GigaCapacity
Free tool
How much power and cooling does an AI GPU rack need?
Updated June 14, 2026

Rack power

AI Rack Power & Cooling Estimator

Use this estimator to translate GPU count, server overhead, rack density, and PUE into rack kW, heat rejection, and facility power.

The output helps buyers decide whether ordinary air cooling is enough, whether rear-door or direct-to-chip cooling should be tested, and what to ask a colocation provider.

Calculator inputs

Scenario defaults are editable. Replace them with current quotes, utility tariffs, tax counsel inputs, or engineering values before relying on the output.

Air or rear-door review

Modeled outcome

Request facility evidence for breaker capacity, heat rejection, liquid loop readiness, and commissioning at this rack density.

GPU load

358 kW

IT load

542 kW

Facility load

677 kW

Rack density

16.93 kW

Cooling load

1,849,042 BTU/hr

Cooling load

154 tons

Annual energy

5,934 MWh

Annual energy cost

$0.5M

Decision checks

  • IT load: 541.9 kW
  • Rack density: 16.9 kW/rack
  • Cooling load: 154.1 tons

How does the rack power model work?

The estimator multiplies GPU count by board power and a server-overhead factor, applies a power safety margin, divides the load across racks, and converts IT power into facility power through PUE. Heat load is reported in kW and tons because nearly all IT power becomes heat. Cooling fit is based on sustained rack density bands, with buyer verification required for allowable inlet conditions, liquid supply temperature, redundancy, leak detection, water treatment, and service procedures.

What should buyers verify before using the result?

  • GPU model, utilization, power capping, network design, and power supply efficiency can materially change the result.
  • A rack-density fit label does not prove a facility can accept the deployment; verify breaker, busway, floor loading, heat rejection, and operating procedures.
  • Liquid cooling type should be matched to actual server cold-plate or immersion certification requirements.

Related guides and tools

Which sources support this rack power model?

SourceUse in this toolLink
ASHRAE TC 9.9 liquid cooling white paperLiquid cooling design context for high-density data centers.Open
ASHRAE 2021 equipment thermal guidelines reference cardThermal class reference for air and liquid-cooled IT equipment.Open
Data Center Dynamics coverage of ASHRAE liquid cooling guidanceIndustry context for newer liquid cooling guidance.Open
Netrality high-density colocation for AI and GPU infrastructureHigh-density rack power and GPU colocation context.Open
TRG guide to colocation for GPUsGPU colocation power and density buying criteria.Open
Flexential liquid cooling for high-density colocationHigh-density colocation and liquid cooling fit context.Open
Uptime Institute Global Data Center Survey 2025Data center efficiency and operations benchmark context.Open
LBNL PUE comprehensive examinationFoundational PUE metric definition and limitations.Open

FAQ: How much power and cooling does an AI GPU rack need?