Retrofit Liquid Cooling Without Rebuilding the Data Center

Operators rank ease of retrofit into existing infrastructure as the single most important test of whether a direct liquid cooling system is viable, ahead of operating cost and ease of maintenance. The hardware that passes that test is already on the market and mounts on a standard rack. What decides the project is the water temperature the existing chiller plant is asked to produce, and the available retrofit paths pull that setpoint in opposite directions. A rear door rated at 50 kW and a cold plate loop carrying a comparable load want facility water roughly 20 °C apart.

Source: Uptime Institute, Cooling Systems Survey 2025, UII Data Report 181 (July 2025)

Three retrofit paths exist and they differ by what they ask of the building

The choice among them is settled at the chiller plant and the pipe route, not at the rack.

The first path is a liquid-to-air coolant distribution unit, a self-contained pumping and heat exchange skid that captures heat in coolant and rejects it straight back into the room air. It needs no facility water connection at all. Schneider Electric describes this type as the easiest to add to a conventional air cooled data center and most cost effective for deployments of 1 to 10 racks, precisely because the white space air system, not a new pipe run, carries the heat away.

The second path is the rear door heat exchanger, an air-to-water coil that replaces the rack’s rear door and is fed either from facility water or from a coolant distribution unit. The third is direct liquid cooling, meaning cold plates clamped to the high power components, sold variously as direct-to-chip or D2C, connected through a closed technology cooling system loop to a liquid-to-liquid coolant distribution unit that hands the heat to the facility water system. Schneider Electric scopes the liquid-to-liquid case at roughly 500 kW or more and 10 or more racks.

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

Ease of retrofit topping the viability list is not a preference for convenience. It is operators reporting that the building, not the rack, is the binding constraint on an air to liquid conversion.

A rear door rating is a statement about the water loop, not about the door

Published rear door capacities reproduce from the rating conditions using nothing but the water flow and the water temperature rise, which means the headline number describes the facility loop the door is attached to.

Delta Electronics rates its CoolDoor D50 at 50 kW, with the rating conditions given as return air at 50 °C, inlet water at 12 °C and outlet water at 20 °C, at a rated water flow of 90 LPM. Run that through the steady flow energy equation:

Q = ṁ × c(p) × ΔT

ṁ = 90 L/min ÷ 60 = 1.5 kg/s (water taken as 1 kg/L) c(p) = 4.18 kJ/kg·K ΔT = 20 °C − 12 °C = 8 K

Q = 1.5 × 4.18 × 8 = 50.2 kW

The published figure is 50 kW. The same substitution on the D30, rated at 55 LPM across the same 8 K rise, returns 30.7 kW against a published 30 kW. Both are upper bounds at the stated conditions rather than measured values.

The derivative is the more useful result: at the rated 90 LPM, the D50 gains or loses 6.3 kW for every 1 K of water temperature rise.

Source: Delta Electronics, CoolDoor 30/50 kW technical specifications (accessed September 2026)

That sensitivity is where a legacy data center upgrade runs into trouble, because the inlet condition is not free. The same specification notes that inlet water above 12 °C requires cooling capacity derating, and that inlet water must stay above the room dew point.

Lenovo’s documentation for its Rear Door Heat eXchanger V2 specifies water above dew point at 18 °C ±1 °C for an ASHRAE Class 1 Environment and 22 °C ±1 °C for Class 2, with a required flow between 22.7 and 56.8 liters per minute. Two published rear door specifications, two water temperatures six degrees apart, for what buyers treat as one product category.

Table 1. Rear door capacities and the conditions attached to themkW · °C · L/min

* Each rating is stated at its own air and water conditions; capacities are not comparable across vendors.

* Lenovo figure is the illustrated performance load, not a single rated point.

Source: Delta Electronics, “CoolDoor 30/50 kW Technical Specifications”; Lenovo, “Rear Door Heat eXchanger V2” (both accessed September 2026)

Source: Lenovo, ThinkSystem Heavy Duty Rack Cabinets documentation, Rear Door Heat eXchanger V2 (accessed September 2026)

The 12 °C condition is a claim on the chiller plant, not a property of the product.

The air side sets the second limit and the servers own it

A passive rear door has no fans, so its rated capacity is also an assumption about how much air the installed servers push.

The Delta rating carries an air condition alongside the water condition: 4016 CFM at the D50 and 3812 CFM at the D30. Checking the air side closes the loop.

Converting 4016 CFM gives 1.895 m³/s, and air at 50 °C has a density near 1.09 kg/m³, so the mass flow is 2.07 kg/s. With c(p) for air at 1.006 kJ/kg·K, removing 50 kW requires an air temperature drop of 24 K, which puts the air leaving the door at 26 °C.

The rating is a room neutral condition that assumes a rack drawing intake air at about that temperature.

Source: Delta Electronics, CoolDoor 30/50 kW technical specifications (accessed September 2026)

Lenovo states the same dependency plainly in its specification table: air movement is provided by the servers and other devices in the rack, with an air temperature drop of up to 25 °C between the air leaving the devices and the air leaving the heat exchanger. A half populated rack, missing filler panels, or a fan curve sitting at minimum speed all reduce the airflow the rating assumed.

Direct liquid cooling does not remove this dependency either, it relocates it. Schneider Electric treats server heat capture ratio as a primary efficiency lever and uses 80 percent capture to liquid as its worked example, leaving 20 percent for air. Memory, storage, power delivery and network components stay on the air path.

