain to a given ColdLogik model. The 1.2 MW figure must remain a supplier-reported hall value. It is not a measured IT load or proof that all cabinets run at 44 kW. An operator interview should confirm installed models and counts. It should cover normal and peak rack loads, water temperatures and fan energy. Ask about leaks, service work and changes made after start-up. Buyers should also ask if the growth required plant, pipe or control changes beyond the rear doors. They should learn how cooling stays live during door service. Annual trend data needs a clear electrical boundary. Without it, the reference cannot predict power usage effectiveness (PUE) or operating cost.
  • NorthC Eindhoven MIXIT chip-cooling loop description: Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes; a customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The public record does not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. The June 2022 date in this record represents the reported mid-2022 delivery period; the cited sources do not publish a precise commissioning day. MIXIT's exact configuration, valve size, sensors, controlled variable and relationship to pumps or a CDU are also undisclosed. Procurement teams should ask NorthC or Hamer for a loop diagram, control sequence, temperature and flow trends, alarm history, maintenance interventions and behavior during server or network changes. They should distinguish heat made available to the campus ring from heat actually reused by a customer and quantify backup rejection when campus demand is absent. Evidence that would strengthen the record includes a named server platform, installed kilowatts, measured pump and cooling energy, water temperatures, uptime and acceptance criteria. Until then, capacity and rack density correctly remain not publicly disclosed. → Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes; a customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The public record does not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. The June 2022 date in this record represents the reported mid-2022 delivery period; the cited sources do not publish a precise commissioning day. MIXIT's exact configuration, valve size, sensors, controlled variable and relationship to pumps or a coolant distribution unit (CDU) are also undisclosed. Procurement teams should ask NorthC or Hamer for a loop diagram, control sequence, temperature and flow trends, alarm history, maintenance interventions and behavior during server or network changes. They should distinguish heat made available to the campus ring from heat actually reused by a customer and quantify backup rejection when campus demand is absent. Evidence that would strengthen the record includes a named server platform, installed kilowatts, measured pump and cooling energy, water temperatures, uptime and acceptance criteria. Until then, capacity and rack density correctly remain not publicly disclosed.
  • Castrol–Hypertec Pangbourne immersion lab description: Status: pilot. Castrol announced on 20 July 2023 that immersion systems were installed and fully functional at its Pangbourne headquarters. The research setup combines Castrol fluids, Hypertec immersion-cooled server expertise, and Submer SmartPod and MicroPod tanks. The partners intended to test fluids, servers, and integrated cooling behavior. This is a real, named physical deployment at Castrol's site, but it is a supplier development laboratory rather than a production data center. The announcement does not disclose IT load, tank count, server models, fluid formulation used in each test, water temperatures, heat-rejection equipment, test protocol, measured efficiency, uptime, or customer acceptance results. It therefore supports interoperability research, not a capacity or savings claim. A buyer evaluating the reference should ask for the exact bill of materials, test duration and duty cycle, material inspections, fluid analyses, failure tests, and changes made after the program. Evidence that would strengthen the record includes published long-duration data, an operator acceptance report, or a commercial site using the qualified combination. The announcement also does not say which Castrol fluid was in each tank, whether tests used production batches, or whether Submer's later DC 20 approval came from this installation. Those links should not be inferred. Due diligence should request baseline and end-of-test fluid properties, server and component inventory, immersed-material exposure time, thermal load profile, pump and heat-exchanger data, maintenance interventions, spills or contamination events, and forensic inspection results. A commercial buyer should separately verify fire strategy, occupational procedures, fluid inventory, lifting and draining, spare capacity, and server-warranty ownership. Because the public source gives no later operating update, continued use and the present configuration are unknown; a current site reference or dated test report is needed before treating Pangbourne as an active qualification facility. → Status: pilot. On 20 July 2023, Castrol said its Pangbourne immersion systems were installed and working. The research setup joins Castrol fluids with Hypertec server skills. It also uses Submer SmartPod and MicroPod tanks. The partners planned to test fluids, servers, and the full cooling setup. This is a real site with named partners. It is a supplier lab, not a production data center. The release gives no information-technology load or tank count. It omits server models and the fluid used in each test. It also omits water temperatures, heat-rejection gear, test steps, measured efficiency, uptime, and customer acceptance. The site proves joint research, not capacity or savings. Buyers should request the full bill of materials. They need test length, duty, material checks, fluid results, fault tests, and later design changes. Long-run data would strengthen the record. So would an operator report or a commercial site with the same mix. The release does not say which Castrol fluid was in each tank. It does not say if tests used sale batches. It also does not tie Submer's later DC 20 approval to this lab. Those links cannot be assumed. Buyers should ask for fluid values before and after each test. They need the server list, time in fluid, heat profile, pump and heat-exchanger data, service work, spills, dirt events, and final inspection. A commercial project also needs fire rules, worker steps, fluid stock, lift and drain plans, spare capacity, and a warranty owner. No later public update confirms current use or setup. A fresh site reference or dated report is needed before calling Pangbourne active.
  • Valvoline–Iceotope 18-month fluid validation description: Status: pilot. Valvoline's December 2025 case study documents an 18-month in-application test of HTF-DE1 fluid in an Iceotope MicroDC system with HPE DL380 servers and NVIDIA A40 graphics processors. The reported load was 5.2 kW. Valvoline says viscosity remained stable, dielectric performance stayed above the Open Compute Project minimum, and forensic inspection found no corrosion or material damage. The named hardware, duration, and post-test inspection make this stronger than a generic compatibility assertion. Important limits remain: the test location is not disclosed, 5.2 kW is not representative of a full high-density rack, raw measurements and uncertainty are not published, and the report comes from the fluid supplier. It does not establish compatibility with other server generations, plastics, cables, storage devices, or operating temperatures. Buyers should request the full protocol, starting and ending fluid analyses, sample history, hardware inspection criteria, excursions, maintenance events, and written Iceotope approval for the exact production system. Independent replication or a named commercial operator would further strengthen the evidence. The source does not state whether the 5.2 kW value is average, maximum, or nameplate load, how continuously the hardware ran, what inlet and outlet temperatures applied, or whether fluid was filtered or topped up. It also does not publish individual dielectric and viscosity readings, laboratory methods, uncertainty, or photographs and measurements from the forensic examination. Procurement teams should ask whether the exact HPE and NVIDIA configurations retained their warranties and whether Iceotope's approval applies to all MicroDC revisions. A production pilot should reproduce the intended materials and temperature range, include representative service events, preserve sealed baseline samples, define pass and fail limits before testing, and assign authority for continued operation after any out-of-limit sample. → Status: pilot. A Valvoline case from December 2025 covers an 18-month test of HTF-DE1 fluid. The test used an Iceotope MicroDC system. It had HPE DL380 servers and NVIDIA A40 graphics processors. The stated load was 5.2 kW. Valvoline says viscosity stayed steady. It says dielectric strength stayed above the Open Compute Project floor. A final inspection found no rust or material harm. The named gear, long test, and later inspection add weight. Limits remain. The site is not named. A 5.2 kW test does not represent a dense full rack. Raw values and uncertainty are not public. The fluid supplier wrote the report. The test does not prove fit with other servers, plastics, cables, drives, or heat ranges. Buyers need the full test plan. They should request fluid results from start to finish, sample history, inspection limits, faults, service events, and written Iceotope approval for the exact system. A repeat test or named user would add confidence. The source does not say if 5.2 kW is average, peak, or nameplate load. It does not say how often the servers ran. Inlet and outlet temperatures are unknown. Filter use and top-ups are also unknown. Individual electrical and viscosity readings are not public. Nor are lab methods, uncertainty, or detailed inspection records. Buyers should confirm that HPE and NVIDIA warranties stayed valid. They should ask which MicroDC versions Iceotope approves. A production test must use the planned materials and heat range. It should include real service work, sealed baseline samples, preset pass limits, and a named owner for action after a bad sample.
