Liquid Cooling Adoption Barriers Still Slow Down Rollouts

A GB200 NVL72 rack draws 132 kW, with 115 kW carried by liquid cooling and the remaining 17 kW carried by air, according to HPE’s published specification for the system. A standard raised-floor tile can move enough air to cool roughly 20 kW before its airflow runs out. Uptime Institute’s 2025 Cooling Systems Survey, covering 861 data center operators, found that the top-cited obstacle to closing that gap is not thermal performance. It is the absence of an agreed hardware standard, named by 39 percent of respondents as a leading barrier to deploying direct liquid cooling.

The 20 kW air cooling limit per rack comes from a floor tile, not from air itself

The commonly quoted 20 to 25 kW air cooling limit per rack is not a property of air as a cooling fluid. It reproduces from two published figures: how much airflow a rack needs per kilowatt of IT load, and how much airflow a single floor tile can deliver.

A widely cited rule of thumb, reported in the paper “Measurement of Perforated Tile Airflow in Data Centers” (ASHRAE Transactions, Vol. 122, Part 1), estimates rack airflow at roughly 212 m³/h, or 125 CFM, per kilowatt of power dissipation. Separately, a calorimetric measurement study published in the ASME Journal of Electronic Packaging (2017) tested perforated tile airflow across a range of 500 to 2,500 CFM and found the measurement method itself became unreliable above roughly 2,400 CFM, where underfloor pressure drop overwhelms fan capacity. In a typical raised-floor layout, the cold aisle is two tiles wide, giving one supply tile per IT cabinet.

Dividing tile airflow by the per-kilowatt airflow ratio gives the ceiling: 2,400 CFM ÷ 125 CFM/kW ≈ 19.2 kW, close to 20 kW. Reaching 25 kW under the same ratio would require roughly 3,125 CFM from a single tile, beyond the range where the ASME study’s own measurement tooling still performed reliably. This is an upper-bound estimate built from two published rule-of-thumb figures, not a measurement of any specific facility, and it assumes full airflow capture with no bypass around the rack.

Source: “Measurement of Perforated Tile Airflow in Data Centers,” ASHRAE Transactions, Vol. 122, Part 1; VanGilder et al., “Measurement of Air Flow Rate Through Perforated Floor Tiles in a Raised Floor Data Center,” Journal of Electronic Packaging (ASME, 2017)

The 20 kW figure describes what one floor tile and one fan curve can deliver, not a thermodynamic ceiling on air as a cooling medium. A facility running a higher-flow grate, a contained cold aisle, or a server model with a lower airflow-per-kilowatt ratio can push past 20 kW without touching a coolant loop, and many enterprise racks already do.

Lack of standardization is the most cited barrier, not cost or performance

Operators name a missing hardware standard, not price or reliability, as the leading reason direct liquid cooling deployment lags behind the technology’s readiness. In Uptime Institute’s 2025 Cooling Systems Survey, 39 percent of the 861 respondents cited lack of standardization across systems as a top barrier, ahead of cost at 38 percent and reliability concerns at 35 percent. Table 1. Barriers to deploying direct liquid cooling% of respondents (n=861)

* Respondents could choose up to two options; shares do not sum to 100%

Source: Uptime Institute, “Cooling Systems Survey 2025: DLC Adoption Remains Slow and Steady” (UII Data Report 181, July 2025)

Source: Uptime Institute, “Cooling Systems Survey 2025: DLC Adoption Remains Slow and Steady” (UII Data Report 181, July 2025)

Schneider Electric’s white paper on direct liquid cooling system challenges lays out why the gap persists at the equipment level. There are currently no industry-accepted standards for quick-connect couplings, manifolds, or piping across vendors, which is why the paper recommends dedicating a single coolant distribution unit (CDU) loop to a single server model rather than mixing vendors on one loop. On the CDU side, ANSI/ASHRAE Standard 127-2020 covers air-conditioning units for data centers, but the addendum that would add direct liquid cooling test conditions and rating equations for CDUs was still in draft at the time of the paper’s writing, so vendors remain free to state efficiency and capacity ratings under whichever fluid temperature, flow rate, and ambient conditions favor their own product.

