Skip to main content
Datacentre Powerdraw and Heat Calculator
Size a GPU build against real datacentre sites, and see how they connect to the grid. Starting with Victoria, expanding statewide and then countrywide. Hardware baselines: NVIDIA DGX H100/H200 user guide (10.2 kW per node, 38,557 BTU/hr, 1105 CFM), NVIDIA DGX GB rack guide (~120 kW per GB200 NVL72 rack) and HPE GB300 NVL72 datasheet (132-140 kW per rack). Grid data: AEMO 2026 ISP, VicGrid, DataVic (compiled October 2026).
Australia
Text size:
A-
100%
A+
High contrast
Step 1: Pick a site
Datacentre site
Map: sites, powerline routes and substations
Pipeline overview
Datacentre pipeline. Each row lists a site, its status, its capacity in megawatts of IT load, and a bar showing relative size.
Site
Status
Capacity (MW)
Relative size
Assumptions and sources
Assumptions: all IT electrical power becomes waste heat (energy conservation); facility load = IT load times PUE; airflow from Q = BTU/hr divided by (1.08 times temperature rise in degrees F); DGX H100 heat uses NVIDIA's published 38,557 BTU/hr per node (about 11.3 kW); GB200 NVL72 uses NVIDIA's ~120 kW per rack; GB300 NVL72 uses HPE's 132-140 kW per rack (140 kW design max); water use = heat load times water intensity (L/kWh) at full-year operation; energy cost = facility draw times 8760 hours times price. Powerline routes and substations marked "inferred" are geographically plausible assignments from public network documents, not confirmed connection designs.
Sources:
NVIDIA DGX H100/H200 User Guide ,
NVIDIA DGX GB Rack Scale User Guide ,
NVIDIA GB200 NVL72 ,
NVIDIA GB300 NVL72 ,
AEMO 2026 Integrated System Plan ,
VicGrid (Terminal Stations in Victoria, Western Metropolitan Melbourne Reinforcement RIT-T) ,
DataVic Energy Facilities ,
RenewMap Victoria datacentre tracker (October 2026), JLL Data Centre Market Dynamics (H1 2025), AirTrunk (December 2025), Firmus Project Southgate (October 2025), NEXTDC, Microsoft Local. Data compiled 2026-10-07. Verify before investment or engineering decisions.