  • NREL ESIF warm-water cooling loop description: Status: operating. Xylem's case study says more than 55 Bell & Gossett products, including e-1510, Series 90, Series 80, and Series 60 pumps, were selected for the National Renewable Energy Laboratory's Energy Systems Integration Facility in Golden, Colorado. The data center uses component-level warm-water cooling for supercomputing, transfers heat through an energy-recovery loop, and reuses available heat in laboratories and offices. NREL confirms that the water-based system was installed in 2012 and describes the operating hierarchy for heat reuse, dry rejection, and cooling towers. This record establishes named equipment families and an operating facility, but it does not disclose which pump serves each loop, current pump quantities by model, selected flow and head, annual pump energy, maintenance history, or attribution of whole-site efficiency to Xylem equipment. The computing platform has also changed over time. Buyers should use the reference to examine controls, low-load operation, redundancy, water chemistry, heat-reuse availability, and maintenance, not to copy a pump schedule. Current drawings, trend data, and service records would be needed for a like-for-like benchmark. The sources identify 2012 as the installation year but do not provide the exact day; the structured 2012-01-01 date is therefore a normalization placeholder, not a claimed commissioning date. NREL's current cooling description also reflects later heat-rejection additions and should not be assumed to match the original configuration. Reference checks should separate the technology loop, energy-recovery loop, process-hot-water interface, thermosyphon, and tower loop, then identify the Xylem equipment and controls in each. Useful evidence would include selected pump curves, annual operating points, standby testing, seal and bearing history, water-treatment records, measured pump energy, periods when reusable heat had no customer, and the present supercomputer's actual supply and return temperatures. → Status: operating. Xylem says more than 55 Bell & Gossett products went into the Energy Systems Integration Facility. The site is at the National Renewable Energy Laboratory in Golden, Colorado. Named pump families include e-1510, Series 90, Series 80, and Series 60. The data center uses warm water close to the computer parts. It sends heat through an energy-recovery loop. Useful heat warms labs and offices. NREL says the water system was installed in 2012. It also describes an order for heat reuse, dry rejection, and cooling towers. The sources prove named product families at a live site. They do not show which pump serves each loop. Current model counts, flow, head, yearly pump power, and service history are unknown. Whole-site efficiency cannot be assigned only to Xylem. The computers have also changed over time. Buyers should study controls, low-load use, standby design, water care, heat demand, and service. They should not copy the old pump schedule. A fair comparison needs current drawings, trends, and service records. The source gives only the year 2012. It gives no exact day. The structured date 2012-01-01 is a format placeholder, not a claimed start date. NREL's current description includes later heat-rejection additions. It may not match the first design. A site review should split the technology, energy-recovery, hot-water, thermosyphon, and tower loops. It should then map Xylem gear and controls to each loop. Useful proof includes pump curves, yearly duty points, standby tests, seal and bearing history, water records, measured pump power, times when no one used the heat, and current supply and return temperatures.
  • Novva Colorado Springs cooling-controls retrofit description: Status: operating. Siemens and Novva Data Centers report that Desigo PXC controllers replaced the prior cooling-control system at Novva's Colorado Springs facility. Holbrook Service arranged the controllers in a flat architecture across four chillers, two chiller plants, cooling towers, pumps, and fans so remaining equipment could respond if one chiller faulted. Siemens says the retrofit supported Novva's transition away from water cooling, eliminated prior weekly service needs, delivered no controller-related central-plant downtime for a year and a half, and saved more than two million kilowatt-hours and $176,000 annually. This is a named operator, site, equipment family, and completed project. The limits are important: Siemens is the source, no independent audit or raw trends are published, and the savings include controller replacement, rewiring, reprogrammed sequences, and a broader cooling-strategy change. The source does not disclose IT capacity, controller count, exact PXC models, baseline dates, weather or load normalization, capital cost, or final water use. Buyers should request Novva's acceptance tests, failure matrix, trend data, network design, maintenance record, and savings method before using the figures in a business case. Siemens' reference page lists project completion in 2023, while the public savings announcement is dated 22 July 2024; the announcedAt field uses the documented announcement date rather than implying that commissioning occurred then. The phrase water-free cooling should also be checked against the source's simultaneous description of chillers and cooling towers, because the public materials do not provide a final water balance or equipment schematic. Reference diligence should ask how the live cutover was staged, which functions remained locally autonomous, what faults were injected, how controller and network redundancy were tested, and whether the reported zero downtime excludes upstream power, mechanical, or sensor events. Raw interval data and a normalized baseline are needed to validate the energy and cost claims. → Status: operating. Siemens and Novva Data Centers say Desigo PXC controllers replaced the old controls in Colorado Springs. Holbrook Service used a flat control design. It covers four chillers, two chiller plants, towers, pumps, and fans. If one chiller fails, the other gear can respond. Siemens says the work helped Novva move away from water cooling. It ended a need for weekly service. Siemens reports no controller-related plant downtime for a year and a half. It also reports yearly savings above two million kilowatt-hours and $176,000. This is a named site and completed project. Siemens is the source. No outside audit or raw trends are public. The savings include new controls, new wires, new code, and a wider cooling change. The source omits information-technology capacity and controller count. It also omits exact PXC models, baseline dates, weather and load adjustments, project cost, and final water use. Buyers need acceptance tests, a fault plan, trend data, network design, service records, and the savings method. Siemens lists completion in 2023. The savings release is dated 22 July 2024. The structured date uses the public release date, not a claimed start date. The phrase water-free cooling needs care. The same source names chillers and cooling towers but gives no final water balance or system drawing. A reference call should ask how the live change was staged. It should ask which controls stayed local and what faults were tested. It should also test controller and network standby design. Reported zero downtime may not cover power, mechanical, or sensor events. Raw interval data and a baseline adjusted for load and weather are needed to prove the savings.
