Artificial intelligence has arrived in the workplace faster than most organisations anticipated. From writing assistance and data analysis to customer service, procurement tools and supply chain optimisation, AI tools are becoming a routine part of how businesses operate. Sustainability teams are no exception — many are now using AI to support carbon accounting, reporting and strategy development.
There is a certain irony in that. Because AI itself carries a significant environmental footprint, and most organisations using it have given little thought to what that means for their sustainability commitments.
Every time a user interacts with a large language model — typing a prompt, generating a report, running an analysis — that request is processed in a data centre. Data centres consume significant amounts of electricity, require continuous cooling, and in many parts of the world are still powered substantially by fossil fuels.
The scale of this energy demand is not trivial. Ireland's data centres consumed approximately 22% of national electricity demand in 2025, according to SEAI data — a figure that reflects the concentration of large-scale data centre infrastructure the country has accumulated over the past two decades. Globally, the energy demand from AI workloads is growing faster than the broader data centre sector. Training large AI models is an extremely energy-intensive process, and inference — the energy used every time someone uses an AI tool — accumulates across millions of daily interactions.
The carbon intensity of that energy consumption depends on where the data centre is located and how it is powered. Data centres running on renewable electricity have a significantly lower carbon footprint than those relying on grid electricity from fossil fuels. Many major AI providers have made renewable energy commitments, but the reality of how those commitments are accounted for — particularly through energy attribute certificates rather than direct use — is more complex than headline claims suggest.
Electricity gets most of the attention in discussions about AI's environmental impact, but water is also a significant resource. Many d Data centres use water for cooling — either directly in cooling towers or as part of evaporative cooling systems. A large data centre can consume millions of litres of water per year.
As AI workloads increase, so does the cooling demand. This is particularly relevant in regions experiencing water stress — and while Ireland is not typically thought of as water-scarce, it is not immune to the pressures that large-scale industrial water use creates on local supply infrastructure.
Water use is increasingly included in Scope 3 emissions and resource efficiency frameworks. Organisations with serious sustainability commitments are beginning to include their data usage and AI consumption in their broader resource accounting, rather than treating it as an invisible overhead.
The carbon footprint of using AI tools sits within an organisation's Scope 3 emissions — specifically under the purchased goods and services category of the GHG Protocol. Most organisations are not yet measuring or reporting this, partly because the data is difficult to obtain and partly because awareness of the issue is still limited.
This is likely to change. As Scope 3 reporting becomes more rigorous under the likes of the CSRD and similar frameworks, the emissions associated with digital tools and services — including AI — will come under greater scrutiny. Organisations that have already started tracking their AI-related emissions will be better prepared than those encountering the issue for the first time in a reporting context.
The answer is not that organisations should stop using AI. The productivity benefits are real, and sustainability work itself is among the domains where AI is proving genuinely useful — helping to process large datasets, automate routine reporting tasks and identify patterns in emissions data.
The answer is that AI use should be considered as part of a sustainability strategy rather than excluded from it. This means a few practical things.
First, understanding which AI tools your organisation uses and, where possible, seeking information from providers about the energy source and carbon intensity of their data centres. Some providers publish this information; others don't. The transparency of this data will improve over time as reporting standards develop.
Second, considering whether AI tools are being used efficiently. Unnecessary prompts, over-generated content and casual use of computationally intensive tools all add to the energy bill without corresponding value. This is analogous to how organisations think about energy efficiency in their physical operations.
Third, including digital consumption — including AI — in broader discussions about Scope 3 emissions. The emissions don't become less real by being in a category that's harder to measure.
There is also a governance dimension to AI that connects to ESG more broadly. The E in ESG covers environmental impact, which includes AI's resource use. The G covers governance, which increasingly includes how organisations make decisions about which AI tools to use, what data they process, and what accountability structures exist around AI-driven decisions.
Organisations developing ESG strategies are increasingly being asked how they govern their use of AI. This is not yet a standard requirement, but it is a direction of travel that responsible organisations are getting ahead of now, rather than waiting to be asked.
Does AI increase carbon emissions?
Yes. AI tools, particularly large language models, require significant computing power in data centres, which consume electricity and generate carbon emissions. The scale of this impact depends on the volume of AI use and the energy source powering the data centres involved. Training large AI models is particularly energy-intensive; the ongoing inference cost of daily AI use also accumulates at scale.
How much energy does artificial intelligence use?
The energy consumption of AI varies significantly depending on the type and scale of the model being run. Training a large AI model can consume hundreds of thousands of kilowatt-hours of electricity. Inference, the energy used when someone sends a prompt or runs an analysis, is less intense per interaction but adds up across millions of daily uses globally. Ireland's data centres, which host much of the AI infrastructure serving European users, consumed around 22% of national electricity demand in 2025.
What is the environmental impact of AI on a business?
For most organisations, AI-related emissions fall within Scope 3, specifically under purchased goods and services. Currently, very few businesses are measuring this element of their footprint, but as reporting frameworks develop and Scope 3 scrutiny increases under CSRD and similar directives, the environmental impact of digital tools, including AI, is likely to receive more attention. Water consumption by data centre cooling is a secondary impact worth understanding.
How can businesses reduce the environmental impact of AI tools?
Practical steps include choosing AI providers with credible renewable energy commitments, using AI tools efficiently rather than generating unnecessary outputs, tracking AI-related digital consumption as part of Scope 3 reporting, and incorporating AI governance into broader ESG frameworks. Monitoring how this area develops in terms of reporting standards and provider transparency is also worthwhile.
Should organisations include AI use in their carbon footprint?
Increasingly, yes. Scope 3 reporting under frameworks like CSRD will encompass emissions from purchased digital services, including AI. Organisations that start tracking their AI-related emissions now — even approximately — will be better prepared for more rigorous requirements in the future and will have a more complete picture of their total environmental footprint.
If your organisation wants to understand how AI use fits into your sustainability and carbon reporting strategy, AD Sustainability can help you think through the Scope 3 implications and what a practical approach looks like. Get in touch today.