Air Cooling Limit per Rack Is a Floor Tile, Not Physics

Uptime Institute puts the air cooling limit per rack at 20 to 25 kW. Schneider Electric’s engineering study puts it at 33 to 35 kW. Neither number describes a property of air. Push 1,900 cfm, the best-in-class floor tile airflow ASHRAE cites, through the sensible heat equation, and the 20 to 25 kW band falls out at a temperature rise of 19 to 24 kelvin across the rack.

Air cooling capacity is a mass flow problem, and the arithmetic fits on one line

Air removes heat in proportion to how much of it reaches the rack and how much warmer it gets on the way through. The governing equation is the sensible heat equation:

Q = V̇ × ρ × c_p × ΔT

Assumptions, with units:

• ρ = 1.18 kg/m³, dry air near 25 °C at 101.325 kPa
• c_p = 1.005 kJ/(kg·K)
• 1 cfm = 0.000471947 m³/s
• Sensible heat transfer only, no condensation at the rack

Substituting one best-in-class floor tile:

• V̇ = 1,900 cfm = 0.8967 m³/s
• ṁ = 0.8967 × 1.18 = 1.0581 kg/s
• ṁ × c_p = 1.0581 × 1.005 = 1.0634 kW per kelvin

Each kelvin of temperature rise across the rack buys 1.06 kW. At a 20 kelvin rise, one tile supports 21.3 kW.

That result is an upper bound, not a measurement. It assumes every cubic meter leaving the tile enters a server, and that none of it bypasses the rack or recirculates out of the hot aisle. Real rooms do worse.

Source: ASHRAE TC9.9, “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” (2021); the 1,900 cfm figure is attributed there to ASHRAE, IT Equipment Design Impact on Data Center Solutions (2016)

The 20 to 25 kW band is not a discovered property of a gas. It is one floor tile multiplied by whatever temperature rise the room operator is willing to accept.

The airflow term runs out first, and the floor tile is where it runs out

Two of the four terms are fixed by nature and two are set by the building. Density and specific heat do not move. Airflow and temperature rise are design choices, and airflow is the one that fails first.

ASHRAE states that servers on the market require 100 cfm or more per U. A single 1,900 cfm tile therefore supports about 19U of such equipment.

The same document notes that a 40 to 50 kW rack could pull up to 5,000 cfm, against that same best-in-class tile figure.

Source: ASHRAE TC9.9, “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” (2021)

A published product specification confirms the arithmetic from the opposite direction. Rittal rates its SK 3312.540 in-row air-to-water unit at a maximum cooling output of 30 kW with an air throughput of 5,000 m³/h, a maximum air inlet temperature of 24 °C, and a water inlet temperature of 15 °C.

Converting and substituting:

• V̇ = 5,000 m³/h = 1.3889 m³/s
• ṁ × c_p = 1.3889 × 1.18 × 1.005 = 1.6471 kW per kelvin
• ΔT = 30 ÷ 1.6471 = 18.2 kelvin

That airflow is 2,943 cfm, roughly one and a half floor tiles’ worth, delivered at the rack face rather than up through the floor.

Source: Rittal, SK 3312.540 Liquid Cooling Package LCP Inline CW specification (state 10 September 2026)

A standards body and a manufacturer land on the same kilowatts per kelvin from opposite ends. The equation is not what constrains anyone. The supply path is.

The published limits disagree because each one measures a different thing

The 20 to 25 kW and 20 to 30 kW ranges circulating in the industry are not competing estimates of one quantity. They are limits on four unrelated pieces of hardware, and each is correct inside its own scope.

Table 1. Four published air cooling limits per rack, and what each one measureskW per rack

Source	Stated limit	What the number constrains	Conditions stated by the source
Uptime Institute, “Rack Density is Rising” (2020)	20–25	Economic crossover to direct liquid cooling and precision air cooling	Survey observation across respondents; no test condition given
Rittal, SK 3312.540 LCP Inline CW specification (2026)	30	Maximum cooling output of one in-row air-to-water unit	Max. air inlet 24 °C, water inlet 15 °C, 5,000 m³/h, four fan modules
Schneider Electric, power distribution study (2021)	33–35	Rack PDU ampacity, not heat removal	63 A rack PDU at 400 V; IEC power environments
Schneider Electric, RDHx retrofit note (2025)	72	Rear door heat exchanger capacity at the cabinet	Motivair ChilledDoor on air cooled servers; existing chilled water system required

* Four different constraints; not competing estimates of one quantity

* Rittal and Motivair values are single product ratings, not industry limits

Source: Uptime Institute (2020); Rittal, SK 3312.540 specification (2026); Schneider Electric (2021, 2025)

Uptime Institute’s figure comes from a survey observation rather than a thermal calculation: above 20 to 25 kW, direct liquid cooling and precision air cooling become more economical and efficient than room air.

Schneider Electric’s number comes from amperes. Its engineers state that 63 A is the maximum ampacity of a rack PDU, and that at 400 V this yields a maximum rack density of 35 kW per rack.

