Data Center Edition

Lab · Planning

The Expansion Brief

Plan a 2 MW hall. Three decisions reveal the trade between space, water, energy and testing.

4 minDecide and compare

A 2 MW hall. Three decisions: rack density, heat rejection and commissioning. Compare what each choice means for space, water, electricity and handover.

Read the explanation

The brief: a new hall for 2 MW of IT equipment, planned in three decisions. There is no single best arrangement; the choices move effort, water and electricity between systems.

Density. The hall can hold 100 racks at 20 kW or 20 racks at 100 kW. That is 2 MW of heat either way (magazine page 102). For the 100 kW racks, this brief assumes liquid cooling at the rack captures 80 per cent of the heat, as the magazine does (page 18): 1.6 MW goes into water and 0.4 MW into room air, which still needs its own cooling. At 20 kW a rack, air cooling remains an option, so the split is a design choice the brief does not fix. Power and heat are five times as concentrated at each 100 kW position, and power distribution, heat removal and access all face that.

Heat rejection. A steady 2 MW for a year (8,760 hours) is 17,520 MWh of heat, and either route has to send all of it outside. An evaporative cooling tower evaporates up to about 26,000 m³ of water a year if all of that heat leaves by evaporation (17,520 MWh × 3.6 GJ per MWh ÷ 2.4 GJ per m³), with no compressor in this simplified model. That is the most that could evaporate, not the water a tower draws: real towers also shed some heat without evaporating, so less water evaporates, and lose further water to blowdown and drift, so a tower draws more water than it evaporates (magazine page 58). Dry coolers with a chiller for the hottest hours evaporate no water at the site, but the chiller’s compressor needs electricity: the heat during chiller hours divided by an illustrative COP of 5. With the chiller on for 10, 30 or 50 per cent of hours, that is about 350, 1,050 or 1,750 MWh a year, and the dry coolers send that electricity outside as heat on top of the 17,520 MWh. Actual COP varies with temperatures and load, so real compressor electricity can be higher or lower. The share depends on climate and design temperatures; liquid cooling at the rack can often run warmer water, which can mean fewer chiller hours (magazine page 34). No real city is modelled. Electricity generation can also use water upstream (IEA, Energy and AI, 2025), so moving load from water to electricity moves part of the impact rather than removing it.

Handover. Every component has passed its own tests, commissioning levels L1 to L3. Functional performance testing at load (L4, each system) and integrated systems testing (L5, the systems together) run before customers move in, which takes extra days. In the magazine’s imagined case (pages 85 to 91), testing at load shows one branch with no flow: a valve left closed, which is corrected and retested before customer equipment depends on it. The magazine places that test at L4 and L5 (page 94). Accepting the component tests keeps the planned date, but L4 and L5 are not run before move-in and the interactions between systems have not been proven at load, so a fault like a closed branch valve would surface later, with live equipment. The ASHRAE, NEMA and PNNL framework (June 2026) describes integrated systems testing, level L5, as the level where systems are tested together.

Your hall and the other path. Racks: 100 at 20 kW, or 20 at 100 kW. Heat into water and into air: not fixed, or 1.6 MW and 0.4 MW. Cooling water evaporated: up to about 26,000 m³ a year with the tower, none at the site with dry coolers. Extra compressor electricity: none with the tower, about 1,050 MWh a year with dry coolers and a chiller for 30 per cent of hours. Handover: L4 and L5 at load with the closed valve fixed before move-in, or the planned date after L3 with such a fault surfacing later. Fans, pumps and the IT load’s own electricity are excluded from both.

What each hall trades. Density did not change the heat, only where it is concentrated and how much goes into water. The tower and the dry coolers trade up to about 26,000 m³ of evaporated water a year against about 350 to 1,750 MWh of compressor electricity, and climate and design temperatures decide where a real hall lands in that range. Testing at load spends extra days before move-in so that a fault like a closed branch valve turns up while no customer depends on it. Magazine page 105 makes the same point about a supplier’s quote: what one line leaves out reappears in another.

Both halls can work. The choices decide where the effort, water and electricity go.

Behind the model

Assumptions

  • The hall’s IT load is a steady 2 MW all year, and all of that electricity ends up as heat.
  • For the 100 kW racks, liquid cooling at the rack captures 80 per cent of the heat. This is the magazine’s assumption (page 18), not a property of liquid cooling.
  • No capture rate is assumed for the 20 kW racks; air cooling remains possible and the split is a design choice.
  • Water evaporated is an upper bound on evaporation, not on the water a tower draws: all heat leaving by evaporation, at about 2,400 kJ per kg near cooling-tower temperatures and 1 kg per litre (2.4 GJ per m³). Heat shed without evaporation would lower it; blowdown and drift, which add to the water drawn, are not modelled.
  • The tower route has no compressor in this simplified model.
  • The chiller has an illustrative COP of 5 and carries the whole 2 MW during chiller hours. Its share of the year (10, 30 or 50 per cent) stands in for climate and design temperatures; no real place is modelled. A real chiller works harder in hot weather, when it runs most, so its COP then is usually below 5 and these figures lean low.
  • The compressor’s electricity leaves through the dry coolers as heat, on top of the IT heat (magazine page 34).
  • The three decisions are treated as independent. In a real hall they interact: liquid cooling at the rack can often run warmer water, which can reduce the hours that need a chiller (magazine page 34).
  • Component tests cover commissioning levels L1 to L3. Testing at load adds L4 and L5 before move-in, where the magazine places its imagined valve test (page 94).
  • Fans, pumps and the IT load’s own electricity are excluded from both halls.
  • The handover outcomes follow the magazine’s imagined commissioning test. No durations, costs or uptime figures are given.
  • Numbers are rounded: water to two significant figures, electricity to the nearest 10 MWh.

Sources

  1. DOE Fundamentals Handbook: Thermodynamics, Heat Transfer, and Fluid Flow, Vol. 1, DOE-HDBK-1012/1-92
    US Department of Energy · June 1992
  2. NIST Chemistry WebBook (SRD 69), water: thermophysical properties of fluid systems
    NIST · Undated web page
  3. Best Management Practice #10: Cooling Tower Management
    US DOE Federal Energy Management Program · Undated web page
  4. Purchasing Energy-Efficient Electric Chillers
    US DOE Federal Energy Management Program · October 2024
  5. AI Data Center Energy Performance Framework: Commissioning & Performance Validation
    ASHRAE, NEMA and PNNL · June 2026
  6. Energy and AI
    International Energy Agency · April 2025
  7. Data Center Edition, Issue 01, pages 9, 18, 34, 58, 85 to 94, 102 and 105
    Data Center Edition · October 2026

Search stories

Your reading list

Your reading list is saved in this browser.