Frugal AI kit: reducing carbon footprint

Free resource in French · Version 2 of October 26, 2026

Frugal AI Kit: The decisions that truly impact your AI project's footprint

What a request, a reasoning method, an agent, or a video actually weighs; what AFNOR, ADEME, and Arcep say; and the ten scoping, purchasing, and management decisions that account for the bulk of the gain, without writing a single line of code. Sourced orders of magnitude, not sensational figures.

📄 Word, 42 pages📊 Excel Workbook, 7 tools🌱 Based on AFNOR Spec 2314🆓 Free
W<1 tMonitoringimpact sheet1 to 10 tReflectionlevers activated> 10 tAction planCO2e per year

The frugal AI kit, in short

What it is
A decision kit for the sponsor of an AI project, who needs to be able to answer "what is the impact of this project on the planet, and what have we done to reduce it?": orders of magnitude, framework, self-diagnosis, necessity test, ten levers, questions and clauses for suppliers, indicators and steering thresholds, roadmap.
For who
Business units that sign up for the deployment of assistants or agents, project managers and AI program managers, IT departments, procurement, CSR and responsible digital initiatives, in groups, banks, insurers, public institutions and industrial companies. No technical skills required.
Format
42-page Word document (8 parts and 6 annexes) and Excel workbook with 7 tools: self-diagnosis, range order of magnitude calculator, decision matrix, supplier evaluation grid, five-indicator dashboard, use case register, before/after arbitration simulator.
What it contains
The 2025-2026 orders of magnitude (sourced query, long context, reasoning, agent, image, video, training, hardware), the five misconceptions and the degree of certainty of each figure; what AFNOR Spec 2314, the Ecolab commitment kit, the State's purchasing form, ADEME, Arcep and the Senate say, and what you are really obligated to do in 2026; the twelve-question self-assessment; the necessity test and the "AI or no AI" decision tree; the ten levers for the sponsor with the documented gain of each, including the case of off-the-shelf assistants, and the simulator that quantifies the before and after; sourcing questions, nine indicators to require and six contractual clauses; five indicators, two thresholds, the frugal AI benchmark, decommissioning and communication without greenwashing; the 90-day and 12-month roadmap; the one-page impact sheet, the ten-rule charter for sobriety of use, the note to the management committee, the thirty-point checklist, the glossary and the dated sources.
What is it based on?
AFNOR Spec 2314 “General Reference Framework for Frugal AI” (June 2024), the Engagement Kit for Frugal AI (Ecolab, Hub France IA, AFNOR, June 2025), the practical guide for responsible purchasing of AI solutions (MiNumEco and State Procurement Directorate, end of 2025), the ADEME opinion of July 22, 2026, the Arcep-PEReN report of May 21, 2026, the Senate report of September 16, 2026, the life cycle analysis of Mistral Large 2 (Carbone 4, ADEME, July 2025), the measures published by Google (including its 2026 environmental report), Microsoft Research, Hugging Face, the IEA, Epoch AI, UNESCO and Sasha Luccioni's teams, the “Omnibus I” directive (EU) 2026/470 on the CSRD, the delegated regulation of 21 September 2026 creating the A to G rating of data centers, and the Impact'IA calculator published by SNCF in 2026.
Price
Free, in exchange for a professional email address. No unsolicited emails.
Update
Version 2, updated October 26, 2026: new before/after arbitrage simulator in the workbook, revised orders of magnitude (agents, long-term context, graphics card manufacturing) with the degree of certainty for each figure, and updated obligations (CSRD after Omnibus I, data center rating A to G adopted on September 21, 2026). A further review is planned upon publication of the first data center sustainability labels (2027) and the next Arcep annual survey.
What it doesn't replace
A life cycle assessment compliant with AFNOR Spec 2314 or ITU-T L.1801, an instrumented measurement of your systems by your technical team, or a legal opinion on your environmental reporting and claims. The kit puts you in a position to decide, ask the right questions, and take control.
Go Further
AI Act kit from the sponsor, sponsor's GDPR kit, AI Systems Governance Kit, AI charter template, AI training in companies.

There are two narratives about AI's footprint, and both are based on real figures. The first says that a query consumes ten times the energy of a Google search and drinks a bottle of water. The second says that a query uses less energy than a second of oven time and that's the end of it. They're simply describing different things: a median query in 2026 versus a query in 2023, on-site cooling water versus total water consumption, a market-based carbon factor after purchasing certificates versus the actual grid mix. Between these two narratives, the sponsor of an AI project can't find the orders of magnitude they need, and, more importantly, what they can actually decide.