Source: Schneider Electric, “Direct Liquid Cooling System Challenges in Data Centers,” White Paper 210, Version 1 (2025)

Two CDUs rated under different, vendor-chosen test conditions are not comparable no matter how mature either vendor’s engineering is, and that comparison problem sits upstream of any single purchasing decision an operator makes. Until the ASHRAE addendum publishes, a specification sheet is a marketing document as much as an engineering one.

Operators cannot hire or train fast enough to run liquid loops

A shortage of trained staff is a separate constraint from equipment maturity, and it will not close on a vendor’s timeline. Uptime’s 2025 survey found 16 percent of respondents cited staff training as a barrier to direct liquid cooling, smaller than the standardization and cost figures but persistent across the last several years of the same survey.

Source: Uptime Institute, “Cooling Systems Survey 2025: DLC Adoption Remains Slow and Steady” (UII Data Report 181, July 2025)

The broader hiring picture explains why. Audrey Escalante, head of global operations learning and development at Digital Realty, told Fierce Network that data center engineers are the role the company most needs to fill, adding: “We don’t have data center engineers coming out the door.” Fierce Network’s reporting cites a JLL midyear 2026 report showing North America has more than 66 GW of data center capacity under construction, a volume that would draw more electricity than all of Germany consumes, while vacancy across the sector has held near 1 percent for three straight years. Separately, Kao Data’s marketing director told Data Center Knowledge that HVAC and thermal specialists are drawing similar demand as operators shift toward liquid and hybrid cooling.

Source: Fierce Network, “Data center labor shortage pushes operators to train, poach and automate” (2026); Data Center Knowledge, “Data Centers Build Talent Pipelines to Close AI-Era Skills Gap” (2026)

Schneider Electric’s white paper frames the operational side of the same problem: facility teams are comfortable running chilled-water plants because they have done it for decades, but distributing coolant directly into a rack is new to most of them and represents a steep learning curve.

Source: Schneider Electric, “Direct Liquid Cooling System Challenges in Data Centers,” White Paper 210, Version 1 (2025)

Training people to run a coolant distribution loop competes for the same electricians and mechanical trades that every other data center project in this construction cycle is also trying to hire. That makes the timeline for closing this barrier a function of the labor market, not of a vendor’s product road map.

When a coolant loop fails, no one is sure who pays

Warranty and repair responsibility between vendors is unresolved in direct liquid cooling in a way it never was in air-cooled facilities, because the coolant infrastructure is physically coupled to the server rather than stopping at a computer room air handler. Schneider Electric’s white paper describes this directly: the physical coupling between servers and coolant distribution complicates the delineation of responsibilities between vendors, and ambiguous or contradictory warranties often lead to finger pointing between vendors, which delays repairs and adds cost for the operator. The paper poses the underlying question plainly: if a leak occurs inside a liquid-cooled server, it may be unclear whether the server vendor, the CDU vendor, or the coolant vendor is responsible.

Source: Schneider Electric, “Direct Liquid Cooling System Challenges in Data Centers,” White Paper 210, Version 1 (2025)

The Open Compute Project’s liquid distribution reference design guidance describes the same ambiguity from the facility side. Where a colocation tenant deploys liquid-cooled IT equipment, ownership of maintaining and warranting coolant quality can be genuinely unclear when a multi-tenant data center operator, an IT equipment vendor, and an IT equipment owner are all involved in the same coolant loop. Table 2. The three barriers, evidence, and sourceSummary

* Percentages are not available for every row; the liability barrier is documented qualitatively, not surveyed

Source: Uptime Institute (2025); Schneider Electric White Paper 210 (2025); Open Compute Project ACF Reference Design Guidance

Source: Open Compute Project, “Data Center Liquid Distribution Guidance & Reference Designs,” Revision 0

A warranty dispute over a coolant leak does not wait for resolution before it costs money. The operator absorbs the downtime and the repair bill while vendors work out whose contract governs, which makes this barrier behave more like an insurance gap than a technical shortfall, and no CDU firmware update closes an insurance gap.

