AI Data Center Power & Cost Calculator
Estimate the electricity load and cost of running a GPU cluster, from the chip level up to the facility meter.
Inputs
Allowed range: 1–5,000,000 (whole numbers only)
Allowed range: 1–3,000
Allowed range: 0–200
Allowed range: 1–4
Allowed range: 0–100
Allowed range: 0–24
Allowed range: 1–366 (whole numbers only)
Allowed range: 0–10
Worked example: electricity cost for 10,000 GPUs
Using the calculator's illustrative defaults: 10,000 GPUs, 700 W per GPU at full compute load, 80% simplified utilization, 15% additional IT overhead, PUE of 1.3, 24 hours per day, 365 days per year, and USD $0.12 per kWh.
- GPU load: 10,000 × 700 × 0.8 ÷ 1,000 = 5,600 kW.
- Total IT load: 5,600 × 1.15 = 6,440 kW.
- Facility load: 6,440 × 1.3 = 8,372 kW (8.372 MW).
- Annual energy: 8,372 × 24 × 365 = 73,338,720 kWh.
- Annual electricity cost: 73,338,720 × $0.12 = $8,800,646.40.
These assumptions describe an example, not a measured facility or a utility quote. The simplified utilization model excludes idle power and changing workloads, and the cost excludes demand charges and taxes. This example stays fixed when you edit the calculator above. Live calculator results use your entered values.
For the grid-connection context, read Why AI Data Centers Have a Power Problem Nvidia Can't Solve.
How this is calculated
1. GPU load
GPU load (kW) = GPU count × GPU power draw at full compute load (W) × utilization ÷ 1000
Utilization is used here as a simplified linear scaling factor on top of the GPU's power draw at full compute load — not a measured power curve. In reality, GPUs draw meaningful power even at idle, and real power draw does not necessarily scale linearly with utilization. This model ignores both effects, so the result is a planning approximation, not a measured or guaranteed load.
2. Total IT load
IT load (kW) = GPU load × (1 + additional IT overhead %)
IT overhead accounts for CPUs, memory, storage, networking, and power supplies around the GPUs themselves.
3. Facility load
Facility load (kW) = IT load × PUE
Power Usage Effectiveness (PUE) captures the overhead of cooling, power distribution losses, lighting, and other non-IT facility draw. A PUE of 1.3 means for every 1 kW delivered to IT equipment, the facility draws 1.3 kW at the meter.
4. Energy consumption
Daily energy (kWh) = Facility load (kW) × hours operated per day
Annual energy (kWh) = Daily energy × operating days per year
Monthly energy (kWh) = Annual energy ÷ 12Calculation convention: monthly figures are always annual energy divided by twelve, not daily energy multiplied by a fixed day count. This keeps the monthly and annual numbers internally consistent regardless of how many days per year you enter.
5. Electricity cost
Annual cost = Annual energy (kWh) × electricity price per kWh
Monthly cost = Annual cost ÷ 12Cost uses a single flat electricity price you provide. It does not model tiered pricing, demand charges, or currency conversion — QAR and USD are treated as independent input currencies, not converted between each other.
Input bounds and validation
Every field has a stated, realistic range, and GPU count and operating days per year must be whole numbers. If any field is empty, out of range, or otherwise invalid, the calculator shows which fields need correction instead of guessing on your behalf — it never substitutes a default or rounded value for something you typed, so a result you see always corresponds exactly to the values shown in the fields. The shareable link only updates once every field is valid, for the same reason. If you've opened a shared link, the calculator reads its values as soon as the page loads in your browser — you may see a brief "Loading calculator…" message rather than a result computed from the wrong numbers while that happens. Calculations run in your browser. After valid edits, inputs may appear in the page address for sharing. Opening or reloading that address sends its query parameters to the hosting service, where they may appear in server logs.