The ΔT you assume is the plant you buy

There is one number in a data hall cooling design that quietly decides how much equipment you buy, how much space the plant takes, and how much fan energy the facility burns for the next fifteen years. It is not the IT load. Everybody argues about the IT load.

It is ΔT.

The arithmetic is trivial. That is the problem.

Sensible heat into a moving fluid is one equation, and it is the same equation whether the fluid is air in a hot aisle or water in a riser:

Q = ṁ · cₚ · ΔT

Rearranged for the thing you actually have to size:

V̇ = Q / (ρ · cₚ · ΔT)

Volume flow is inversely proportional to ΔT. Nothing subtle happens here. A 1 MW hall at a 10 K air rise needs roughly 83 m³/s of supply air. The same hall at 14 K needs about 59 m³/s.

That is a 29 % reduction in air, for a number somebody typed into a spreadsheet. It is fewer units, smaller ducts, less plant room, less fan power, and a meaningfully different capital cost. And the calculation supporting it fits on one line.

The trouble is that a one-line calculation attracts no scrutiny. Nobody asks a design review to defend V̇ = Q/(ρcₚΔT). So the assumption rides through untouched, and by the time anyone measures the real ΔT the plant is bought.

A large measured ΔT is not proof of anything

Here is the part that catches people. A high return-air temperature looks like success. It looks like heat transfer.

It very often is not.

Return air gets hot for three quite different reasons, and only one of them is the one you wanted:

1. Real heat pickup. Air passes through the servers, takes their heat, and leaves. This is the case the equation assumes.
2. Bypass. Supply air short-circuits back to the return without ever passing through a rack. This does nothing useful, and it lowers the return temperature — so at least it fails in the direction of honesty.
3. Recirculation. Hot exhaust is drawn back across the front of a rack. The equipment gets warm inlet air, and your return sees a temperature that suggests the plant is working beautifully. Meanwhile the inlet temperature at the top of the cabinet is climbing toward the point where the servers throttle themselves.

Recirculation raises measured ΔT and degrades the thing the cooling exists to protect. If your evidence for a large design ΔT is "we measured it on a similar facility," you need to know which of these three you measured.

Containment is what buys you the ΔT

The reason to commit to hot- or cold-aisle containment is not neatness. It is that containment is the mechanism that makes the assumed ΔT physically true. It removes path 2 and path 3 and leaves you with path 1.

Which gives a simple rule:

If you have not committed to containment, you have not earned the ΔT. Design for the smaller number.

I would rather install a slightly larger air path against a conservative ΔT and find out I have headroom, than design against 14 K, discover 9 K in commissioning, and find that the fix is a plant room that is already full.

The redundancy arithmetic has the same problem

The other number that gets asserted rather than calculated is the redundancy label. "N+1" is a claim about capacity, and it is true only if the remaining units can carry the full load at their real, derated capacity on a design-day outdoor condition — not their nameplate rating in a catalogue at 35 °C.

Two things routinely go wrong:

• The unit count assumes nameplate capacity, so at the design-day condition N+1 is quietly N.
• The stated N does not match the installed count, because the load grew after the plant was selected and nobody re-ran the sum.

Both are arithmetic. Both are checkable in an afternoon. Both are found far more often during a failure than during a design review.

I put this check into the cooling load calculator on the bench, because it is exactly the kind of thing a tool should refuse to let you skip.

PUE does not change the air you move

One more that comes up constantly. A mechanical PUE multiplier scales the heat you must ultimately reject outdoors. It does not change the air or the water you must move inside the hall. Those are set by the IT load alone.

Multiplying the internal airflow by PUE oversizes the whole chilled water plant. I have seen it happen, and it is an expensive way to be conservative, because it buys margin in the one place you did not need it.

What I actually do

• Write the ΔT down as an explicit assumption in the design basis, with the containment strategy it depends on named alongside it. If the containment gets value-engineered out later, the link is on the record.
• Size the air path for the conservative ΔT and the plant for the design one, so the mistake, if it comes, is recoverable by turning a fan up rather than by buying a building.
• Ask for measured inlet temperatures at the top of the cabinets, not the return temperature at the unit. The return tells you about the room. The inlet tells you about the equipment you are being paid to protect.
• Re-run the redundancy arithmetic against derated capacity at the design-day condition, every time the load estimate moves.

None of this is difficult. It is one equation and some honesty about where the inputs came from. But the equation is so small that it slips past review, and the assumption behind it ends up being the largest single decision in the mechanical design.

Write it down. Then defend it.