This toolkit is based on the French standard definition, AFNOR Spec 2314: the frugality of an AI service depends "on a redefinition of uses or performance requirements." Redefining uses and requirements is a sponsor decision before it's a code optimization. And the data confirms this: the significant differences in footprint (a factor of 59 between a small and a large model, a factor of 30 on average and up to 700 for reasoning mode, a factor of 8 to 10 depending on the hosting country, and approximately half the energy used by halving the length of responses) are determined by the scope, the choice of solution, and the purchase. The toolkit brings together these levers, the official standards on which they are based, and the tools to implement them.

Kit overview: what you receive

Before you leave your email, here's exactly what the kit contains. Everything is in French, each figure has its source and date, and version 2 was updated on October 26, 2026.

"AI or no AI" tree · impact index

Standard text query = 1

Solution without AI (existing rule or tool)≈ 0 Small specialized model: classify, extract0,03 to 0,3 Limited response or framed RAG: write concisely1 to 3 Image, if the text is not enough2 to 10 Agent, for rare tasks3 to 170 Advanced model and reasoning10 to 130 Exceptional video300 to 3,000

Before/After Simulator · Demonstration Scenario

Internal assistant, 2,000 users, 6,6 million requests per year: approximately 10 tonnes of CO2e per year before, approximately 1,3 tonnes afterthat is 87% less for a comparable service provided.

Excerpts from the decision tree and arbitration simulator of the kit (simplified illustration, values ​​taken from the kit).

📄 The Word document, 42 pages

8 sections: orders of magnitude and five misconceptions, framework and timeline 2026-2028, self-assessment, framing, ten levers, procurement, management, 90-day and 12-month roadmap. 6 ready-to-use appendices: one-page impact sheet, ten-rule sobriety charter, note to the management committee, thirty-point checklist, glossary, dated sources.

📊 The Excel workbook, 7 tools

Self-diagnosis, order-of-magnitude calculator (kWh, kgCO2e, litres), decision matrix, supplier grid with fifteen criteria, twelve-month dashboard, use case register, before/after simulator. A references tab provides the source and date of each hypothesis.

🧾 Parts for your suppliers

Sourcing questions for the publisher, integrator and host, nine indicators to require with the correct answer and the red flag, six written clauses and the weighting rule of 10% minimum of the grade.

Receive the complete kit by email →

1. The true weight of AI, in orders of magnitude

The values ​​below come from the providers themselves (Google, OpenAI, Mistral), research laboratories (Hugging Face, Microsoft Research, universities), and agencies (IEA, ADEME, Arcep), all published between 2024 and 2026. They are expressed in watt-hours per request: 1,000 Wh equals 1 kWh, roughly enough to fully charge one hundred smartphones. For each value, the toolkit provides the source, date, and confidence level, and explains why the CO2 and water emissions for the same request vary by a factor of 30 to 170 depending on the scope considered.

Subject Order of magnitudeSources
A classification or extraction using a specialized model0,002 to 0,05 WhLuccioni, Jernite, Strubell (ACM FAccT 2024)
A standard text request to an assistant0,2 to 0,4 WhGoogle 0,24 Wh (August 2025), OpenAI 0,34 Wh, Epoch AI 0,3 Wh, Microsoft Research 0,31 Wh, IEA 0,10 to 0,40 Wh (April 2026)
A long or long-context query, state-of-the-art model1 to 40 WhEpoch AI (2025): approximately 2,5 Wh with 7,500 input words, approximately 40 Wh with 75,000 words; Jegham et al. (2025)
A query in reasoning mode3 to 40 WhAI Energy Score v2 (×30 on average, up to ×700 for the same model), Microsoft Research (×13), Jegham et al. (o3 approximately 12 Wh)
An agent's request1 to 50 Wh per request, more for a long taskIEA, “Key Questions on Energy and AI” (2026), cited by Our World in Data: approximately 1,1 Wh for a standard agent request, approximately 50 Wh with reasoning; KAIST (2026): up to 136 times a chatbot request
A generated image0,5 to 3 WhMIT Technology Review and ML.ENERGY (May 2025), Luccioni 2024, IEA 2026
A five-second generated video100 to 950 WhHugging Face (July 2025), MIT Technology Review, IEA 2026; doubling the duration or quality multiplies the energy by approximately four (Delavande et al. 2025, cited by ADEME)
Training a large model20,400 tCO2e for Mistral Large 2, including 18 months of operationMistral AI, Carbone 4, ADEME (July 2025); manufacturing accounts for 11% of emissions (ADEME, 2026)
Manufacturing a data center graphics card60 to 320 kgCO2e per GPUADEME (July 2026 opinion, eight maps analyzed); NVIDIA, HGX H100 product review
The hosting country46 gCO2e/kWh in France, 210 in the European Union, 384 in the United States, 458 worldwideADEME Base Empreinte 2025, Ember 2026 (life cycle); the IEA gives lower values ​​(170 for the European Union, 435 for the world) in another convention: do not confuse the two
Data centers around the world485 TWh in 2025, approximately 950 TWh expected in 2030IEA, “Key Questions on Energy and AI” (2026); The Shift Project envisions up to 1,500 TWh in 2030 without a change in direction

