Eleven 1.5 MW CDUs Equaling 12 MW Is Coincidence, Not Margin

Eight working units at 1.5 MW nameplate equal 12,000 kW [MODELED]. Ten surviving units at a 1.2 MW design derate also equal 12.0 MW [MODELED]. Both land on the 12 MW [TARGET] billable information-technology load, which is 100 % of that load. They are not the same claim. The first counts groups after each spare is surrendered — 3+1, 3+1 and 2+1, three plus three plus two working units, 8 × 1.5 = 12,000 kW [MODELED]. The second counts the fleet after one failure, at a derate the nameplate is not allowed to impersonate: 10 × 1.2 = 12.0 MW [MODELED]. Eleven times 1.5 is 16.5 MW installed [MODELED]. The 12.0 MW is what remains after redundancy is taken. The hydraulic basis prints that first equality in quotation marks and then demolishes it: the surface tidiness is a coincidence, not design margin [FACT]. The equation ignores two effects that both push the same way — secondary pump heat and nameplate derating.

Eight Working Units Equal 12 MW, Then Pump Heat Shorts It

Pump heat is the one the same document quantifies. The secondary loop has to reject the liquid-side information-technology heat plus the shaft work the pumps put into the fluid, Q_sec = Q_liq + η_h · P_pump, with the heat fraction taken conservatively at 1.00. Carrying the electrical-basis figure of 385 kW [MODELED] for pump and control power, the secondary duty at a liquid fraction of 1.00 is 12,385 kW against 12,000 kW of N capacity — short by 385 kW, or 3.2 % [MODELED]. The document places the crossover at a liquid fraction of about 0.97 [SENSITIVITY].

Nameplate derating is the second effect. It is held open rather than quantified here. Whether 1.5 MW was calibrated on water or on the 25 % glycol mixture the secondary loop is assumed to carry is an unclosed item with a named owner [HOLD]. The nameplate is a capability at a stated primary inlet temperature, secondary supply temperature and temperature difference. It is not a condition-independent constant.

A Second Calculation Route Also Reaches 12.0 MW

A separate parametric engine expands thirty-six scenarios across three branches, three liquid fractions and four temperature differences. It reaches 12.0 MW for the same eleven-unit branch by a different road. It designates one standby unit, counts ten available after a single failure, and applies a 1.2 MW design derate rather than the 1.5 MW nameplate. Its own label on that result is that the branch meets the requirement with zero arithmetic margin and is explicitly not engineering-accepted [MODELED].

Two documents, two redundancy accountings, and the same 12.0 MW. One counts groups. The other counts the fleet. They would diverge immediately if the grouping changed. They already do, one branch over: the twelve-unit branch is 13.5 MW [MODELED] of N capacity in the grouping table and 13.2 MW [MODELED] in the derated engine. The figure survives being recomputed, so it starts to look confirmed. It is not the same claim twice.

Two 3+1 Zones Hold 4.5 MW; the 2+1 Zone Holds 3.0 MW

The zone table shows where the 12.0 MW actually sits. Two zones of 3+1 hold 4.5 MW of N capacity each, 37.5 % of the site [MODELED]. One zone of 2+1 holds 3.0 MW, 25 % [MODELED]. Zone load has to follow those shares, or redundancy is uneven across the floor.

Against a single unit failure, the margin narrows as the liquid fraction rises: 20.2 % at 0.80, 7.3 % at 0.90, 1.8 % at 0.95 [SENSITIVITY], and a deficit at 1.00 — 144 kW in the larger zones and 96 kW in the smaller one [MODELED]. The 7.3 % at a fraction of 0.90 still has to absorb nameplate condition deviation, uneven flow distribution and fouling allowance before it is spendable.

Maintenance Plus Failure, Not Single Failure, Is the Test

The case that removes the arrangement from contention is not single failure. It is planned maintenance with a failure on top. The annual work list makes that pairing unavoidable: heat-exchanger cleaning, pump seal and bearing replacement, filter and side-stream media changes, fluid sampling and control-valve actuator calibration. During any of those the unit is out. Faults do not pause for maintenance.

With one unit in maintenance and one failed, the 3+1 zones drop to two units, 3.0 MW against a zone load of 4,194 kW — short 1,194 kW, or 28.5 % [MODELED]. The 2+1 zone drops to one unit, 1.5 MW against 2,796 kW — short 1,296 kW, or 46.3 % [MODELED]. That zone carries 25 % of the information-technology load, about 2.7 MW at a liquid fraction of 0.90, which the document puts at roughly twenty-one racks of 130 kW [MODELED]. The 130 kW density is a local hydraulic point in that basis. It is not a freeze of rack count.

Both studied electrical topologies offer a path-maintenance-plus-single-failure capability. Overall availability is then capped by the cooling side rather than by the electrical side [MODELED].

Limits and open items

Confirmed: 16.5 MW installed and 12.0 MW of N capacity follow from the stated grouping; the 12.0 MW match with billable IT is a coincidence, not design margin. [FACT]

Modelled and not frozen: the 12,385 kW secondary duty and 385 kW shortfall, the 3.2 % and 0.97 crossover, the 10 × 1.2 MW route, the 13.5 MW and 13.2 MW twelve-unit pair, the zone shares, the 20.2 / 7.3 / 1.8 % single-failure margins, the 144 kW and 96 kW deficits, the 1,194 kW and 1,296 kW maintenance-plus-failure shortfalls, and the 2.7 MW / 130 kW exposure. Every figure inherits an unfrozen liquid fraction, a nameplate whose calibration fluid is undeclared, and a 385 kW pump allowance the same document reworks elsewhere. None is a purchase quantity. The eleven-, twelve- and fifteen-unit branches are held in parallel with none selected. [MODELED]

Open: the liquid fraction; the nameplate calibration fluid; the 385 kW pump allowance versus the same document's bottom-up rework; and whether the owner's service terms require single-failure capability while a unit is in planned maintenance. [HOLD]

No Engineer of Record has reviewed or sealed any of this. No branch is released for purchase.

Which Accounting Produced the 12.0 MW Figure

If the owner's service terms require the arrangement to hold single-failure capability while a unit is in planned maintenance, the eleven-unit branch and the twelve-unit branch both fail by the table above, and the comparison restarts at fifteen. If the terms require only single failure, the eleven-unit branch sits on a margin of 1.8 % at the representative liquid fraction — which is to say, on the coincidence. The question that decides which of those sentences applies is contractual rather than hydraulic. The question worth putting to anyone repeating 12.0 MW is narrower: which of the two accountings produced it, and what happens to the figure when the grouping changes.

© 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., “Eleven 1.5 MW CDUs Equaling 12 MW Is Coincidence, Not Margin,” https://www.kkdatasvc.com/lab/direct-liquid-cooling-and-cdu/11-x-1-5-mw-equals-12-mw-by-coincidence-not-margin/.