• √3 × 400 V × 63 A = 43.6 kVA
• 43.6 × 0.8 continuous derating = 34.9 kW

Source: Uptime Institute, “Rack Density is Rising” (December 7, 2020); Schneider Electric, “New Research Studies Impact of Rack-based Liquid Cooling and High Power Densities on Power Distribution” (March 25, 2021)

Quoting any of these as the air cooling limit per rack is a category error. The 20 to 25 kW number belongs to a room, the 35 kW number belongs to a power cord, and the 72 kW number belongs to a door.

Widening the temperature rise moves the constraint onto the hot aisle

Airflow is capped by the floor, which leaves ΔT as the only free term, and that term is bounded by people rather than by thermodynamics.

ASHRAE runs the case directly. At 27 °C, the upper end of the recommended inlet range, with a maximum recommended dew point of 15 °C and a rack ΔT of 15 °C, easy work in the hot aisle is unrestricted.

Moderate work at those same conditions requires a 30 minute break every hour and a prescribed minimum water intake. Move the inlet to the A1 allowable upper dry bulb of 32 °C with the same 15 °C rack ΔT, and no work of any kind in the hot aisle meets the OSHA guidelines without risk of heat stress.

Source: ASHRAE TC9.9, “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” (2021), citing OSHA Technical Manual TED 1-0.15A (2020)

The inlet is moving the wrong way at the same time. The fifth edition of Thermal Guidelines for Data Processing Environments adds an H1 class for high density equipment, with an allowable range of 15 °C to 25 °C and a recommended range of 18 °C to 22 °C.

Dense hardware is pulling the cold aisle down while the equation asks for a wider spread. Those two requirements point in opposite directions, and the hot aisle absorbs the difference.

Fan power reverses the answer before the air does

Buying more airflow costs power, and past roughly 25 kW the cost stops being marginal.

ASHRAE records that server fan power fell from levels as high as 20 percent of server power to as low as 2 percent during the multicore period. A fan power percentage of 10 percent to 20 percent is not uncommon again for some of the denser servers.

In a 50 kW rack, that translates to at least 5 kW of fan power.

The consequence is not thermal. Server fans draw from the same UPS source as the servers, so retaining air cooling while fan power climbs from 2 percent to 10 percent equates to reducing UPS capacity by 8 percent. The UPS sizing assumption fails before the cooling does, and it fails quietly, because nothing gets hot.

Source: ASHRAE TC9.9, “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” (2021)

Noise closes the remaining room. Data center A-weighted sound pressure levels regularly meet or exceed 80 dBA, and a 30 percent increase in air mover speed brings an expected increase of 6 dB. Raising fan speed satisfies the equation and produces a rack that eats its own power budget and cannot be staffed.

Table 2. What one 1,900 cfm floor tile supports as the temperature rise widenskW per rack, °C

Rise across rack (K)	Rack power supported (kW)	Exhaust at 24 °C inlet (°C)
10	10.6	34
15	15.9	39
20	21.3	44
25	26.6	49
30	31.9	54

* Upper bound; assumes no bypass and no recirculation

* Air at 1.18 kg/m³ and 1.005 kJ/(kg·K); sensible heat only

* Red rows span the commonly quoted 20 to 25 kW band

* Third column is exhaust, not inlet; hot aisle heat stress limits bind before the last row

Source: calculated from ASHRAE TC9.9, “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” (2021), 1,900 cfm per floor tile

Most operators never test any of this. Uptime Institute’s 2025 survey of more than 800 owners and operators put the average of modal rack densities at almost 9 kW.

Excluding sites whose typical density is 30 kW or above, that average falls to 7.5 kW, and more than 80 percent of respondents reported no racks above 30 kW at all.

Source: Uptime Institute, Global Data Center Survey 2025 (July 2025; fielded April to May 2025, n=710 for modal density)

The 20 to 25 kW figure has survived fifteen years of warnings partly because most of the industry has never come close enough to test it.

What this analysis does not cover

The 1,900 cfm floor tile figure is a best-in-class value ASHRAE cites from a 2016 book, not a measurement of any particular room, and rooms without raised floors have no floor tiles at all. The temperature rise column in Table 2 is arithmetic rather than a test result: it assumes perfect containment with no bypass and no recirculation, and it ignores the pressure drop through the server chassis that actually sets how much air a fan can move. Rittal’s 30 kW rating carries three stated conditions that travel with it, and a unit run at a different water inlet temperature will not produce that number. Uptime Institute’s density figures are self-reported by survey respondents, and the 2020 article stating the 20 to 25 kW crossover and the 2025 survey draw on different samples five years apart, so the two are not directly comparable. Whether a given room can deliver its design airflow to a given cabinet is a question for computational fluid dynamics and site measurement, and any density change at the rack requires engineering review of the electrical and mechanical systems feeding it.