Three conclusions emerge, and they structure the toolkit. Inference, meaning usage, now dominates the training: your usage decisions carry more weight than the technical choices of a laboratory. Large discrepancies are decisions, not optimizations: none of the levers below require a single line of code. And figures are interpreted within their scope: a frugal sponsor doesn't calculate down to the gram; they know which questions to ask and reject figures without context. The toolkit even classifies its own figures into four levels of certainty: robust (relative discrepancies, carbon factors by country), accurate to within 50% (median text query, PUE), fragile (agents, video, water, manufacturing), and unknown (the footprint per query of most assistants on the market).

2. What the framework says, and what actually obliges you

In two years, France has produced a unique body of work on frugal AI: AFNOR Spec 2314 and its thirty-one best practices (June 2024), the Engagement Kit with fifteen best practices from Ecolab, Hub France IA, and AFNOR (June 2025), the practical guide for responsible procurement of AI solutions from DINUM and the State Procurement Directorate (end of 2025), the ADEME opinion (July 22, 2026), the Arcep report (May 21, 2026), and the Senate report (September 16, 2026). The kit extracts what a sponsor can use directly: the three cumulative conditions for a frugal service, the six-step process, sourcing questions and standard clauses, and the seven algorithmic criteria of the RGESN.

It must also be said without mincing words: No legislation currently imposes a direct environmental obligation on a company that deploys an AI assistant or agent.The obligations fall on model providers (energy documentation for general purpose models, Annex XI of the AI Act), on data centers (European reporting, sustainability rating A to G adopted by the Commission on September 21, 2026, first ratings from August 15, 2027), on cloud providers (collection by Arcep, which a draft decision of July 2026 extends to generative AI services of more than one million users) and on public buyers. They concern you indirectly: because they produce data that you can demand, because the purchased AI falls within your scope 3 (the CSRD, refocused by the "Omnibus I" (EU) 2026/470 directive on companies with more than 1,000 employees and more than €450 million in turnover, must be transposed by March 19, 2027 at the latest), and because your environmental claims have been regulated since September 27, 2026. The client's leverage is therefore contractual and organizational.

DateWhat happensWhat you can do with it
August 2, 2026Commission enforcement powers on energy documentation of general purpose models (AI Act, Annex XI).Request documented energy consumption data for the model in your tenders.
September 27, 2026 Application of Directive 2024/825 on environmental claims.Review all "responsible AI" or "green AI" communications with chapter 4 of AFNOR Spec 2314.
End 2026Arcep's decision to extend its collection to generative AI services; closure on December 14 of the European consultation on minimum performance standards for data centers; follow-up to the Senate report (eco-score of models).Prepare indicators compatible with a future eco-score: gCO2e per unit of work, location-based.
March 2027Transposition of the “Omnibus I” directive (no later than March 19); annual Arcep collection by March 31; first data from cloud providers published in the spring.Check your liability to the CSRD; compare your hosting providers on public data.
2rd quarter 2027Expected proposal from the Commission on minimum energy performance standards for data centres.Ask your hosting providers about their future trajectory: a site below the minimum requirements will have to invest or close.
August 2, 2027Compliance of general purpose models prior to August 2025.All your suppliers then have energy documentation: no more excuses.
August 15, 2027First sustainability ratings A to G for data centers of 500 kW and above (delegated regulation of 21 September 2026, subject to no objection from Parliament and the Council).Require the data center rating in your hosting contracts and renewals.
Exercise 2027Simplified ESRS mandatory for companies within the scope of the CSRD (financial years beginning on or after January 1, 2027).Integrate purchased AI within scope 3 with collected supplier data.
2028Review of the data centre rating system (no later than 31 December 2028); Commission report on the standardisation of energy efficiency of AI systems (AI Act).Anticipate efficiency requirements in future specifications.

3. The ten levers of the sponsor

Each lever is documented by at least one source published between 2024 and 2026 and can be decided without writing a single line of code. The toolkit provides, for each lever, the decision-making process, what the data reveals, the question to ask your team or supplier, and the pitfall to avoid. The first four levers account for the majority of the gains.