  • Illinois data-center cooling-water treatment retrofit description: Status: operating, based on a supplier case summarized by Veolia on 23 July 2025. Veolia says an Illinois data center had low cooling-tower cycles of concentration and high blowdown under its prior alkaline treatment program. Veolia installed a TrueSense Ready-Set-Go controller to monitor pH and feed sulfuric acid, together with remote monitoring and alarms. The company reports that tower cycles doubled, cooling-water demand fell by 50%, and annual savings reached 12 million gallons and $150,000. The operator, city, facility capacity, number and type of towers, baseline period, final cycles, source-water chemistry, acid dose, measurement method, and independent verification are not disclosed. Those omissions prevent normalization by IT load, weather, or water quality and make this a supplier-reported reference rather than an audited benchmark. Acid-feed systems also create safety and corrosion risks if controls fail. Buyers should request the water model, materials review, setpoints, interlocks, containment, alarm response, trend data, calculation method, and operator reference before using the savings in a business case. The source publication date is not identified as the installation or commissioning date, so the deployment schedule and length of demonstrated operation remain unknown. “Doubled” tower cycles is also incomplete without the starting and ending values, seasonal source-water range, blowdown meter data, cooling load, and weather. Reference diligence should verify acid-storage and transfer safeguards, secondary containment, compatible materials, probe calibration, dosing proof, high- and low-pH interlocks, response after a failed sensor or stuck pump, and the remote alarm escalation record. An operator contact, water and sewer invoices, laboratory results, corrosion and deposition trends, and a calculation normalized by heat rejection would materially strengthen both the operating label and the savings claims. → Status: operating, based on a Veolia case dated 23 July 2025. Veolia says an Illinois data center had low tower cycles and high blowdown. Its old treatment program was alkaline. Veolia added a TrueSense Ready-Set-Go controller. The controller watched pH and fed sulfuric acid. Remote checks and alarms were also used. Veolia says tower cycles doubled. It reports a 50% cut in cooling-water demand. It also reports yearly savings of 12 million gallons and $150,000. The operator and city are not named. Site capacity, tower count, tower type, baseline period, final cycles, source-water chemistry, acid dose, test method, and outside review are unknown. These gaps prevent a fair adjustment for information-technology load, weather, or water quality. This is a supplier case, not an audited result. Acid feed can harm people and equipment if controls fail. Buyers need the water model and a review of wet materials. They need setpoints, interlocks, containment, alarms, trends, the savings math, and an operator contact. The release date is not the install or start date. Project timing and proven run time remain unknown. The word doubled is incomplete without start and end cycle values. It also needs seasonal water data, blowdown meter records, cooling load, and weather. A reference check should cover acid storage and transfer, secondary containment, material fit, probe checks, dose proof, pH interlocks, failed sensors, stuck pumps, and alarm escalation. Operator input, water bills, sewer bills, lab results, rust and deposit trends, and savings adjusted for rejected heat would make the claims much stronger.
• Oct. 4, 2026

  Research expansion published: 90 records

  • Smardt New York-region data-center chiller replacement description: Status: operating, based on Smardt's report of early commissioning with stable operation and efficiency. The unnamed New York-region operator required two replacement AD120 water-cooled chillers, each configured with three TT350 oil-free magnetic-bearing compressors, while preserving a fixed commissioning schedule. Smardt says one unit was damaged in third-party transit and that it reorganized production and sourcing to deliver a replacement in three months; this is useful evidence of project response, but the account remains supplier-authored and the operator is not named. Public material does not disclose cooling capacity, water temperatures, refrigerant, measured efficiency, redundancy at design load, acceptance criteria, rack density, compute platform, final commissioning date, or long-term operating results, so buyers should treat it as a delivery reference and request an operator contact and acceptance data. → Status: operating, based on Smardt's report of early commissioning with stable operation and efficiency. The unnamed New York-region operator required two replacement AD120 water-cooled chillers, each configured with three TT350 oil-free magnetic-bearing compressors, while preserving a fixed commissioning schedule. Smardt says one unit was damaged in third-party transit and that it reorganized production and sourcing to deliver a replacement in three months; this is useful evidence of project response, but the account remains supplier-authored and the operator is not named. Public material does not disclose cooling capacity, water temperatures, refrigerant, measured efficiency, redundancy at design load, acceptance criteria, rack density, compute platform, final commissioning date, or long-term operating results, so buyers should treat it as a delivery reference and request an operator contact and acceptance data. The record is strongest as evidence of factory and supply-chain response after transport damage, not as a thermal-performance benchmark. “Stable efficiency” is not accompanied by a rating method, measured kilowatts per ton, load point, condenser-water condition, observation period, or independent witness. Due diligence should ask whether both chillers reached final acceptance, whether the damaged machine was rebuilt or replaced, what commissioning defects remained, and whether the three-month statement ran from authorization to factory shipment or site operation. A reference call should also cover packaging and transport controls, spare compressors and electronics, startup staffing, alarm and controls integration, restart tests, maintenance access, and service response. For design comparison, request the exact AD120 selection, certified duty, electrical demand, refrigerant, dimensions, operating hours, availability, and capacity after one compressor, one chiller, or a supporting plant component is unavailable.