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

Sizing the residual air system to the pre-retrofit load is the most expensive way to be safe, and sizing it to zero is the fastest way to lose a rack. Neither number is available from a vendor datasheet.

The three paths pull the chilled water setpoint in opposite directions

A rear door retrofit and a cold plate retrofit are not two versions of the same project. They ask the plant to move the other way.

Schneider Electric’s specification guidance for a liquid-to-liquid coolant distribution unit lists entering fluid temperature on the facility side at 27 °C to 42 °C, with state of the art approach temperatures of 2 °C to 4 °C and some vendors as high as 6 °C. Rear door specifications, as above, sit between 12 °C and 18 °C. When a single chiller plant serves both the liquid cooled and the air cooled load, the paper describes the result as a narrow band of usable chilled water setpoints, and puts the operator in front of two options: keep one low temperature plant and forgo the efficiency the liquid cooled load could have delivered, or build a separate high temperature plant and pay for it.

The efficiency at stake is not rhetorical. The same paper cites roughly 1 to 2 percent chiller efficiency gain for every 0.6 °C increase in chilled water temperature, rising to 2 to 4 percent on variable speed machines.

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

Read that way, the rear door is the conservative retrofit in every sense. It reuses the plant the building already has, at the plant’s existing temperature, and it forfeits the economizer hours that were the reason to consider liquid in the first place.

Geometry and pipe routing decide the schedule before thermodynamics does

The constraints that stop retrofit projects are dimensional, and they are all published.

Door depth is the first. The Delta unit is listed at 600 x 345 x 1970 mm, while the Lenovo door is 129 mm deep. Adding 345 mm to the back of every rack in a row moves the hot aisle boundary, and in a hall with fixed containment panels that is a demolition item rather than a door swap.

Weight follows. The Delta doors are 90 kg and 98 kg net, while the Lenovo door is 39 kg empty and holds approximately 9 liters of water. That mass hangs cantilevered on the rack’s rear hinge line, on a floor specified for a lighter cabinet.

Pipe routing is the second. Delta lists 1 inch connections on the D30 and 1 1/4 inch on the D50, top or bottom feed, at a maximum operating pressure of 10 bar. Bottom feed in a raised floor hall means running supply and return through the same plenum that still has to deliver air to every rack that was not retrofitted.

Source: Delta Electronics, CoolDoor 30/50 kW technical specifications; Lenovo, ThinkSystem Rear Door Heat eXchanger V2 documentation (accessed September 2026)

The third is stranded capacity, which is specific to liquid. Schneider Electric works an example of a 300 kW coolant distribution unit provisioned for anywhere between 3 racks at 100 kW and 10 racks at 30 kW. Commit to the low density case and the branch piping is oversized; commit to the high density case and seven of ten distribution runs sit unused. Air has no equivalent failure mode, because air is not plumbed to individual servers.

Table 2. What each retrofit path asks of the existing buildingfacility water, °C

* Water temperatures are the published conditions for each device class, not one interoperable range.

Source: Delta Electronics, “CoolDoor 30/50 kW Technical Specifications”; Lenovo, “Rear Door Heat eXchanger V2”; Schneider Electric, White Paper 210 (2025)

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

Limits of vendor ratings

Every capacity figure here is a vendor rating at vendor conditions, not a measurement. The 50 kW and 30 kW door ratings are stated at specific return air temperatures, water temperatures and flow rates, and the arithmetic above confirms internal consistency, not field performance. No public test standard currently rates coolant distribution units the way ANSI/ASHRAE Standard 127 rates computer room air conditioners, so capacities from different manufacturers are not directly comparable. The Uptime Institute figures come from a survey of 1,033 respondents conducted between April and June 2025, weighted toward enterprise owner-operators and design firms, which is a population that may under-represent hyperscale practice. Structural adequacy of a specific floor, the fire and leak detection consequences of bringing water into a specific hall, and the correct dew point control strategy for a specific climate are engineering review items and cannot be settled from published specifications.

Retrofit homework that starts at the chiller plant

Check the plant’s achievable supply water temperature before evaluating any rack level device. It sets which of the three paths is available.

Ask each vendor for capacity at your water temperature and your rack airflow, not at their rating conditions, and ask in writing.

Measure the actual airflow the installed servers deliver at the rack positions being retrofitted. A passive door inherits it and cannot improve it.

Size the residual air system from the server vendor’s stated heat capture ratio, not from a category average.

Confirm rear aisle clearance and rack rear hinge loading against the specific door depth and weight, since these vary by a factor of two and a half across published models.

If one chiller plant will serve both the retrofitted and the untouched racks, price the separate high temperature plant as the base case rather than the exception.

References

Uptime Institute, “DLC adoption remains slow and steady,” UII Data Report 181, Cooling Systems Survey 2025, July 2025. https://intelligence.uptimeinstitute.com/sites/default/files/2025-07/UI%20Field%20181_Data%20center%20cooling.pdf

Schneider Electric, White Paper 210, “Direct Liquid Cooling System Challenges in Data Centers,” Version 1, 2025. https://media.datacenterdynamics.com/media/documents/WP210_V1_EN.pdf

Delta Electronics, “Rear Door Heat Exchanger (RDHx), CoolDoor 30/50 kW, Technical Specifications.” https://www.deltapowersolutions.com/en/mcis/data-center-cooldoor-30kw-50kw-specifications.php

Lenovo, “Rear Door Heat eXchanger V2,” ThinkSystem Heavy Duty Rack Cabinets documentation. https://pubs.lenovo.com/hdc_rackcabinet/options_42u_heat_exchange_door