AI’s Energy Footprint Belongs in the Business Case

Look at both sides of the energy equation

AI can be proposed as a way to reduce building energy use while depending on additional computing, storage and communication. Those effects should be considered together, but not combined carelessly. A global data-centre forecast is context for the technology system; it is not an estimate of the electricity consumed by a particular workplace application.

IEA’s 2025 report estimates global data-centre electricity use at about 415 TWh in 2024 and projects about 945 TWh in its 2030 Base Case [IEA]. These totals include non-AI workloads. They cannot be labelled AI-only consumption, and the projected value must remain visibly distinct from the historical estimate.

Define the application boundary

For an individual project, identify where computation occurs and what equipment is added. An on-site server may have a visible electricity load; a cloud service may provide only partial information. Include data transfer, storage and sensor maintenance where they are material and measurable. State exclusions instead of assuming the unmeasured part is zero.

The boundary should match the claim. A statement about reduced HVAC electricity is not automatically a claim about lower lifecycle emissions. Different electricity sources, hardware manufacture and service usage may matter to the broader assessment. Keep each measure separate until there is a defensible basis for combining them.

Figure 1. Global data-centre electricity consumption Source: IEA (2025), Energy and AI, Energy demand from AI, CC BY 4.0. Original Osmos chart; rounded source values; no IEA endorsement.

Prepared 1 September 2026.

What this means: The projected global footprint is context—not a calculation of an FM application’s net benefit.

Choose the simplest sufficient method

Osmos recommends comparing the AI proposal with a simpler rule, schedule correction or engineering intervention. If the simpler option solves the operational problem reliably, additional computational complexity needs a reason. This is not an argument against AI; it is an argument for matching the method to the decision.

Hong and Li recommend documenting computational resources in building-AI studies [HONG]. In procurement, a related request is useful: ask the supplier to describe its model, deployment pattern, update frequency and available resource-use information. The answer may be incomplete, but the gap should be visible.

Protect the quality of the savings claim

The building-side benefit still needs a credible baseline, comparable operating conditions and service guardrails. Do not subtract a rough cloud estimate from an unverified building saving and call the result a precise net benefit. Use ranges or separate reported quantities where uncertainty is substantial.

Distinguish energy, cost and carbon. A cheaper tariff can reduce expense without reducing electricity, and a lower-carbon supply can change emissions without changing consumption. A useful business case explains these differences rather than allowing one favourable measure to stand in for all three.

Illustrative decision rehearsal

Imagine two proposals for an energy diagnostic task. One repeatedly processes a large stream of raw data through a remote model; another uses a smaller local rule and sends only exceptions for further analysis. The more complex system may be better, but that conclusion requires evidence about diagnostic value, operational effort and resource use. Complexity alone is not a benefit.

The project team should compare the same task and service requirement. Ask what additional decisions the complex option enables, how frequently they are needed and whether the extra information changes action.

Avoid using an energy estimate to dismiss a system whose reliability benefit is essential, but make the trade-off explicit.

In the first review cycle, record available resource-use data and explain its boundaries. Supplier estimates may cover inference but not training, or computing but not cooling. Those differences should remain visible.

Do not combine incompatible estimates into a precise total simply to fill a reporting template.

The resulting decision can be expressed in plain language: the system delivered a measured operational benefit, required specified additional infrastructure and left certain resource effects unquantified. That is a more defensible statement than claiming that any AI-enabled efficiency project is automatically sustainable.

It also gives the organisation a practical agenda for better data as the service expands, contracts or moves to a different hosting arrangement.

Set a review point as use expands

A pilot may use little computing because it serves a few assets. Portfolio rollout can change data volume, model frequency and storage requirements. Revisit the boundary when scale changes and check whether the operational benefit persists. Retiring unused models and unnecessary data retention can be part of good stewardship.

IEA’s figures are a scenario-based macroeconomic context, not a project calculator. This article offers no universal energy payback for AI. The practical requirement is transparent accounting: report the benefit that was measured, the infrastructure that was counted and the uncertainty that remains.

Source notes

[IEA] International Energy Agency. Energy and AI. IEA, Paris, 2025-04-10. Executive summary; Energy demand from AI; AI for energy optimisation and innovation. Accessed 1 September 2026.

https://www.iea.org/reports/energy-and-ai

[HONG] Tianzhen Hong and Han Li. Good practices for documenting AI-based studies on energy and buildings. Energy & Buildings / Elsevier; author copy hosted by Lawrence Berkeley National Laboratory, 2026-01-20. Sections 2, 3.1–3.6 and 4; pp. 1–4. DOI: 10.1016/j.enbuild.2026.117043. Accessed 1 September 2026. https://eta-publications.lbl.gov/sites/default/files/2026-06/1-s2.0-s0378778826001039-main.pdf

Editorial and visual note

This is original Osmos Global analysis informed by the cited publications. Reported findings are distinguished from Osmos recommendations and illustrative scenarios. Source findings and trademarks remain attributable to their owners. Original visual designs do not imply endorsement by source organisations. The content is general research and does not replace site-specific professional advice.

IEA data attribution: IEA (2025), Energy and AI, IEA, Paris. Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Visuals are newly designed by Osmos; any calculation is identified in its caption.