LeverDocumented GainSource
1. The smallest model capable of the task, by default; the larger model upon justification.Factor 59 between Llama 3.1 8B and 405B; up to 90% savings with small, specialized modelsMIT TR and ML.ENERGY 2025; UNESCO and UCL 2025; Arcep 2026
2. Reasoning, advanced search, and agents are disabled by default.Factor 30 on average and up to 700 for reasoning; 1 to 50 Wh per agent requestAI Energy Score v2; Microsoft Research 2025; IEA 2026; ADEME 2026
3. Classify rather than generate, text rather than image, image rather than videoFactor of up to 1,450 between classification and image generationLuccioni 2024; IEA 2026
4. Short answers and a controlled contextHalving the length of responses reduces energy consumption by 54%; with long attached documents, it's the input that weighs in (approximately 2,5 Wh for 7,500 words, 40 Wh for 75,000).UNESCO and UCL 2025; Epoch AI 2025
5. Low-carbon hosting, proven on a location-based basis46 gCO2e/kWh in France compared to 458 worldwide: a factor of 8 to 10ADEME, Ember 2026, Arcep 2026
6. Reuse rather than retrain: prompt, framed RAG, light fine-tuningRAG: 20 to 51% less energy with an adjusted perimeterGuo, Gao, Bogner 2026; AFNOR Spec 2314; RGESN
7. Caching, request grouping, and routing"One complex prompt is better than five simple prompts"Razorfish and Green IT 2024; Mistral 2025
8. Require compressed models and an efficiency ratingQuantization: 22 to 44% less energy; up to 73% with a set of optimizationsUNESCO and UCL 2025; Fernandez, Luccioni et al. 2025; Hugging Face
9. Postpone non-urgent treatments to off-peak hours.From 9 to 58 gCO2e/kWh per year in France: factor 6 (secondary lever)RTE 2025; MiNumEco fact sheet
10. Manage, raise awareness, stop: five indicators, two thresholds, one lead, one charterOnly 12% of organizations measure the footprint of their generative AI.Capgemini 2025; Ecolab 2025 Commitment Kit

For each lever, the kit provides the decision, the question to ask, and the pitfall to avoid. Receive the ten detailed levers and the simulator →

4. The case of off-the-shelf assistants

Most large organizations don't build AI systems from scratch; they purchase licenses for Copilot Chat (included in eligible Microsoft 365 subscriptions) or Microsoft 365 Copilot (often called "Copilot Premium"), ChatGPT Enterprise, Gemini, Claude, or Mistral's Le Chat. The prevailing view is that the environmental footprint is the vendor's responsibility. Their contribution is real; yours remains, and the toolkit details it in seven key areas: default settings (template and modes), disabling unnecessary "augmented" features (which ADEME has recommended since July 2026), the number of licenses and their usage rate, the inference location and your tenant's carbon footprint, usage reports as measurement proxies, user awareness training, and the transparency clause in the framework agreement. It lists what a commissioning body can demand from a vendor in 2026: three of these items correspond to what the vendor already provides to regulators.

To move from levers to decision-making, the workbook includes a before/after trade-off simulator: it takes the current allocation of a use case by task type, the target allocation, and three adjustments (avoided requests, shorter responses, hosting), and provides a range for electricity, emissions, and water consumption, before and after, lever by lever. In the kit's demonstration scenario (an internal assistant for 2,000 users, 6,6 million requests per year), emissions decrease from approximately 10 tons of CO2e per year to approximately 1,3 tons in typical value, representing an 87% reduction for comparable service; the new task allocation and hosting within the European Union have the same impact as all the other levers combined.

5. Buying: questions, indicators, clauses

Since no text mandates it, your contract will bind your supplier. The toolkit adapts the sourcing questions and standard clauses from the government's practical guide to private contracts: separate questions for the model publisher, the integrator, and the hosting provider; nine indicators to require, each with the correct answer and a red flag (country and inference data center, PUE and WUE, footprint per unit of work and method, location-based agreement, training energy, efficiency score, model size, bounding options, usage report); six pre-written clauses (transparency, reporting, proportionality and non-complexity, retraining and versions, data lifecycle, skills); and the weighting rule, a minimum of 10% of the overall score, recommended by the French Digital Affairs Directorate (DINUM) and the State Procurement Department. The workbook's grid scores fifteen criteria in five categories and ranks your candidates out of 100.

6. Management: five indicators, two thresholds, one lead

The minimum sustainable metrics include: the number of requests, estimated electricity consumption, estimated location-based emissions, the proportion of requests in high-cost mode, and the monthly cost, with the license utilization rate serving as a contextual indicator. Two thresholds, taken from the governance rule published by SNCF (the French National Railway Company) with its Impact'IA calculator, are also relevant: beyond one tonne of CO2e per year, a documented analysis is required; beyond ten tonnes, a mandatory reduction plan is necessary. A designated frugal AI lead, as requested in the Ecolab Commitment Kit, maintains the use case register and prepares the quarterly review for the steering committee. The kit also addresses what all guides overlook: the criteria for stopping, retraining, and decommissioning, and how to discuss these publicly without violating the Environmental Claims Directive.

