COMPUTE & ENERGY PLANNER
See the resource assumptions.
Compare a clearer scenario.
Explore accelerator operating hours, electricity and an indicative energy cost using your own inputs. This is an arithmetic scenario, not a workload benchmark or a capacity quotation.
TRANSPARENT METHOD
What the calculation includes.
Arithmetic
Operating hours = hours per day × days in the month.
Accelerator-hours = accelerator count × operating hours.
IT power (kW) = accelerator count × entered watts ÷ 1,000 × power percentage ÷ 100 + other IT watts ÷ 1,000.
IT electricity = IT power × operating hours. Facility electricity = IT electricity × PUE. Electricity cost = facility electricity × price per kWh.
Interpretation and limits
PUE is the ratio of facility energy to IT equipment energy over a consistent boundary. See the U.S. Department of Energy metering guide. Here a user-entered constant is applied to an illustrative operating schedule.
The calculation excludes electricity outside that schedule, hardware, cloud-service charges, software, support, implementation, taxes and demand charges. A managed-service bill is not the same as its underlying energy cost.
Validate actual workload performance separately. These power assumptions cannot establish training time, tokens per second, GPU memory needs or electrical-system design.
Context: IEA: Powering Artificial Intelligence (14 October 2026). The formulas are disclosed arithmetic, not an IEA sizing method.
START WITH A BETTER QUESTION
Where could AI make a difference?
Turn an idea into a focused, measurable pilot brief.