  • BAC 160 MW HPC HXV cooling project description: Status: operating, based on BAC's statement that the installed HXV system delivered water temperatures meeting the customer's specification and supported the information-technology load. The company says the 160 MW operating-power high-performance-computing customer previously used open cooling towers with water-cooled chillers and selected HXV hybrid coolers to combine dry and evaporative operation without chillers. BAC's case study reports comparative estimates and savings, but the operator and location are withheld and no independent meter data, equipment count, water temperatures, weather file, baseline boundary, commissioning report, or observation period is published. The record therefore establishes a real supplier-documented deployment and architecture, not independently verified PUE, WUE, energy, water, or cost performance; procurement teams should request the underlying model, measured post-commissioning data, and a customer reference. → Status: operating, based on BAC's statement that the installed HXV system delivered water temperatures meeting the customer's specification and supported the information-technology load. The company says the 160 MW operating-power high-performance-computing customer previously used open cooling towers with water-cooled chillers and selected HXV hybrid coolers to combine dry and evaporative operation without chillers. BAC's case study reports comparative estimates and savings, but the operator and location are withheld and no independent meter data, equipment count, water temperatures, weather file, baseline boundary, commissioning report, or observation period is published. The record therefore establishes a real supplier-documented deployment and architecture, not independently verified PUE, WUE, energy, water, or cost performance; procurement teams should request the underlying model, measured post-commissioning data, and a customer reference. The stated 160 MW is customer operating power, not a published tower thermal duty, and should not be used to infer equipment quantity or unit size. The case also does not say whether the HXV system serves the entire campus, one phase, or a defined subset of load. Qualification should request the exact baseline and proposed system boundaries, hourly weather and load assumptions, dry, evaporative, and combined mode hours, makeup and blowdown, fan and pump energy, leaving-water temperatures, reserve criterion, and hot-day capacity. Operators should ask how the plant responds to water restrictions, poor water quality, plume conditions, freezing weather, a fan or spray-pump failure, and loss of common controls. Acceptance records and at least one year of utility and maintenance data would be more decision-useful than supplier percentage claims alone.
  • Güntner Zhangjiakou immersion-cooling heat rejection description: Status: operating. Güntner says a data center run by an unnamed large Chinese e-commerce company has used two-phase immersion cooling in Zhangjiakou since 2017, with four V-shape VARIO dry coolers and hydroBLU adiabatic assistance providing 1,060 kW aggregate heat-rejection capacity. The architecture condenses vapor inside the immersion chambers and transfers that heat to the outdoor coolers; water is used for adiabatic assistance only when ambient conditions require it, according to the supplier. Güntner also reports large energy and cost savings versus conventional systems, but does not name the operator, publish the baseline, equipment schedule, water use, weather normalization, raw metering, rack density, compute platform, availability, or independent verification. Buyers can use the case to validate architecture and scale, but should request site contacts and measured annual fan, pump, and water data. → Status: operating. Güntner says a data center run by an unnamed large Chinese e-commerce company has used two-phase immersion cooling in Zhangjiakou since 2017, with four V-shape VARIO dry coolers and hydroBLU adiabatic assistance providing 1,060 kW aggregate heat-rejection capacity. The architecture condenses vapor inside the immersion chambers and transfers that heat to the outdoor coolers; water is used for adiabatic assistance only when ambient conditions require it, according to the supplier. Güntner also reports large energy and cost savings versus conventional systems, but does not name the operator, publish the baseline, equipment schedule, water use, weather normalization, raw metering, rack density, compute platform, availability, or independent verification. Buyers can use the case to validate architecture and scale, but should request site contacts and measured annual fan, pump, and water data. The evidence does not establish whether 1,060 kW is installed nameplate, design-day duty, or measured rejected heat, nor does it disclose the fluid temperatures, glycol, approach to ambient, fan power, or reserve margin. Four units may provide staging, but their N, N+1, or other resilience role is not stated. A comparable-project review should ask for the selected model and fan configuration, hydroBLU activation threshold, annual and peak water use, pad maintenance and treatment, dry-only output, winter sequence, alarm history, coil cleaning, and capacity after one cooler or common pump loop is lost. Immersion-system due diligence also needs the intermediate exchanger and pump boundary, because outdoor cooler performance alone does not validate chamber condensation, dielectric-fluid compatibility, server service procedures, or end-to-end cooling availability.
  • Infosys Hyderabad data-center heat exchangers description: Status: operating. SWEP reports that Infosys and project design lead Schneider Electric developed several data-center cooling projects between 2016 and 2020 using B439 and B649 brazed plate heat exchangers to separate primary cooling sources from the secondary loop. At the Hyderabad campus, multiple B649 units have 6-inch ports and a reported flow of 350 m³/h each; SWEP says the units were running without problems when the case was published. The named operator, design partner, product models, hydraulic role, period, and per-unit flow make this a useful deployment reference. Evidence remains supplier-authored, however, and public material does not disclose unit count, thermal megawatts, temperatures, pressure drop, rack density, compute platform, PUE, measured savings, maintenance history, or independent acceptance data. → Status: operating. SWEP reports that Infosys and project design lead Schneider Electric developed several data-center cooling projects between 2016 and 2020 using B439 and B649 brazed plate heat exchangers to separate primary cooling sources from the secondary loop. At the Hyderabad campus, multiple B649 units have 6-inch ports and a reported flow of 350 m³/h each; SWEP says the units were running without problems when the case was published. The named operator, design partner, product models, hydraulic role, period, and per-unit flow make this a useful deployment reference. Evidence remains supplier-authored, however, and public material does not disclose unit count, thermal megawatts, temperatures, pressure drop, rack density, compute platform, PUE, measured savings, maintenance history, or independent acceptance data. Flow is not thermal capacity without inlet and outlet temperatures and fluid properties, so the reported 350 m³/h must not be converted into megawatts from assumptions. The phrase “without any problems” also lacks an operating period, availability definition, alarm log, leakage history, fouling trend, or maintenance record. Buyer reference questions should cover exact article numbers and plate counts, water chemistry on both sides, approach temperature, pressure loss, pump energy, strainer and filtration practice, cleaning frequency, isolation and bypass, spare units, and the method for detecting internal cross-contamination. The design team should ask how parallel exchangers are balanced, what capacity remains during cleaning or replacement, whether performance was field verified, and which measured savings Infosys attributed specifically to the exchangers rather than to the wider cooling design.