7. What the kit contains

The Word document has 42 pages, in eight parts and six appendices: the user manual and the three conditions for frugality; what AI really weighs (orders of magnitude, why the figures diverge, five misconceptions, three conclusions, what is certain and what is not); what the framework says (reference points, actual obligations, 2026-2028 timetable); the self-diagnosis in twelve questions; the framing (necessity test, "AI or no AI" decision tree, sufficient performance level, task family, usage volume and AFNOR formula, rebound effects, impact sheet); the ten levers, the case of off-the-shelf assistants and the quantified trade-off before/after; the purchase (sourcing questions, indicators, six clauses, weighting); the management (indicators, thresholds, reference point, charter, decommissioning, communication); the 90-day and 12-month roadmap. The appendices provide the one-page impact sheet, the ten-rule code of conduct for responsible consumption, the note to the management committee, the thirty-point checklist, the fifteen-term glossary, and dated sources.

The Excel workbook includes the tools in seven tabs, plus a reference tab where each quantified hypothesis has its source and date: the self-assessment that calculates your profile and the phase to prioritize; the calculator that transforms your volumes by task type, your hosting country, and your PUE into an annual range of kWh, kgCO2e, and liters, with the level relative to thresholds and the savings from hosting in France; the decision matrix for each option, with the impact index of each; the supplier evaluation grid with fifteen weighted criteria; the twelve-month dashboard; the use case register with estimated footprint, level, retraining and discontinuation criteria; and the trade-off simulator that compares the before and after of a use case, in ranges and lever by lever. For more detailed calculations including training and manufacturing, the kit refers to the SNCF's Impact'IA calculator.

The kit, part by part: what each part contains and what it's for. The templates to fill out are in the Word document and the Excel workbook.

Part 1. The true weight of AI

Why the figures don't agree with each other

Each source chooses its scope. Google includes on-site cooling water (0,26 mL) but not electricity generation water, and converts energy using a market-based factor of 94 gCO2e/kWh, compared to 345 gCO2e/kWh in 2024 based on location data. Meta reports 0 tCO2e for Llama 3.1 training based on market data, and 11,390 tCO2e based on location data. Hence the rule for reading the toolkit: before any figure, ask about the scope, the carbon convention, the water usage, and the query.

Five misconceptions, and what the data says.

“The problem is training”: according to the IEA, energy consumption “has already decisively shifted from training to inference.” “Each query is negligible”: Google reduced the median energy of a Gemini query by a factor of 33 in twelve months, but reports a 37% increase in electricity demand, a 34% increase in water consumption, and an 18% increase in emissions for 2025 (2026 environmental report): this is the Jevons paradox. “Our supplier is carbon neutral”: the manufacturing of the hardware, which generates 60 to 320 kgCO2e per graphics card according to ADEME, is excluded from almost all per-query figures.

"Every conversation swallows a bottle of water."

The figure of "500 mL per 10 to 50 responses" (Ren et al., 2023) concerned GPT-3 and included water used for electricity generation; Google reports 0,26 mL when only on-site cooling is considered; Mistral finds 45 mL per response in a full analysis. The 170-fold difference stems from the scope. Most of the water is indirect and location-dependent: French data centers directly withdrew 575,000 m³ in 2024, but approximately 6,5 million m³ when including the water needed for their electricity generation (Arcep, May 2026). Demand WUE and location data.

Part 2. What the framework says: standards, obligations, timetable

AFNOR Spec 2314, "General reference framework for frugal AI" (June 2024)

Free of charge and developed at the initiative of Ecolab, it establishes an evaluation method based on life cycle analysis and thirty-one best practice sheets categorized by benefits and effort. Its six-step process serves as a project plan: define the need, identify all solutions, with or without AI, evaluate each solution throughout its life cycle, integrate the evaluation into decision-making, implement tools for monitoring actual usage, and redefine roles. As of October 26, 2026, it remains an AFNOR Spec, a consensus document and not an officially recognized standard, with no standard to replace it: the future lies with CEN-CENELEC, the ITU, and ISO. Cite it as a methodological reference, not as a certification.

RGESN, General Eco-design Framework for Digital Services, 2024 version

Of its 78 criteria, seven relate to algorithms and are presented as such to a provider: need for training (9.1), complexity proportionate to use (9.2), limited training and data (9.3 and 9.4), optimized retraining (9.5), compression (9.6), inference strategy (9.7). Criterion 1.1 requires that the service has "been favorably evaluated in terms of utility, taking into account its environmental impacts."

