IT Services in Greenhouse Gas Accounting: How to Integrate AI into a Responsible Digital Strategy
Incorporatingdigital services into the carbon footprint®has become a major challenge for companies in France. Digital technology is now a driver of innovation, but the rapid rise of artificial intelligence—and generative AI in particular—is changing the scale of its impacts: more computing and processing power, more technical resources, more electricity consumed…and also greater pressure on water and natural resources.
In other words: if an organization wants to seriously manage its greenhouse gas (GHG) footprint, it can no longer settle for a generic IT audit. It is essential to understand where the footprint is generated (training, usage, data centers, hardware purchases) and to use decision-making tools toprevent AI from becoming a “gray area”in the carbon footprint.
That’s exactly whereDecarbo’Solution® comes in: offering solutions to make what is often invisible in the digital world measurable and manageable.
Contents
Understanding the Carbon Footprint of AI in Organizational Transformation
Carbon Accounting for AI: What Are We Really Talking About?
When analyzing the overall environmental impact of new technologies, many peoplelimit their perspective to the electrical energy required during the use phase. However, serious research requires a life-cycle approach to assess the resources used and data flows:
- Upstream (procurement/manufacturing): servers, GPUs, storage, networks → "embedded" emissions that are often overlooked.
- Training: a process that requires an enormous amount of computing power and electrical energy.
- Inference (use): request processing, IT integrations, and continuous learning workflows.
- Facilities: data centers, cooling, pressure drops, and the strain on water resources.
- End of Life: Equipment Replacement Cycle Driven by the Pursuit of Performance.
For a company, the challenge is tolink this reality to its decarbonization strategy: identifying the carbon emissions generated by AI and understanding how this affects the organization’s transition.
Management with Decarbo’Solution®:
- Decarbo’Supply® helpsmap and decarbonize purchases (servers, GPUs, cloud services in France or abroad).
- Decarbo’Target® incorporatesthis new emissions source into the overallclimate trajectory for real-time monitoring.
Why "giant" models cause orders of magnitude to skyrocket
Large language models (LLMs) rely on massive architectures comprising tens of billions of parameters. Here, the challenge is no longer just aboutoccasional usage, but rather the constant demand for computing power and server infrastructure.
Real-world example:LLaMA (Meta), a powerhouse in the field of artificial intelligence. TheLLaMA technical reportnotes that training the 65B model (a large-scale system) required 2,048 A100 GPUs over 21 consecutive days.
- An A100 processor draws about 400 W under load.
- Direct consumption: approximately 413 MWh for the calculation operation.
- When data center efficiency (PUE) is factored in, the total exceeds 450 MWh.
These figures clearly illustratethe scale of the challenge:AI isn’t just software—it’s a major industry.
Data Centers: Beyond Carbon, the Challenge of Water
We tend to focus oncarbon emissions, but artificial intelligence also poses a water management problem. Research conducted byUC Riversideshows that training GPT-3 (via Google or OpenAI) for two weeks can cause up to 700,000 liters of water to evaporatefor cooling purposes.
The report published in*Communications of the ACM*in 2025also points out that this “invisible” footprint must be factored into any responsible digital assessment.
What are the main regulations, standards, and methodologies for GHG accounting?
Identifying AI Challenges: Life Cycle Analysis
Training: The Most Critical Phase for Footprint Formation
Initial training gets the most media attention because that’s when carbon emissions skyrocket. This process requiresmassive amounts of processing timeand aconstant supply of electricity to power the models. In France, the challenge for companies is to decide between developing a proprietary tool and using off-the-shelf solutions, which are often more optimized.
Choosing a large-scale model directly impacts the natural resources used.Decarbo’Supply®steps in here to establish sustainability criteria right from the supplier selection stage, thereby preventing unnecessary carbon emissions from the very start of the project.
Day-to-Day Operations: The Invisible Flow
Once AI is deployed, it becomes a recurring service. Each request requires energy to process data and transmit information. When you add up the millions of daily interactions on platforms like Google, the total energy consumed far exceeds initial estimates.
It is this digital “background noise” thatrisks derailing the company’s transition to a low-carbon future. By integrating these uses intoDecarbo’Target®, we transform a vague technology expense into a precise control center aligned with the company’s environmental strategy.
Systems and Transparency: The Need for Clarity
Today, the main obstacle to a serious carbon strategy is the lack of transparency surrounding data centers. Without knowing where the data is stored or what the electricity mix is, the carbon footprint remains an estimate.
Groups such as theShift Projectemphasizethisneed for transparency: to ensure that impacts are comparable, companies of all sizes must require their service providers to provide actual dataon the energy consumption of their hosted models.
First, the use of GCI has led to a significant reduction in the uncertainty of the GHG inventory—particularly for Scope 3—from 50% to 5%. The group therefore has data that is 95% accurate.
Using supplier data collected and analyzed by the GCI tool, in-depth analyses were conducted internally on the impact of 35,000 products, grouping suppliers into 170 categories. This represents more than 80% of purchases by volume and value.
Taking Action on AI: Actionable Leverage Points and Strategies
Putting Artificial Intelligence Under Control
The primary means of reducing the footprint lies in governance. Without a clear framework, these new processes spread unchecked. A mature organization in Francemust identify its use cases and classify them by business value in order to establish strict access rules.
Decarbo’Target®serves as the technological backbone here, linking each machine development project to management’s environmental goals and making it possible to track the impacts just as one would track a financial budget.
Opt for simple technical solutions
Simplicity does not hinder innovation; rather, it involves aligning resources with actual needs. Instead of systematically relying on models with billions of parameters, we should prioritize more compact and optimized systems.
Limiting unnecessary training iterations and promoting the sharing of infrastructure can drastically reduce energy consumption.With Decarbo’Supply®, companies can secure a sustainable footprint for years to come.
Merging Innovation with Climate Strategy
The ultimate goal is to incorporate the impact of these technologies into the core of emissions accounting. It is no longer a matter of treating the technological aspect as a separate issue, but rather of being able to precisely quantify what portion of the burden stems from the increased use of these applications.
This approach makes it possible to simulate future scenarios: What is the trade-off between the performance of generative AI and energy efficiency? This is how we transform an innovation project into a driver of sustainable growth.
Artificial intelligence—and more specifically, its generative form —offers significant opportunities but also entails a massive increase in environmental impact. Deploying thousands of processors for large-scale systems like those ofMetaor Google’s services places a strain on natural resources.
To prevent these technologies from becoming a blind spot in companies’ trajectories, rigorous oversight is essential. This is the very purpose ofDecarbo’Solution®’s support: to make AI an ally in the transition. This vigilance must be accompanied by an awareness of other sources ofpollution—such asdata centersandemail—to ensure a truly comprehensive strategy.