  • MITA Barcelona data-processing center cooling towers description: Status: operating, based on MITA Group's 2023 case study describing the completed supply and the towers' role in maintaining equipment performance. TORRAVAL Cooling, a MITA Group company, supplied sixteen CTFP 2436 forced-draft open-circuit towers with laminar fill for a large Barcelona data-processing center operated for an international information-technology company. The source states a 20 MW required load expandable to 40 MW plus 5 MW of emergency supply; it does not explain whether those values are information-technology, electrical, or thermal capacity, so this record preserves the source wording rather than converting them into cooling duty. The rooftop towers were arranged in two facing rows to use limited space and support acoustic control, with maintenance walkways and ladders included. Public evidence does not name the operator or installer, disclose tower water temperatures and flow, measured energy or water use, redundancy, commissioning tests, rack density, compute platform, availability, or an operator-authored account, so buyers should request the approved schedule, acceptance data, annual operating records, and a reference contact. → Status: operating, based on MITA Group's 2023 case study describing the completed supply and the towers' role in maintaining equipment performance. TORRAVAL Cooling, a MITA Group company, supplied sixteen CTFP 2436 forced-draft open-circuit towers with laminar fill for a large Barcelona data-processing center operated for an international information-technology company. The source states a 20 MW required load expandable to 40 MW plus 5 MW of emergency supply; it does not explain whether those values are information-technology, electrical, or thermal capacity, so this record preserves the source wording rather than converting them into cooling duty. The rooftop towers were arranged in two facing rows to use limited space and support acoustic control, with maintenance walkways and ladders included. Public evidence does not name the operator or installer, disclose tower water temperatures and flow, measured energy or water use, redundancy, commissioning tests, rack density, compute platform, availability, or an operator-authored account, so buyers should request the approved schedule, acceptance data, annual operating records, and a reference contact. The case's separate 15 kW statement for each tower is not treated as heat-rejection capacity here because its meaning is unclear against the cited multi-megawatt requirement; it may describe installed motor power, but the source does not say. That ambiguity should be resolved from the equipment schedule rather than inferred. Qualification should request design wet bulb, hot- and cold-water temperatures, cell flow, fan power, drift, evaporation and blowdown, cycles of concentration, sound measurements, plume review, chemical treatment, hygiene plan, rooftop vibration and structural reactions, and capacity with one cell or common pump unavailable. It should also confirm whether the expansion allowance was physically installed, reserved in the layout, or only a future design intention.
  • Fugaku Stäubli CGD cooling connections description: Status: operating. Stäubli says Fujitsu selected CGD metal couplings to connect liquid-cooled blades to Fugaku's chassis and to permit tool-free hot swapping with flush-face, non-spill connections. Its current data-center page reports 400 racks, more than 150,000 processors and 384 connection points per rack. RIKEN independently confirms that development was completed and shared use began on 9 March 2021. The sources establish an operating named system and connector role, but they do not publish coolant chemistry, coupling size, pressure drop, leak history or maintenance replacement rate. Buyers should treat Fugaku as evidence of deployment scale, not as proof that the same connector selection fits a different rack or coolant. → Status: operating. Stäubli says Fujitsu selected CGD metal couplings to connect liquid-cooled blades to Fugaku's chassis and to permit tool-free hot swapping with flush-face, non-spill connections. Its current data-center page reports 400 racks, more than 150,000 processors and 384 connection points per rack. RIKEN independently confirms that development was completed and shared use began on 9 March 2021. The sources establish an operating named system and connector role, but they do not publish coolant chemistry, coupling size, pressure drop, leak history or maintenance replacement rate. Buyers should treat Fugaku as evidence of deployment scale, not as proof that the same connector selection fits a different rack or coolant. The Stäubli story was written while Fugaku was being installed, whereas RIKEN provides the firmer operating milestone; neither source is a connector acceptance report. The public record does not identify the number of coupling pairs in active coolant service versus other reported connection points, the pressure and temperature envelope, the guide mechanism's tolerance, or any leakage events over time. Due diligence should ask Stäubli and Fujitsu for the exact CGD configuration, qualification standard, coolant and material stack, hot-swap procedure, inspection interval, spare strategy and field-return data. A prospective buyer should also reproduce blade insertion and removal with the proposed chassis tolerances and verify pressure drop, residual spill and air inclusion after repeated cycling. Fugaku supports confidence that CGD can be engineered into a very large system; it does not establish an independently measured failure rate or lifecycle cost.
  • DataBank ATL1 ColdLogik rear-door cooling description: Status: operating. USystems and DataBank describe ColdLogik rear-door heat exchangers at ATL1 in Georgia Tech's CODA development. The published design point is 50 kW per enclosure with 73°F warm water, with a stated path to 100 kW per rack after limited infrastructure changes. DataBank's current facility page confirms ATL1 remains an operating site hosting Georgia Tech's High-Performance Computing Center and advertises liquid-cooled cabinet capacity. The vendor article's savings claims are not backed by a disclosed measurement protocol, and the installed door count, actual IT load and current water conditions are not public. The record therefore supports the architecture and operating site, not the full claimed efficiency comparison. → Status: operating. USystems and DataBank describe ColdLogik rear-door heat exchangers at ATL1 in Georgia Tech's CODA development. The published design point is 50 kW per enclosure with 73°F warm water, with a stated path to 100 kW per rack after limited infrastructure changes. DataBank's current facility page confirms ATL1 remains an operating site hosting Georgia Tech's High-Performance Computing Center and advertises liquid-cooled cabinet capacity. The vendor article's savings claims are not backed by a disclosed measurement protocol, and the installed door count, actual IT load and current water conditions are not public. The record therefore supports the architecture and operating site, not the full claimed efficiency comparison. DataBank's current page lists 7.1 MW of facility critical IT load, but that figure must not be attributed to the ColdLogik subsystem; the cooling source does not state how many racks or megawatts use rear doors. Likewise, the 100 kW statement describes potential after infrastructure changes rather than demonstrated operating rack density. Procurement teams should request current rack counts, door models, supply and return temperatures, water flow, fan power, maintenance history and measured seasonal cooling energy. They should also ask how ATL1 maintains temperatures while a door is open or isolated, how leaks are detected, and whether server-fan settings changed. An operator reference should separate original design targets from present operation and identify any control, hinge, hose or water-quality lessons learned since commissioning.