Decision criteria

• Establish which term binds in your own room before quoting a limit. Measure delivered airflow at the cabinet face, not the nameplate output of the air handler.
• Check rack PDU ampacity and distribution voltage before assuming a thermal ceiling. At 400 V and 63 A the electrical limit arrives around 35 kW whatever the cooling can do.
• Treat any vendor kW rating as inseparable from its stated air inlet and water inlet temperatures. Change either one and the rating no longer applies.
• Track fan power as a share of IT power on the densest racks you operate. Past 10 percent, the UPS is being consumed by cooling that never appears on the cooling bill.
• Measure hot aisle dry bulb and dew point at the height where technicians work, and set the maximum ΔT from what that allows rather than from what the equation permits.
• If the cold aisle set point has to come down for new equipment, count the economization hours lost before counting the cooling capacity gained.

FAQ

Is the air cooling limit per rack 20 kW, 25 kW, or 30 kW?

All three appear in reputable sources because they describe different constraints. Uptime Institute’s 20 to 25 kW is an economic crossover observed across survey respondents, above which direct liquid cooling and precision air cooling become more economical and efficient. Rittal’s 30 kW is a single product rating at a maximum air inlet of 24 °C and a water inlet of 15 °C. Neither is a property of air, and neither transfers to a room with different airflow at the cabinet.

Why does Schneider Electric say 35 kW when others say 25 kW?

Schneider Electric’s number is electrical, not thermal. Its engineering study states that 63 A is the maximum ampacity of a rack PDU, and that at 400 V this yields a maximum rack density of 35 kW per rack, with the achievable range put at 33 to 35 kW. The study covered IEC power environments and was aimed at power distribution above 35 kW, the range where liquid cooled systems may come into play. Above 63 A, the study notes, the plug standards themselves have to change.

How much airflow does a high density rack actually need?

ASHRAE states that a 40 to 50 kW rack could require up to 5,000 cfm, against a best-in-class floor tile output of 1,900 cfm. Servers on the market require 100 cfm or more per U, so one tile supports about 19U of that equipment. The figure scales linearly with the temperature rise allowed across the rack, which is why the same rack power gets quoted at very different airflow numbers.

Can air cooling handle racks above 50 kW?

At the rack, yes, using a heat exchanger at the cabinet rather than air from the room. Schneider Electric states that rear door heat exchangers such as the Motivair ChilledDoor cool up to 72 kW per rack on air cooled servers, and that a Motivair heat dissipation unit reaches 150 kW per rack while rejecting that heat back into the white space. The trade-off is that room air conditioning then has to absorb it, which raises PUE and may exceed the spare capacity of the existing system. Any such retrofit requires engineering review of the chilled water and heat rejection systems in place.

Related reading

Anchor text and placement for internal links:

• In the mass flow section, on “the sensible heat equation”, link to a piece deriving the equation and its unit conversions for facility engineers.
• In the temperature rise section, on “ASHRAE H1 class”, link to a walkthrough of the A1 through A4 and H1 environmental classes in the fifth edition.
• In the fan power section, on “fan power as a share of server power”, link to a piece on how the fan affinity laws set the power penalty for airflow.
• In the closing criteria, on “rack PDU ampacity”, link to a piece on 400 V rack distribution and the 63 A ceiling.

Bottom-of-page links:

• Reading a liquid cooling specification: what CDU and cold plate ratings actually promise
• ASHRAE W classes and why supported facility water temperatures are regressing
• Rear door heat exchangers versus direct-to-chip cold plates in an existing room
• What PUE hides about cooling energy at high rack power density

References

ASHRAE Technical Committee 9.9 / “Emergence and Expansion of Liquid Cooling in Mainstream Data Centers” / ASHRAE / 2021 / https://www.vertiv.com/49db1c/globalassets/documents/white-papers/ashrae_tc0909_emergence_and_expansion_of_liquid_cooling_in_mainstream_data_centers_5_may_2021_332771_0.pdf (hosted copy)

Uptime Institute / “Rack Density is Rising”, Andy Lawrence / Uptime Institute Journal / December 7, 2020 / https://journal.uptimeinstitute.com/rack-density-is-rising/

Uptime Institute / Global Data Center Survey 2025, Keynote Report 180 / Uptime Institute Intelligence / July 2025 / https://datacenter.uptimeinstitute.com/rs/711-RIA-145/images/2025.Annual.Survey.Report.pdf

Schneider Electric / “New Research Studies Impact of Rack-based Liquid Cooling and High Power Densities on Power Distribution”, Gia Wiryawan / Schneider Electric Blog / March 25, 2021 / https://blog.se.com/datacenter/cloud-and-service-providers/2021/03/25/research-studies-impact-rack-based-liquid-cooling-high-power-densities-power-distribution/

Schneider Electric / “Upgrade legacy data centers for AI workloads with RDHx liquid cooling”, Steven Carlini / Schneider Electric Blog / November 11, 2025 / https://blog.se.com/datacenter/2025/11/11/upgrade-legacy-data-centers-for-ai-workloads-with-rdhx-liquid-cooling/

Rittal / SK 3312.540 Liquid Cooling Package LCP Inline CW, LCP Inline CWG, product specification / Rittal / state 10 September 2026 / https://www.rittal.com/pdf-creator/variant/in-en/3312540