AI Energy Claim Evaluation Record

AI Energy Claim Evaluation Record

Required boundary

An interpretable energy claim identifies the unit, denominator, system boundary, workload, model or service, training or inference phase, time, geography, facility overhead, electricity mix, measurement or modeling method, allocation rule, and uncertainty.

Electricity and power are different. Kilowatt-hours measure energy. Kilowatts or megawatts measure a rate. Emissions require an electricity mix and time or contractual accounting method.

Aggregate evidence

The IEA's 2025 Energy and AI report estimates that data centers used about 415 TWh in 2024, around 1.5 percent of global electricity, and models roughly 945 TWh in 2030 in its base case.

Those are global data-center values. They include more than generative AI and cannot be divided by an assumed prompt count to create a reliable per-prompt number.

The IEA's scenarios range widely because adoption, model and hardware efficiency, utilization, infrastructure, and bottlenecks are uncertain.

The IEA's April 2026 update says global data-center electricity demand grew in 2025 and reports additional investment and power-procurement context. It remains system-level evidence.

Claim audit

First identify what the number measures. Find the denominator. Separate measured from modeled and present use from forecast. Check geography and time. Separate IT load from facility use and electricity from emissions. Preserve the range and sensitivity.

Efficiency can reduce energy per task while total demand grows because more tasks run or tasks become larger. Both statements can be true.

Conclusion boundary

A global electricity projection can support infrastructure planning. It cannot establish the footprint of one Microsoft, Slack, or other workplace feature without vendor and workload measurements.