  • West Cambridge Data Centre ColdLogik cooling description: Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later increased cabinet density from 30 to 44 kW and raised data-hall capacity from 900 kW to 1.2 MW. Those figures come from the cooling supplier rather than an independently published acceptance test, but the university confirms the operating facility and continued research-computing role. Public sources do not disclose the current door count, inlet-water temperature, annual cooling energy, redundancy arrangement or whether every hall uses the same design. Procurement teams should use the reference to ask for operator contacts and measured seasonal data. → Status: operating. The University of Cambridge states that research computing moved into the £20 million West Cambridge Data Centre in 2015. USystems reports that ColdLogik rear-door heat exchangers later increased cabinet density from 30 to 44 kW and raised data-hall capacity from 900 kW to 1.2 MW. Those figures come from the cooling supplier rather than an independently published acceptance test, but the university confirms the operating facility and continued research-computing role. Public sources do not disclose the current door count, inlet-water temperature, annual cooling energy, redundancy arrangement or whether every hall uses the same design. Procurement teams should use the reference to ask for operator contacts and measured seasonal data. The university history confirms relocation and continued use of the facility but does not independently attribute the later capacity increase to a particular ColdLogik model. The 1.2 MW figure should therefore remain a supplier-reported hall value, not a measured IT load or proof that all cabinets operate at 44 kW. A reference interview should establish the installed models and quantities, typical and peak rack loads, water temperatures, fan energy, leak events, maintenance burden and changes made after commissioning. Buyers should also ask whether capacity growth required plant, piping or controls changes beyond the rear doors, and how cooling continuity is maintained during door service. Annual trend data with a defined electrical boundary would be needed before using the reference to predict PUE or operating cost.
  • NorthC Eindhoven MIXIT chip-cooling loop description: Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes; a customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The public record does not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. → Status: operating. Grundfos reports that NorthC, installer Hamer and Grundfos developed a MIXIT proof of concept for direct chip cooling beginning in 2021 and delivered it in mid-2022. The case says water circulates through chip-mounted cooling blocks and that MIXIT responds to rapid temperature changes; a customer quotation states that the system was still running reliably nearly three years later. NorthC independently confirms that its High Tech Campus site supports liquid cooling and exchanges waste heat and cold water through a campus ring. The public record does not identify server models, installed thermal capacity, rack density, loop temperatures, pump energy or measured heat recovery. The deployment demonstrates control integration, not a quantified whole-site efficiency result. The June 2022 date in this record represents the reported mid-2022 delivery period; the cited sources do not publish a precise commissioning day. MIXIT's exact configuration, valve size, sensors, controlled variable and relationship to pumps or a CDU are also undisclosed. Procurement teams should ask NorthC or Hamer for a loop diagram, control sequence, temperature and flow trends, alarm history, maintenance interventions and behavior during server or network changes. They should distinguish heat made available to the campus ring from heat actually reused by a customer and quantify backup rejection when campus demand is absent. Evidence that would strengthen the record includes a named server platform, installed kilowatts, measured pump and cooling energy, water temperatures, uptime and acceptance criteria. Until then, capacity and rack density correctly remain not publicly disclosed.
  • Digital Realty HKG10 Armstrong chilled-water plant rooms description: Status: operating. Armstrong reported in September 2016 that it supplied off-site-manufactured, integrated chilled-water plant rooms for the second phase of HKG10, then a Digital Realty and CenturyLink joint venture. The packages were assembled and tested at Armstrong's Halesowen factory and used variable-primary distribution. Digital Realty's current HKG10 page confirms the facility is operating at 33 Chun Choi Street with N+1 cooling, although it does not identify the plant supplier. Public sources do not disclose pump models, plant tonnage, measured efficiency, water use, commissioning results or how much of today's facility remains served by the 2016 packages. The record supports a real delivered plant reference, with performance claims still supplier reported. → Status: operating. Armstrong reported in September 2016 that it supplied off-site-manufactured, integrated chilled-water plant rooms for the second phase of HKG10, then a Digital Realty and CenturyLink joint venture. The packages were assembled and tested at Armstrong's Halesowen factory and used variable-primary distribution. Digital Realty's current HKG10 page confirms the facility is operating at 33 Chun Choi Street with N+1 cooling, although it does not identify the plant supplier. Public sources do not disclose pump models, plant tonnage, measured efficiency, water use, commissioning results or how much of today's facility remains served by the 2016 packages. The record supports a real delivered plant reference, with performance claims still supplier reported. Digital Realty's current N+1 statement describes the facility today but does not establish the redundancy arrangement of each 2016 package or prove that the original plant remains unchanged. Armstrong's article describes factory integration and testing without publishing the test schedule, witnessed results, controls architecture or post-commissioning energy data. A buyer evaluating packaged plant should request the original and current equipment schedules, package boundaries, design water temperatures, pump curves, staging logic, failure tests, transport splits and site reconnection work. Reference questions should cover factory-versus-field defects, commissioning duration, spare parts, controls updates and service response in Hong Kong. Measured annual chiller, pump and heat-rejection energy under local weather would be needed before using this project as an efficiency benchmark; no such dataset appears in the cited public sources.
  • Castrol–Hypertec Pangbourne immersion lab description: Status: pilot. Castrol announced on 20 July 2023 that immersion systems were installed and fully functional at its Pangbourne headquarters. The research setup combines Castrol fluids, Hypertec immersion-cooled server expertise, and Submer SmartPod and MicroPod tanks. The partners intended to test fluids, servers, and integrated cooling behavior. This is a real, named physical deployment at Castrol's site, but it is a supplier development laboratory rather than a production data center. The announcement does not disclose IT load, tank count, server models, fluid formulation used in each test, water temperatures, heat-rejection equipment, test protocol, measured efficiency, uptime, or customer acceptance results. It therefore supports interoperability research, not a capacity or savings claim. A buyer evaluating the reference should ask for the exact bill of materials, test duration and duty cycle, material inspections, fluid analyses, failure tests, and changes made after the program. Evidence that would strengthen the record includes published long-duration data, an operator acceptance report, or a commercial site using the qualified combination. → Status: pilot. Castrol announced on 20 July 2023 that immersion systems were installed and fully functional at its Pangbourne headquarters. The research setup combines Castrol fluids, Hypertec immersion-cooled server expertise, and Submer SmartPod and MicroPod tanks. The partners intended to test fluids, servers, and integrated cooling behavior. This is a real, named physical deployment at Castrol's site, but it is a supplier development laboratory rather than a production data center. The announcement does not disclose IT load, tank count, server models, fluid formulation used in each test, water temperatures, heat-rejection equipment, test protocol, measured efficiency, uptime, or customer acceptance results. It therefore supports interoperability research, not a capacity or savings claim. A buyer evaluating the reference should ask for the exact bill of materials, test duration and duty cycle, material inspections, fluid analyses, failure tests, and changes made after the program. Evidence that would strengthen the record includes published long-duration data, an operator acceptance report, or a commercial site using the qualified combination. The announcement also does not say which Castrol fluid was in each tank, whether tests used production batches, or whether Submer's later DC 20 approval came from this installation. Those links should not be inferred. Due diligence should request baseline and end-of-test fluid properties, server and component inventory, immersed-material exposure time, thermal load profile, pump and heat-exchanger data, maintenance interventions, spills or contamination events, and forensic inspection results. A commercial buyer should separately verify fire strategy, occupational procedures, fluid inventory, lifting and draining, spare capacity, and server-warranty ownership. Because the public source gives no later operating update, continued use and the present configuration are unknown; a current site reference or dated test report is needed before treating Pangbourne as an active qualification facility.