The standards and metrics that are coming

ITU-T Recommendation L.1801, approved on February 26, 2026, is the first comprehensive international methodology that compares a service with and without AI. SCI for AI (Green Software Foundation, December 17, 2025) measures carbon intensity per functional unit: per token, per image, per page. The AI ​​Energy Score (version 2, December 4, 2025) rates models from one to five stars across ten tasks: the simplest free comparison tool for a buyer.

Part 3. Self-assessment in twelve questions

Lecture du résultat

Twelve questions, each scored 0, 1, or 2, for a total of 24 points: framing (out of 6), solution selection (out of 6), purchase (out of 4), and management (out of 8). These questions assess whether the footprint has become a factor in your decisions. Scores from 0 to 8 indicate a "Discovery" profile: start with the default model and expensive options. Scores from 9 to 16 indicate a "On the Way" approach: focus on the lowest-scored phase. Scores from 17 to 24 indicate a "Frugal" approach: document your decisions so you can discuss them publicly without risk.

Part 4. Framing: Decide before you build

The necessity test

Five questions, in a meeting with the business and the IT department: what problem, for whom, and how will we know it's solved? Does a solution without AI solve 80% of it? What is the simplest form of AI that is sufficient? What level of performance is sufficient? What usage volume and what rebound effects? The acceptance threshold is defined before testing; we first test the smallest plausible model on fifty real-world cases, and we only scale up based on measured failures.

The "AI or no AI" decision tree

Eight questions, each with a default orientation and an impact index (standard text query = 1): solution without AI if an existing rule or tool is sufficient 80% of the time (index close to 0); small specialized model for classifying or extracting (0,03 to 0,3); bounded response or framed RAG for writing short (1 to 3); cutting-edge model and reasoning on justification (10 to 130); agent for rare tasks (3 to 170 depending on autonomy); image if text is not sufficient (2 to 10); exceptional video (300 to 3,000).

Estimating the usage volume: the formula a·x + b + c·y

AFNOR bases its calculations on a year of service: a, the cost of an inference; x, the number of queries; b, the amortized training; c·y, the retraining. For a purchased model, the key factor is a·x. Example from the kit: 500 users with 10 queries per working day, i.e., 1,1 million per year, generate approximately 400 kWh and 220 kgCO2e in standard text; 13,000 kWh and 7 tonnes in continuous reasoning; from 7 to over 35 tonnes with ten agent queries per day per person, depending on the assumption used. Hence the importance of measuring your agents before generalizing them (at the global emission factor of 458 gCO2e/kWh, including PUE and manufacturing).

The one-page impact sheet

Intended for the committee that decides on the launch, it brings together the need, the AI-free alternative studied, the solution chosen, the estimated volume, the annual order of magnitude in range, the level in relation to the thresholds, the levers activated, the supplier's responses, the indicators and the stopping criterion, as well as three probable rebound effects (ten versions of a text, one summary per meeting) with their control measure.

Part 5. Choosing the solution: the ten levers of the sponsor

Lever 5: low-carbon, location-based hosting

The inference process runs in France or another European country with low-carbon electricity, with supporting evidence available on a location-based basis; market-based methods are not accepted. In the workbook simulator, switching from the "unknown country" assumption to the European Union reduces emissions by more than half, and switching to France by nearly tenfold, for the same usage. The pitfall: confusing data location with computing location. Ask for the location of the inference, not the storage location.

Lever 6: Reuse rather than retrain

In order: a well-constructed prompt, a well-defined RAG, and light fine-tuning; retraining remains the exception, triggered by measured degradation rather than a schedule. Adjusting the similarity threshold of a RAG reduces energy by 20 to 51% without loss of accuracy (Guo, Gao, and Bogner, 2026). The pitfall: the "all-SharePoint" RAG, which is more expensive for each question and less relevant.

What a sponsor can expect from an assistant publisher in 2026

The inference location of its tenant and the associated location-based factor; the PUE and WUE of the data centers; the median and reasoning mode energy per request, including the perimeter; the documented energy consumption of the model (Annex XI of the AI Act); an efficiency rating; the group deactivation of expensive modes; and a monthly usage report. Three of these pieces of information are already produced for regulators.

Part 6. Purchase and contract: demanding, weighing, writing

Sourcing issues, by supplier type

To be submitted in writing, with answers attached to the contract. To the software publisher: What is the smallest model in your catalog capable of handling our use cases? To the systems integrator: What AI-free or lighter alternative have you considered, and how do you limit token usage? To the hosting provider: What are the PUE, WUE, and CUE values, and a footprint report per tenant on a location-based basis, including server manufacturing costs?

