# Energy forecasts, geography and measurement boundaries

Canonical: https://semiaifoundry.com/learn/atlas/energy-forecast-boundaries/
Revision: 2026-10-07
Rights: https://semiaifoundry.com/ai-atlas/licenses.html

Global AI infrastructure · Atlas v1.3.0

# Energy forecasts, geography and measurement boundaries

Distinguishes estimated electricity use from conditional demand projections.

[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#energy-forecast-boundaries)Facility phases to region and world; geography schematic

Learning objectiveRead a dated demand forecast without confusing geography, AI scope or power and energy.

## The mechanism

Electricity demand reports combine observations, estimates and scenarios. A global all-data-center forecast includes many uses beyond AI, while an AI-focused subset requires its own definition. The IEA’s April 2026 update and LBNL’s June 2026 US report differ in geography, publication timing and methodology. Their totals cannot be added or treated as competing measurements of the same scope. Historical forecast revisions are useful evidence about changing assumptions, but an older projection should not silently replace a newer one. Power is an instantaneous rate; annual energy integrates changing load, including non-IT overhead when the report’s boundary includes it.

1.  Identify the report date, geography and whether AI is the whole scope or a subset.
2.  Separate estimated historical demand from projected future demand.
3.  Check whether the measure includes cooling and other facility loads.
4.  Compare revisions only after aligning definitions and explaining changed assumptions.
5.  Keep local deliverability and useful workload capacity separate from broad annual electricity totals.

## Science and assumptions

### Energy from average power

$$
E=P_{\mathrm{avg}}\Delta t
$$

Energy is average power times the duration over which that average applies.

GW times hours gives GWh; divide GWh by 1000 for TWh. Installed peak power is not an interval-average load.

Symbols and units

E

Energy over the averaging interval

GWh

P\_avg

Average real power over that same interval and boundary

GW

Δt

Interval duration

h

Pavg in GW and Δt in hours gives GWh; divide by 1,000 for TWh.

Assumptions Average must cover the same interval and boundary; installed or contracted peak power is not Pavg.

## Engineering example

Synthetic constant 1 GW load over 8,760 hours uses 8.76 TWh. A 1 GW nameplate facility does not necessarily use that energy because its average operating load can be lower.

## Distinctions and tradeoffs

A global data-center forecast is not an AI-only measurement, and a 2030 scenario is not observed consumption.

Broader forecasts help planning but hide local grid and workload constraints; granular estimates need more uncertain input data.

AI buildout affects electricity systems, but no annual energy total provides a universal conversion into model throughput or capability.

## Specifications and interfaces

Units

MW power and MWh/TWh energy

1 TWh equals one million MWh; time is essential.

Status

Estimated baseline versus forecast scenario

Both need source date, observation year and scope.

### Interfaces

Metering, shipment and scenario inputs Input

Defined geography, device mix and operating assumptions

Demand estimate or forecast Output

Scoped energy with period and uncertainty

## Check your understanding

### Why can a US report not be added to a global total?

1.  US electricity uses different physical units
2.  Forecasts can never discuss regions
3.  The global scope already includes the United States

Answer and reasoning

The global scope already includes the United States

The geographies overlap, so adding them double-counts the US contribution.

## Evidence and scope

IEA baseline and projection are separate claim records. LBNL’s laboratory abstract was retrieved; inaccessible full-report methodology is not represented as independently rechecked.

No current calculator directly models this lesson’s main mechanism. Use the library for related bounds while retaining their stated omissions.

## Sources

-   [Key Questions on Energy and AI: Executive Summary](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)
-   [United States Data Center Energy Usage Report: 2025 Update](https://bies.lbl.gov/publications/united-states-data-center-energy-2025)

[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/energy-forecast-boundaries-a7a1ef8a72b5c20c.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)

## Continue exploring

### Foundations

-   [Dynamic load, reserve and power control](https://semiaifoundry.com/learn/atlas/rack-dynamic-power/)

### Related lessons

-   [Water withdrawal, consumption and cooling circulation](https://semiaifoundry.com/learn/atlas/water-boundary-accounting/)
-   [Grid interconnection and deliverable electrical capacity](https://semiaifoundry.com/learn/atlas/grid-interconnection-readiness/)