High Knowledge · Planning tools
Data centres and Edge AI:
Calculate your requirements
Compare energy scenarios and assess your AI requirements using your own assumptions, transparent calculations and a clear next planning step.
Data centres · Energy efficiency
What changes with a different PUE?
Compare the annual electrical energy demand of two infrastructure scenarios with the same average IT load. Enter the project-specific target value for your planned Blue DC scenario.
The difference concerns total electrical energy demand at the same IT load. It is not a Blue DC guarantee or evidence of CO₂ reduction, waste heat use or legal compliance.
Calculation and limitations
Energy demand (MWh) = average IT load (kW) × operating hours × PUE ÷ 1,000. Change = (planned scenario − reference) ÷ reference × 100. Relative change is undefined when reference consumption is zero.
Both PUE values must use the same system boundary and time period. Weather, partial load, redundancy, cooling and auxiliary equipment affect actual values. Annual comparisons require representative annual values. This calculation uses constant or energy-weighted figures. Waste heat credits are not deducted from electrical energy demand.
Sovereign AI · On-premises
How much AI capacity do you need?
Estimate daily request volume and the memory required for model weights alone. The results provide a starting point for benchmarking your tasks and data.
The time average is not a required benchmark throughput: concurrency, peak loads and response times determine actual sizing. Weight memory excludes KV cache, runtime buffers and quantisation overhead. It is not a GPU recommendation.
Calculation and next planning steps
Requests/day = users × requests per user. Tokens/day = requests × (input tokens + output tokens). Average output rate = requests × output tokens ÷ usage hours ÷ 3,600. Weight memory in GiB = billion parameters × 10⁹ × bits per parameter ÷ 8 ÷ 2³⁰.
Detailed sizing also considers context length, concurrent sessions, time to first token, desired output speed, knowledge retrieval, document processing, redundancy and operating windows. For MoE models, enter the total number of stored parameters; active parameters alone do not describe weight memory. Sovereignty also depends on data flows, permissions, maintenance and external connections.
From an initial estimate to sound planning.
Use the results for your first project discussion. We assess load profiles, availability and operating conditions to develop suitable infrastructure.
Feasibility study for your data centre · Sovereign AI in your organisation
Calculator inputs are processed locally in your browser. Example values provide guidance and are not guaranteed product specifications.