  • Valvoline–Iceotope 18-month fluid validation description: Status: pilot. Valvoline's December 2025 case study documents an 18-month in-application test of HTF-DE1 fluid in an Iceotope MicroDC system with HPE DL380 servers and NVIDIA A40 graphics processors. The reported load was 5.2 kW. Valvoline says viscosity remained stable, dielectric performance stayed above the Open Compute Project minimum, and forensic inspection found no corrosion or material damage. The named hardware, duration, and post-test inspection make this stronger than a generic compatibility assertion. Important limits remain: the test location is not disclosed, 5.2 kW is not representative of a full high-density rack, raw measurements and uncertainty are not published, and the report comes from the fluid supplier. It does not establish compatibility with other server generations, plastics, cables, storage devices, or operating temperatures. Buyers should request the full protocol, starting and ending fluid analyses, sample history, hardware inspection criteria, excursions, maintenance events, and written Iceotope approval for the exact production system. Independent replication or a named commercial operator would further strengthen the evidence. → Status: pilot. Valvoline's December 2025 case study documents an 18-month in-application test of HTF-DE1 fluid in an Iceotope MicroDC system with HPE DL380 servers and NVIDIA A40 graphics processors. The reported load was 5.2 kW. Valvoline says viscosity remained stable, dielectric performance stayed above the Open Compute Project minimum, and forensic inspection found no corrosion or material damage. The named hardware, duration, and post-test inspection make this stronger than a generic compatibility assertion. Important limits remain: the test location is not disclosed, 5.2 kW is not representative of a full high-density rack, raw measurements and uncertainty are not published, and the report comes from the fluid supplier. It does not establish compatibility with other server generations, plastics, cables, storage devices, or operating temperatures. Buyers should request the full protocol, starting and ending fluid analyses, sample history, hardware inspection criteria, excursions, maintenance events, and written Iceotope approval for the exact production system. Independent replication or a named commercial operator would further strengthen the evidence. The source does not state whether the 5.2 kW value is average, maximum, or nameplate load, how continuously the hardware ran, what inlet and outlet temperatures applied, or whether fluid was filtered or topped up. It also does not publish individual dielectric and viscosity readings, laboratory methods, uncertainty, or photographs and measurements from the forensic examination. Procurement teams should ask whether the exact HPE and NVIDIA configurations retained their warranties and whether Iceotope's approval applies to all MicroDC revisions. A production pilot should reproduce the intended materials and temperature range, include representative service events, preserve sealed baseline samples, define pass and fail limits before testing, and assign authority for continued operation after any out-of-limit sample.
  • NREL ESIF warm-water cooling loop description: Status: operating. Xylem's case study says more than 55 Bell & Gossett products, including e-1510, Series 90, Series 80, and Series 60 pumps, were selected for the National Renewable Energy Laboratory's Energy Systems Integration Facility in Golden, Colorado. The data center uses component-level warm-water cooling for supercomputing, transfers heat through an energy-recovery loop, and reuses available heat in laboratories and offices. NREL confirms that the water-based system was installed in 2012 and describes the operating hierarchy for heat reuse, dry rejection, and cooling towers. This record establishes named equipment families and an operating facility, but it does not disclose which pump serves each loop, current pump quantities by model, selected flow and head, annual pump energy, maintenance history, or attribution of whole-site efficiency to Xylem equipment. The computing platform has also changed over time. Buyers should use the reference to examine controls, low-load operation, redundancy, water chemistry, heat-reuse availability, and maintenance, not to copy a pump schedule. Current drawings, trend data, and service records would be needed for a like-for-like benchmark. → Status: operating. Xylem's case study says more than 55 Bell & Gossett products, including e-1510, Series 90, Series 80, and Series 60 pumps, were selected for the National Renewable Energy Laboratory's Energy Systems Integration Facility in Golden, Colorado. The data center uses component-level warm-water cooling for supercomputing, transfers heat through an energy-recovery loop, and reuses available heat in laboratories and offices. NREL confirms that the water-based system was installed in 2012 and describes the operating hierarchy for heat reuse, dry rejection, and cooling towers. This record establishes named equipment families and an operating facility, but it does not disclose which pump serves each loop, current pump quantities by model, selected flow and head, annual pump energy, maintenance history, or attribution of whole-site efficiency to Xylem equipment. The computing platform has also changed over time. Buyers should use the reference to examine controls, low-load operation, redundancy, water chemistry, heat-reuse availability, and maintenance, not to copy a pump schedule. Current drawings, trend data, and service records would be needed for a like-for-like benchmark. The sources identify 2012 as the installation year but do not provide the exact day; the structured 2012-01-01 date is therefore a normalization placeholder, not a claimed commissioning date. NREL's current cooling description also reflects later heat-rejection additions and should not be assumed to match the original configuration. Reference checks should separate the technology loop, energy-recovery loop, process-hot-water interface, thermosyphon, and tower loop, then identify the Xylem equipment and controls in each. Useful evidence would include selected pump curves, annual operating points, standby testing, seal and bearing history, water-treatment records, measured pump energy, periods when reusable heat had no customer, and the present supercomputer's actual supply and return temperatures.