The indicators to require, and the correct answer

For each indicator, the toolkit provides the correct answer and a red flag. Location of inference: a named country and site, not "our data is in Europe." PUE: less than 1,3, with WUE. Footprint per unit of work: a median value and a reasoning-mode value, with scope, method, and date. Carbon convention: market-based alone is a red flag. For training energy, the answer "confidential" after August 2, 2027, is another.

Six clauses for a private contract

Environmental transparency, quarterly reporting, proportionality and simplicity, retraining and versions (keeping a previous version for twelve months), data lifecycle, skills. The rubric scores fifteen criteria in five categories (transparency 30%, efficiency 25%, hosting 20%, contract 15%, governance 10%), and the environment accounts for at least 10% of the overall score, as recommended by the MiNumEco guide.

Part 7. Steering and governance: five indicators, two thresholds, one lead

The five indicators

Each case has its own trigger: a 30% increase in requests in a single month warrants an explanation; estimated emissions are compared to the thresholds of one and ten tons per year; when more than 10% of requests are processed in high-cost mode, settings are reviewed; and when license utilization falls below 40%, licenses are reassigned, training is provided, or the system is shut down. The kgCO2e per case processed is the maturity indicator.

The charter for sobriety in use

Ten rules, each justified by an order of magnitude: ask yourself if AI is necessary, ask a complete question the first time, ask for the short version, reserve reasoning for problems that deserve it, reserve agents and in-depth research for rare and long tasks, no image when text is sufficient, do not regenerate "just to see", keep the default model, reuse validated prompts, report unnecessary uses to the referent.

Stop, retrain, decommission

Three rules were written from the outset. Discontinuation: utility rate below 20% for two quarters, business objective not achieved, or redundancy with a standard function. Retraining: only upon measured degradation below the acceptance threshold, never by schedule. Decommissioning: data deleted or archived, licenses released, indexes destroyed, and a line item recorded in the registry.

Part 8. 90-day and 12-month roadmap

The first 90 days

Days 1 to 30, gain clarity: appoint the lead user, inventory use cases and licenses, complete the self-assessment, and adopt the charter. Days 31 to 60, quantify and resolve: run the five largest cases through the calculator, simulate the largest case before and after using the arbitration simulator and draft its impact assessment, configure default settings, and consult with the three main suppliers. Days 61 to 90, manage: dashboard, initial committee review, and transparency and reporting clauses for the first renewal.

Flavien Besset, founder of NextStart.AI

Who designed this kit?

Flavien Besset, founder of NextStart.AI

This toolkit compares French standards (AFNOR Spec 2314, Ecolab Commitment Kit, State Responsible Purchasing Form), the positions of ADEME, Arcep, and the Senate, and the measures published by suppliers and laboratories in 2025-2026 with the decisions made by sponsors in the AI ​​adoption programs we support in large organizations. With the NextStart.AI team: more than 5,350 professionals trained in some thirty organizations (Airbus, CNES, CEA, Engie, Bristol Myers Squibb, etc.), average rating of 9,2/10.

Download the frugal AI kit (Word + Excel, free)

42 pages, 8 parts and 6 annexes, plus a workbook of 7 tools. Written by NextStart.AI based on AFNOR Spec 2314, the Ecolab Commitment Kit, the State's responsible purchasing sheet, the positions of ADEME, Arcep and the Senate, measures published by suppliers and laboratories in 2025-2026, and the AI ​​adoption programs that we support in large organizations.

  • Sourced orders of magnitude and five misconceptions to forget
  • The necessity test, the "AI or no AI" decision tree, the ten levers with their gains, and the before/after simulator
  • The questions, indicators, and six clauses for your suppliers, including assistant publishers
  • The five indicators, the two thresholds, the charter for responsible consumption, the report to the management committee, the Excel spreadsheet

    Associated supportChoose your use cases →The use cases of each department, quantified and arbitrated, then the 12-month roadmap with objectives and budget.3 weeks12-month roadmapConsulting mission

    8. Frequently Asked Questions

    Does a ChatGPT query really consume ten times the resources of a Google search?

    This figure is from 2023, based on an assumption of 2,000 tokens per response. Measurements and estimates for 2025-2026 converge on 0,2 to 0,4 Wh for a median text query (Google, OpenAI, Epoch AI, Microsoft Research, IEA). But the median masks the tail of the distribution: a reasoning mode, a long context, or an agent costs ten to a thousand times more. The real question is no longer "how much per query" but "which model, which mode, how many tokens?"

    Are we obligated to measure the footprint of our AI?

