Stratégie IA · RSE

The Carbon Footprint of AI Is Becoming a Selection Criterion

By Anis Hammouche·July 6, 2026·9 min read

Until now, when you thought about artificial intelligence, you probably thought about license costs, setup time, and training your teams. A new parameter is joining the list, and it does not come from your IT department. It comes from your stakeholders, your ESG function, and soon your key accounts: what is the environmental footprint of the AI you deploy?

The question is no longer theoretical. The environmental reports published in 2026 by the major cloud providers show a marked rise in their emissions, and the cause is clear: the AI race, which is heavy on energy and water. For an executive, this has two very concrete consequences, and neither forces you to give up AI. The first is rising ESG pressure. The second is a useful reminder: a targeted AI use case consumes less than a broad, vague rollout, which also makes it a better decision for your budget.

What the 2026 figures say

The major cloud providers publish an environmental report every year. The 2026 edition confirms a clear trend. Over one year, Google's total emissions rose by roughly 25 %, and Amazon's by roughly 16 %. Over a longer horizon, the gap is even more telling: about 82 % growth for Google and about 58 % for Amazon since 2019. Google's annual electricity consumption itself grew by roughly 37 % in 2025.

These figures share one cause, acknowledged by the providers themselves: AI. Training and running models takes a great deal of electricity, and cooling data centres takes a great deal of water. In other words, the adoption wave that is sold to you as obvious carries a physical cost, and that cost shows up in the balance sheets of the largest technology companies in the world.

For you, as an executive, the point is not to comment on these figures. It is to understand what they signal: the footprint of AI is moving out of the expert debate and into decision criteria. A supplier, a client, or an investor can now ask you not whether you use AI, but how, and at what environmental cost.

ESG pressure will no longer spare you

For a few years, AI was presented as pure efficiency, with no visible trade-off. That period is ending. Your stakeholders are starting to connect digital usage with carbon footprint, and AI is becoming a point of attention within ESG programmes.

In practice, this means the questions will change. A client auditing your environmental commitments will no longer settle for your recycling policy. They may ask about the AI use cases you have deployed, what they consume, and why you selected them. If your only answer is "everyone is doing it", you will struggle. If you can show that each selected use case answers a measured need, on a defined scope, you are in a strong position.

The good news is that the answer to ESG pressure and the answer to the budget question are the same. An AI use case you can justify with a quantified gain is also one you can justify on environmental grounds, because it is bounded and because it replaces one spend of human or material energy with another, measured one. The vague rollout, by contrast, holds up neither in front of your finance director nor in front of your ESG lead.

Restraint is also a sound financial decision

There is a misconception to clear up: restraint does not mean giving up AI. It means choosing. A broad, poorly targeted rollout consumes across the board, in licenses, in compute, in team time, and often with no clear return. A precise use case consumes at a single point, with a gain you can observe.

Environmental footprint and budget follow the same logic. The broader and vaguer a use case, the more resources it draws for an uncertain result. The tighter and more measured a use case, the more favourable the ratio between what it costs and what it returns, financially and in energy terms alike. Choosing restraint is therefore not an environmental sacrifice. It is the same discipline that protects your budget.

The table below puts the two approaches side by side.

Criterion	Broad, vague AI use	Targeted AI lever
Cost	Licenses and compute across the whole scope, uncertain return	Spend concentrated on one use with an observed gain
Footprint	Broad consumption, hard to tie to a benefit	Consumption limited to the strict minimum
Measurement	No quantified basis, "we must do AI"	Gain in euros or hours, verifiable before and after
Scope	"Transform the company with AI"	One function, one task, one expected result
ESG justification	Impossible to defend under an audit	Each use tied to a real need

The right column describes a use case you can defend both to your leadership team and to a stakeholder attentive to your footprint. The left column describes what the trend pushes you to fund, and what will expose you to both criticisms at once.

What the Scan phase rules out before committing a single resource

Choosing a precise lever rather than a sprawling rollout is not a slogan. It is a sorting exercise, and it is the role of the Scan phase in the S3 method. Scan writes no code. Scan measures and rules out.

In practice, we take your candidate use cases and answer three questions for each. What is the real gain: how many hours or euros, on which function, verified against your day to day. How feasible is it: does the data exist, does the tool hold up in production. What is the scope: is the task bounded enough to ship in a few weeks, or is it an endless transformation project. A use case that fails these questions is a decorative one, and a decorative use case consumes without returning anything.

This is where the environmental logic and the financial logic meet within the method. By ruling out decorative use cases before the Solve phase, Scan spares you the budget and the energy you would have spent on what would have produced nothing. You keep only the levers that pass the test, on a tight scope. The result is a written ranking: at the top, the use cases that create measurable value, ready for Solve. At the bottom, those the trend would have funded and that you drop without regret, for your budget as much as for your footprint.

AI in a company does not require automating everything. It requires choosing few, and choosing well. That choice protects you from the rising ESG criticism, and it serves your budget in the same move.

Frequently asked questions

Is AI's footprint really my problem as an executive? It is becoming one. The 2026 environmental reports from major cloud providers show a marked rise in emissions, attributed to AI. Your stakeholders now make the connection, and an ESG audit can cover your AI use cases. This is no longer a topic confined to cloud providers, it is a criterion that reaches all the way up to your decisions.

Do I have to give up AI to cut my footprint? No. Giving up is not the question, choosing is. A broad, vague rollout consumes with no clear return. A precise lever consumes little and delivers a measured gain. Cutting the footprint means ruling out decorative use cases, not useful ones. You keep the benefit of AI while removing the waste.

How does restraint serve my budget? A broad, poorly targeted use draws licenses, compute, and team time for an uncertain result. A tight use concentrates the spend on one point with an observed gain. The ratio between cost and return is better, financially and in energy terms alike. The same discipline serves both.

How do I connect this criterion to a concrete decision? Through the Scan phase. It takes your candidate use cases, measures the real gain of each, and rules out the decorative ones before committing a single resource. You come out with a quantified ranking that tells you which levers to keep. The decision no longer rests on the trend, but on a measurement you can defend to your finance director and to your ESG lead alike.