The Core Principle: Conservation of Energy
Every watt of electrical power delivered to IT equipment — servers, GPUs, storage, and networking gear — has to go somewhere according to the first law of thermodynamics. In a data center, essentially all of that power ends up as heat. A tiny fraction leaves as light (status LEDs) or sound (fan noise), but for any practical sizing purpose, this fraction is negligible. The standard data center design assumption is therefore straightforward: IT load in kilowatts equals heat load in kilowatts, and the cooling system has to reject that same amount of energy to keep equipment within its operating temperature range.
Why This Matters for Cooling Plant Sizing
Because IT load converts essentially 1:1 to heat, cooling plant sizing starts directly from the facility's IT power budget rather than from a separate thermal survey — a data center provisioned for 4 MW of IT load needs cooling infrastructure sized to reject roughly 4 MW of heat (plus a design margin for growth and part-load inefficiency), regardless of what specific equipment is installed. This 1:1 relationship is what allows early-stage cooling capacity planning to happen well before final rack layouts and equipment selections are finalized — the electrical load budget alone is enough to size the chiller plant's order of magnitude.
Converting Between Units: kW, BTU/hr, and Tons
Data center cooling capacity is specified in different units depending on context — electrical/IT teams think in kilowatts, while mechanical/HVAC teams and chiller nameplates typically use tons of refrigeration or BTU/hr. The conversions are fixed: 1 kW = 3,412.14 BTU/hr, and 1 ton of refrigeration = 12,000 BTU/hr = 3.517 kW. So a facility with 4,000 kW of IT load needs to reject roughly 4,000 × 3,412.14 = 13,648,560 BTU/hr, or about 1,137 tons of cooling capacity, before adding any design margin.
Why a Safety/Design Margin Is Still Added on Top
Even though the IT-load-equals-heat-load relationship is a hard physical constraint, cooling plants are still designed with a margin above the nameplate IT load figure — typically 10-20% — to account for: growth headroom (racks are rarely populated to their absolute maximum from day one), non-IT heat sources (lighting, people, envelope gains, though these are usually small relative to IT load in a dense data center), and redundancy/part-load operating margin (N+1 or 2N cooling plant configurations need to handle full load even with one unit offline). The margin isn't compensating for uncertainty in the physics — it's compensating for real operational and redundancy requirements layered on top of a well-understood baseline.
What Changes Between Liquid and Air Cooling — and What Doesn't
The total heat that must be rejected doesn't depend on whether it's removed via air-cooled CRAH units, direct-to-chip liquid cooling, or immersion — that total is fixed by the IT load itself. What changes between cooling methods is how efficiently that heat is moved out of the facility, how much auxiliary energy (fans, pumps) is required to move it, and at what temperature the heat can be usefully rejected to the outdoors or to a heat-recovery system. This is why comparing cooling methods (see the companion Liquid Cooling vs. Air Cooling explainer) is fundamentally a question of transport efficiency, not a question of how much total heat exists to remove.