Ask the layout author where the air terminals stand when the liquid fraction is written as 1.00. The sensitivity table in the hydraulic basis has four rows, and the last one prints 12,000 kW [TARGET] of liquid-side heat and 0 kW of air-side heat under a liquid fraction of 1.00 [MODELED]. The entry against that row is annotated a theoretical limit. Directly beneath the table the author rules the row out of the design space. A liquid fraction of 1.00 does not exist in engineering. The typical residual air-side share is 3 % to 8 % [MODELED]. Design must proceed at f ≤ 0.97, and air-side terminals may not be configured as zero. That 3 % to 8 % is graded an assumption, not a project figure. It is the only quantitative support the prohibition has.
Even Full Cold-Plate Coverage Leaves Residual Air Heat
The list of what stays on air is specific enough to argue with, which is what makes the bound checkable. Even with every graphics processor, central processor, switch fabric, memory module and voltage regulator under a cold plate, the rack still contains power-supply internal losses, optical modules, management boards, the fans themselves, and network equipment that is not liquid-cooled at all.
Nothing in that list is exotic. Nothing in it is a design choice available to the facility. The fans are the sharpest item: fan power is itself part of the residual, so the air-side load includes the cost of moving the air that removes it.
Air Residual Must Be Designed Independently, Not as Zero
The number the design actually carries is not the bound but the representative value one step inside it. At a liquid fraction of 0.95 the air-side residual is 600 kW; at 0.90 it is 1,200 kW [MODELED]. The heat-rejection basis accepts both as an inherited interface condition and states the rule in the same line it states the values: air-side residual must not be taken as zero. The air-side terminals are to be designed independently rather than treated as a rounding error on the liquid system.
Neither figure is a field result. Both are 12,000 kW multiplied by one minus an assumed fraction. The fraction is the same unclosed input that governs the liquid-side unit count.
f = 1.00 Would Delete Required Air-Terminal Floor Positions
The space basis is where the prohibition stops being a caveat and starts costing something, because a terminal that may not be zero has to stand somewhere in a row.
That document divides the residual by zone and by terminal. At a liquid fraction of 0.95, each of three zones carries 200 kW [MODELED]. At an assumed sensible capacity of 50 kW to 100 kW per in-row or rear-door unit, that is two to four units, and with one spare per zone, three to five [MODELED]. At 0.90 each zone carries 400 kW, giving five to nine [MODELED]. The recommendation is to reserve in-row positions for three as the lower envelope, four as the baseline and six as the upper, and to carry the nine-unit case separately as a row-length increment rather than folding it into the baseline.
Moving the liquid fraction from 0.95 to 0.90 takes in-row air-terminal positions from four to between five and nine, which lengthens each zone's rows by one to five terminal widths [MODELED]. Two earlier documents had quantified that same unclosed input against capacity. This is its path into geometry.
f = 1.00 Makes White-Space Area Look Slightly Better
Reported by the same space basis: a white-space intensity of 39 to 45 m² per MW for the baseline layout, against a typical 40 to 90 m² per MW for direct liquid cooling and 300 to 800 for air [MODELED]. Landing at the bottom of the liquid range is recorded as self-consistent with the premise — only three to six air terminals across the whole site, no ducts, no raised-floor supply — and simultaneously as a reason not to submit that figure as a permit-stage area.
The zero row would make that intensity lower still, and it would do so by deleting a terminal count the same document has just derived. That is the practical damage a theoretical limit does when it escapes its label: it does not produce an obviously wrong number. It produces a slightly better one.
A Residual Below 3% Would Make the 0.97 Bound Wrong
Held open: the 3 % to 8 % residual band is an assumption with no project evidence behind it, and the liquid fraction that sets the 600 kW and 1,200 kW figures is not frozen. Nothing above reflects Engineer of Record review, equipment-supplier confirmation, commissioning acceptance or observed operation. No terminal quantity here is a procurement figure. [HOLD]
The condition that would change the conclusion is narrow and specific. If the information-technology vendor's written per-rack heat allocation returns a residual above 3 %, the bound holds and only the terminal count moves. If it returns a residual below 3 % — by moving power supplies, optics and network equipment outside the rack thermal boundary rather than by cooling them — then f ≤ 0.97 is the wrong bound and the row count derived from it is wrong with it. The bound is an assumption about rack construction, not a law. It is owned by whoever specifies the rack.
© 2026 K&K Data Service Inc. All rights reserved. Reproduction or republication is permitted only with clear attribution to K&K Data Service Inc. and a working hyperlink to the canonical URL of this article. Excerpts must preserve the technical context, maturity labels, assumptions, and limitations. No excerpt may imply project approval, field validation, certification, or endorsement that the original article does not state.
Source: K&K Data Service Inc., “Why 100% Liquid Cooling Is Not a Design Case,” https://www.kkdatasvc.com/lab/direct-liquid-cooling-and-cdu/100-percent-liquid-cooling-does-not-exist/.
Discuss This Constraint. If your project record shows a different result, or the same failure domain under another name, we want to see it. Email inquiry@kkdatasvc.com or use the contact page.