How to read the survey numbers

Uptime Institute’s barrier question changed format between years: the 2024 survey (n=86 for that question) asked respondents to choose up to three barriers, while the 2025 survey (n=861) asked for up to two, so the percentages in this piece should not be read as a clean year-over-year trend line. The survey population also skews toward operators, vendors, and consultants who choose to respond to Uptime’s outreach, concentrated in North America and Europe, which may not match the barrier mix a colocation provider in a different region would report. Schneider Electric’s white paper documents vendor-observed challenges and recommended practice, not a measured incidence rate of warranty disputes or leak events. The floor-tile calculation in this piece combines two published rule-of-thumb figures rather than a single measured facility, and it assumes full airflow capture with no underfloor plenum leakage, which published leakage studies put at 5 to 15 percent of available cooling air in a typical raised-floor facility. This piece also does not attempt to rank the three barriers against each other in dollar terms, or to resolve any question of safe system design, which requires facility-specific engineering review.

Contract and staffing steps before the first coolant loop

• If a CDU specification sheet does not state the fluid temperature, flow rate, and ambient conditions its rating was tested under, treat two vendors’ numbers as non-comparable until ANSI/ASHRAE Addendum b publishes and both are retested against it.
• If your facilities team has not operated a technology cooling system (TCS) loop before, budget a training and vendor-support period measured in months, not a single onboarding session.
• If your rack power density need is under roughly 20 kW, a higher-flow tile or a contained cold aisle may extend air cooling further before a coolant loop is required at all.
• Get written agreement from the IT equipment vendor and the CDU vendor on who pays for damage from a leak inside the server before signing, not after a leak occurs.
• If you plan to mix server models or vendors on a single coolant loop, expect warranty disputes; Schneider Electric’s guidance recommends segregating vendors into separate loops.
• Treat any market-size forecast for liquid cooling, including Dell’Oro Group’s, as a single firm’s scope-dependent projection rather than a fixed number, and check the base year and market definition before comparing it to another forecast.

References

Uptime Institute, “Cooling Systems Survey 2025: DLC Adoption Remains Slow and Steady,” UII Data Report 181, July 2025, https://intelligence.uptimeinstitute.com/sites/default/files/2025-07/UI%20Field%20181_Data%20center%20cooling.pdf Uptime Institute, “2024 Cooling Systems Survey: Direct Liquid Cooling Results,” May 2024, https://datacenter.uptimeinstitute.com/rs/711-RIA-145/images/2024.Cooling.Survey.Report.pdf Schneider Electric, “Direct Liquid Cooling System Challenges in Data Centers,” White Paper 210, Version 1, 2025, https://media.datacenterdynamics.com/media/documents/WP210_V1_EN.pdf Open Compute Project, “Data Center Liquid Distribution Guidance & Reference Designs,” Revision 0, https://ocp-all.groups.io/g/OCP-Cooling-Environments/attachment/11/3/OCP%20ACF%20Reference%20Design%20Guidance.pdf HPE, “NVIDIA GB200 NVL72 by HPE” product specification page, https://buy.hpe.com/us/en/compute/rack-scale-system/nvidia-nvl-system/nvidia-gb200-nvl72-by-hpe/p/1014890104 “Measurement of Perforated Tile Airflow in Data Centers,” ASHRAE Transactions, Vol. 122, Part 1, accessed via ResearchGate, https://www.researchgate.net/publication/311964403_Measurement_of_Perforated_Tile_Airflow_in_Data_Centers VanGilder et al., “Measurement of Air Flow Rate Through Perforated Floor Tiles in a Raised Floor Data Center,” Journal of Electronic Packaging (ASME, 2017), https://asmedigitalcollection.asme.org/electronicpackaging/article/139/1/011007/372660 Fierce Network, “Data center labor shortage pushes operators to train, poach and automate,” 2026, https://www.fierce-network.com/cloud/data-center-labor-shortage-pushes-operators-train-poach-and-automate Data Center Knowledge, “Data Centers Build Talent Pipelines to Close AI-Era Skills Gap,” 2026, https://www.datacenterknowledge.com/training-certifications/data-centers-take-training-into-their-own-hands-amid-talent-shortages Dell’Oro Group, “Data Center Liquid Cooling Market to Approach $7 Billion by 2029 as AI Deployments Accelerate,” January 8, 2026, https://www.delloro.com/news/data-center-liquid-cooling-market-to-approach-7-billion-by-2029-as-ai-deployments-accelerate/