    No, not directly. The AI Act mandates energy documentation for general-purpose model providers, not for the companies that use them; the REEN law and Arcep rulings target local authorities, data centers, and cloud providers. Indirectly, purchased AI falls under Scope 3 of your reporting, and your environmental claims are subject to regulations. The toolkit explains what you are and aren't obligated to do, and what you can require from your suppliers based on their own obligations.

    Our tools are Copilot or ChatGPT Enterprise: what can we do?

    Many things to consider: choosing the default model and modes, disabling unnecessary "enhanced" features as recommended by ADEME, measuring license usage, obtaining the locus of inference and carbon footprint of your tenant, using usage reports as a metric, training users, and introducing a transparency clause for renewals. The toolkit dedicates a section to this and lists what a sponsor can expect from a software vendor in 2026.

    Does the kit calculate our exact footprint?

    No, and no one can do it today: 84% of the 754 models identified by Luccioni et al. have no environmental information available (study cited by Arcep in May 2026). The calculator provides a low-to-high range based on published work units, your volumes, your hosting country, and your PUE, accurate enough to classify your use cases against thresholds and make informed decisions; the trade-off simulator compares the before and after of a use case, lever by lever. For a life cycle assessment, it refers users to the AFNOR method and to a consulting firm.

    Why are there thresholds of one and ten tonnes?

    Because they are simple, public, and proven: these are the ones that SNCF applies to its digital projects and which it made public with its Impact'IA calculator in 2026. One ton of CO2e is roughly equivalent to a round trip from Paris to New York per passenger; ten tons, to slightly more than the annual carbon footprint of a French resident. You can adapt them to your own needs, but keep two: one that requires reflection, one that requires action.

    What to start with ?

    Through the two most cost-effective and quickest decisions: the lightest default model capable of handling the task, and the expensive modes (reasoning, agents, image, video) disabled by default. Then, through the twelve-question self-assessment, which tells you which phase to work on next. The 90-day roadmap in the kit is written for an organization starting from scratch.

    9. Going further

    This kit is part of the AI ​​project sponsor's kit series: the AI Act kit covers what applies to a deployer and the AI ​​literacy program, the GDPR kit covers the qualification of projects and the file that the DPO expects, the AI and Social Dialogue Kit prepares the consultation of the CSE, the Security and data kit addresses the project's security risks, and the kits Shadow IA, AI Sovereignty et Responsible AI complete the series. The AI Systems Governance Kit provides the register and qualification grid, the AI charter template sets the rules of use, within which the responsible use charter of this kit fits. Organizations that want to quickly pass the necessity test on their use cases and configure their assistants to be more energy-efficient do so within the framework of the AI training in companies and the mission Governing your agents.

    Read also : AI Project Sponsor Security and Data Kit, Sponsor's AI Act Kit: The 2026-2027 Roadmap, GDPR Kit for AI Project Sponsors, AI systems governance: the 2026 toolkit et AI Charter in Business: The 2026 Model (Downloadable).

    The entire series of free kits. To drive adoption: AI charter template, AI training plan template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, AI-powered time-saving kit, Soft Skills Kit for AI, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Cowork kit, 50 Copilot M365 prompts, Copilot Studio kit, Deploy Microsoft 365 Copilot kit, Copilot Chat Kit, Copilot Excel kit, Copilot Teams and Outlook kit, Copilot Word and PowerPoint kit, Configure Claude kit, Deploying Claude in the Workplace Kit, Claude Agents Kit, Claude Skills kit, Claude's Marketing Skills, Skills Claude HR, Claude Plugins Kit, Claude Code kit, Claude Design kit et Vibe coding kitFor the AI ​​project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, AI and Social Dialogue Kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit, communicator's AI kit, AI Procurement Kit, AI Kit for Researchers et AI kit Customer service.

    10. Sources

    The figures and rules cited on this page and in the kit are based on the following dated sources. All are free. The kit lists approximately forty more in the appendix.

    Back to top
    All NextStart.AI offersEight consulting missions, seven pillars, around twenty formats, eighteen professions, five tools
    Around thirty major organizations were supported.Large groups, public agencies and networks: from acculturation to production

    Insurance and social protection

    Public sector and research

    Consumption and retail

    The NextStart.AI collectiveConsultants, trainers and industry experts: more than 5 350 trained professionals since 2023
    All free resourcesDiagnostics, Finders, prompt library, kits by profession, by tool, and for managing an AI program
    The NextStart.AI blogComparisons of tools, practices, agents, governance: analyses and feedback from the field
    Let's talk about your AI projectResponse within 24 business hours
    AI agents in businessUnderstanding, choosing the tool, training the teams, deploying
    Delegate entire tasksClaude Cowork, Copilot Cowork, ChatGPT Work, Vibe Work