Explore how equipment and power shape an AI deployment.
Loading the scenario model…
Behind the claim
Opening the evidence record…
Reference facility power: 14.4 MW. Average facility power: 10.8 MW. Annual facility energy: 94.61 GWh.
≈ 120 kW / rack · 72 GPUs
One 72-GPU rack. Approximately 120 kW is a vendor reference envelope, not a guaranteed electrical design limit.
Illustrative starting assumptions. Change them to explore your scenario.
14.4MW
For 100 racks at reference load and 1.2 PUE.
12 MW IT × 1.2 PUE = 14.4 MW
Illustrative capacity estimate. PUE is an energy ratio applied here as a simplified power multiplier; this is not a guaranteed peak or connection requirement.
9 MW IT × 1.2 PUE = 10.8 MW
One constant PUE is applied to reference and average load; real facility overhead varies with operating conditions.
94.61GWh / year
10.8 MW × 8,760 h ÷ 1,000 = 94.608 GWh/year
A 365-day planning year at the stated average power load. No separate ramp or downtime schedule is inferred.
7,200GPUs
100 racks × 72 GPUs/rack = 7,200 GPUs
An equipment count, not a comparison of compute performance across GPU generations.
9MW thermal
9 MW IT ≈ 9 MW thermal
Steady-state conversion of IT electrical energy to heat. Facility losses and the engineering design of heat removal are excluded.
Fraction of reference electrical power, not GPU workload utilization. Changes average power and energy, not installed equipment.
Total facility energy ÷ IT energy. The same assumed ratio is used at reference and average load.
No water assumption selected or no benchmark available.
Water estimation is off. Choose a historical benchmark or enter site WUE to estimate site water consumption.
This rack uses liquid cooling alongside air-cooled components. Liquid cooling alone does not determine site water consumption.
Reference profiles describe different equipment, not equivalent performance or a ranking of current products.
Pin this scenario, then change an assumption. Compare the power, energy, and equipment counts on the same scale.
Power is allocated to the modeled equipment and its facility overhead. External networking and storage loads, redundancy design, transients, transmission losses, construction and embodied resources, and site constraints are excluded. This is a reference planning estimate, not an electrical or mechanical design.
100 racks entered by you
Whole deployment units. A power budget is evaluated at reference load, independently of average power load.
100 racks × 120 kW/rack ÷ 1,000 = 12 MW
Published equipment power scaled to the deployment count.
14.4 MW facility − 12 MW IT = 2.4 MW
Modeled cooling and electrical overhead at reference load.
12 MW × 75% = 9 MW
Average power load is an electrical assumption, not GPU workload utilization. Zero represents zero average modeled IT draw, not a prediction of idle power.
9 MW × 8,760 h ÷ 1,000 = 78.84 GWh/year
IT energy excludes facility overhead and is the denominator used for site WUE.
Available in power-budget mode.
The remainder after allocating whole equipment units. It is not a modeled redundancy reserve.