  • Novva Colorado Springs cooling-controls retrofit description: Status: operating. Siemens and Novva Data Centers report that Desigo PXC controllers replaced the prior cooling-control system at Novva's Colorado Springs facility. Holbrook Service arranged the controllers in a flat architecture across four chillers, two chiller plants, cooling towers, pumps, and fans so remaining equipment could respond if one chiller faulted. Siemens says the retrofit supported Novva's transition away from water cooling, eliminated prior weekly service needs, delivered no controller-related central-plant downtime for a year and a half, and saved more than two million kilowatt-hours and $176,000 annually. This is a named operator, site, equipment family, and completed project. The limits are important: Siemens is the source, no independent audit or raw trends are published, and the savings include controller replacement, rewiring, reprogrammed sequences, and a broader cooling-strategy change. The source does not disclose IT capacity, controller count, exact PXC models, baseline dates, weather or load normalization, capital cost, or final water use. Buyers should request Novva's acceptance tests, failure matrix, trend data, network design, maintenance record, and savings method before using the figures in a business case. → Status: operating. Siemens and Novva Data Centers report that Desigo PXC controllers replaced the prior cooling-control system at Novva's Colorado Springs facility. Holbrook Service arranged the controllers in a flat architecture across four chillers, two chiller plants, cooling towers, pumps, and fans so remaining equipment could respond if one chiller faulted. Siemens says the retrofit supported Novva's transition away from water cooling, eliminated prior weekly service needs, delivered no controller-related central-plant downtime for a year and a half, and saved more than two million kilowatt-hours and $176,000 annually. This is a named operator, site, equipment family, and completed project. The limits are important: Siemens is the source, no independent audit or raw trends are published, and the savings include controller replacement, rewiring, reprogrammed sequences, and a broader cooling-strategy change. The source does not disclose IT capacity, controller count, exact PXC models, baseline dates, weather or load normalization, capital cost, or final water use. Buyers should request Novva's acceptance tests, failure matrix, trend data, network design, maintenance record, and savings method before using the figures in a business case. Siemens' reference page lists project completion in 2023, while the public savings announcement is dated 22 July 2024; the announcedAt field uses the documented announcement date rather than implying that commissioning occurred then. The phrase water-free cooling should also be checked against the source's simultaneous description of chillers and cooling towers, because the public materials do not provide a final water balance or equipment schematic. Reference diligence should ask how the live cutover was staged, which functions remained locally autonomous, what faults were injected, how controller and network redundancy were tested, and whether the reported zero downtime excludes upstream power, mechanical, or sensor events. Raw interval data and a normalized baseline are needed to validate the energy and cost claims.
  • Illinois data-center cooling-water treatment retrofit description: Status: operating, based on a supplier case summarized by Veolia on 23 July 2025. Veolia says an Illinois data center had low cooling-tower cycles of concentration and high blowdown under its prior alkaline treatment program. Veolia installed a TrueSense Ready-Set-Go controller to monitor pH and feed sulfuric acid, together with remote monitoring and alarms. The company reports that tower cycles doubled, cooling-water demand fell by 50%, and annual savings reached 12 million gallons and $150,000. The operator, city, facility capacity, number and type of towers, baseline period, final cycles, source-water chemistry, acid dose, measurement method, and independent verification are not disclosed. Those omissions prevent normalization by IT load, weather, or water quality and make this a supplier-reported reference rather than an audited benchmark. Acid-feed systems also create safety and corrosion risks if controls fail. Buyers should request the water model, materials review, setpoints, interlocks, containment, alarm response, trend data, calculation method, and operator reference before using the savings in a business case. → Status: operating, based on a supplier case summarized by Veolia on 23 July 2025. Veolia says an Illinois data center had low cooling-tower cycles of concentration and high blowdown under its prior alkaline treatment program. Veolia installed a TrueSense Ready-Set-Go controller to monitor pH and feed sulfuric acid, together with remote monitoring and alarms. The company reports that tower cycles doubled, cooling-water demand fell by 50%, and annual savings reached 12 million gallons and $150,000. The operator, city, facility capacity, number and type of towers, baseline period, final cycles, source-water chemistry, acid dose, measurement method, and independent verification are not disclosed. Those omissions prevent normalization by IT load, weather, or water quality and make this a supplier-reported reference rather than an audited benchmark. Acid-feed systems also create safety and corrosion risks if controls fail. Buyers should request the water model, materials review, setpoints, interlocks, containment, alarm response, trend data, calculation method, and operator reference before using the savings in a business case. The source publication date is not identified as the installation or commissioning date, so the deployment schedule and length of demonstrated operation remain unknown. “Doubled” tower cycles is also incomplete without the starting and ending values, seasonal source-water range, blowdown meter data, cooling load, and weather. Reference diligence should verify acid-storage and transfer safeguards, secondary containment, compatible materials, probe calibration, dosing proof, high- and low-pH interlocks, response after a failed sensor or stuck pump, and the remote alarm escalation record. An operator contact, water and sewer invoices, laboratory results, corrosion and deposition trends, and a calculation normalized by heat rejection would materially strengthen both the operating label and the savings claims.
• Oct. 4, 2026

  Research expansion published: 90 records

  • Smardt New York-region data-center chiller replacement Added Smardt New York-region data-center chiller replacement
  • BAC 160 MW HPC HXV cooling project Added BAC 160 MW HPC HXV cooling project
  • Güntner Zhangjiakou immersion-cooling heat rejection Added Güntner Zhangjiakou immersion-cooling heat rejection
  • Infosys Hyderabad data-center heat exchangers Added Infosys Hyderabad data-center heat exchangers
  • MITA Barcelona data-processing center cooling towers Added MITA Barcelona data-processing center cooling towers
  • Fugaku Stäubli CGD cooling connections Added Fugaku Stäubli CGD cooling connections
  • DataBank ATL1 ColdLogik rear-door cooling Added DataBank ATL1 ColdLogik rear-door cooling
  • West Cambridge Data Centre ColdLogik cooling Added West Cambridge Data Centre ColdLogik cooling
  • NorthC Eindhoven MIXIT chip-cooling loop Added NorthC Eindhoven MIXIT chip-cooling loop
  • Digital Realty HKG10 Armstrong chilled-water plant rooms Added Digital Realty HKG10 Armstrong chilled-water plant rooms
  • Castrol–Hypertec Pangbourne immersion lab Added Castrol–Hypertec Pangbourne immersion lab
  • Valvoline–Iceotope 18-month fluid validation Added Valvoline–Iceotope 18-month fluid validation
  • NREL ESIF warm-water cooling loop Added NREL ESIF warm-water cooling loop
  • Novva Colorado Springs cooling-controls retrofit Added Novva Colorado Springs cooling-controls retrofit
  • Illinois data-center cooling-water treatment retrofit Added Illinois data-center cooling-water treatment retrofit

Formats and licence

• CSVAvailable at /datasets/liquid-cooling-deployments/data.csv, RFC 4180 quoted.
• JSONAvailable at /datasets/liquid-cooling-deployments/data.json, with the same columns and row order.
• ParquetNot published. The row counts here do not justify it yet, and offering a format we do not regenerate would be worse than not offering it.
• APINo query API. The JSON file is the whole dataset, so fetching it is the API for now.

Licensed CC BY 4.0: Free to use, republish, and build on, including commercially, with attribution to DataCenterChill and a link to the dataset page. Read the licence.