Understanding AI and its Environmental Implications
Understanding AI and its Environmental Implications
How organisations can approach more responsible use
May 2026
Artificial Intelligence (AI) tools are now mostly a part of everyday business life, helping teams to work faster, communicate better, and even to make smarter decisions. As adoption grows rapidly across sectors, so does an important question: what is the environmental cost of this technology and how does it fit with a responsible business agenda?
The good news is that understanding AI's environmental footprint is becoming clearer and for businesses committed to ESG, getting to grips with it now is a valuable step.
WHY AI'S ENVIRONMENTAL IMPACT IS A GROWING ESG CONSIDERATION
AI use is becoming a material factor in environmental-focused reporting and for good reason. The infrastructure that powers AI requires significant energy to run, uses water in the cooling process, and relies on hardware that has its own carbon cost across its lifecycle.
It helps to understand two distinct phases of AI's energy use:
Training: when an AI model is built from scratch, this is the most energy-intensive phase. It is carried out once by the AI provider, not by the businesses using the tool.
Inference: when a business or individual then uses this AI tool, for example to send a query, generate a document or analyse data, this ongoing phase is where that organisation's direct environmental impact sits.
However you look at it, the scale of energy consumption in AI creation and use is significant. According to the International Energy Agency, data centres powering AI and cloud computing consumed around 415 terawatt-hours (TWh) of electricity globally in 2024. This equates roughly to the annual electricity use of around 40 million homes, or the entire electricity demand of countries such as Norway or Sweden. The IEA projects that this could more than double to around 945 TWh by 2030, similar to the total annual electricity consumption of Japan (IEA, 2025).
Water use is a growing secondary concern. Data centres require cooling to operate safely, and many rely on evaporative systems that use significant volumes of water. This is increasingly relevant in regions already experiencing water stress, where growing competition for water resources may create operational and environmental challenges for data centre infrastructure (Net Zero Insights, 2025).
Hardware manufacture and disposal also carry an embedded environmental cost that is easy to overlook, although this sits primarily with device manufacturers and AI providers rather than with businesses using the tools.
HOW DOES THIS SHOW UP IN AN ORGANISATION’S CARBON FOOTPRINT?
For most organisations using cloud-based AI services through tools like Microsoft Copilot, ChatGPT, Google Gemini, or similar, the majority of AI-related emissions are likely to sit under Scope 3, as the underlying computing infrastructure is operated externally by AI and cloud providers. However, electricity used by employee devices, office networks, and workplace energy consumption associated with accessing these tools may still contribute to Scope 2 emissions through purchased electricity use.
Measuring AI-related emissions precisely remains a challenge. Most AI providers do not yet publish granular energy or water consumption data at the level of individual queries or products, which makes accurate calculation difficult. However, this is changing in very encouraging ways. The IEA published a landmark Energy and AI report in April 2025, followed by an updated analysis in April 2026, that provides businesses with a clearer evidential basis for understanding their AI-related environmental impact. This recent analysis highlights two important trends: AI-related electricity demand is rising rapidly, but improvements in hardware efficiency mean that the energy required per individual task is gradually decreasing (IEA, 2025; IEA, 2026). The challenge is that overall usage is still growing faster than efficiency gains, meaning the total environmental footprint of AI continues to rise even as individual queries become less energy intensive.
Engaging positively with this now, even with estimated figures, reflects good ESG practice and positions organisations well as reporting expectations continue to evolve.
PRACTICAL STEPS FOR ORGANISATIONS USING AI
We don't need to wait for perfect data to begin taking a responsible approach to AI's environmental impact. Here are some practical steps organisations can take at this stage:
Take stock of which AI tools your organisation uses: you can’t manage what you don’t measure. A simple audit of AI tools might include identifying which are actively used across the organisation, how many employees use them, what business functions they support, whether paid enterprise subscriptions are in place, and whether any internal guidance exists around their use.
Review the environmental commitments of AI providers: Major providers such as Microsoft, Google, and Anthropic increasingly publish sustainability information relating to renewable energy use, carbon commitments, and data centre efficiency. However, reporting standards and levels of detail still vary significantly between providers.
Consider how AI tool use could fit within your broader Scope 2 and 3 reporting, where appropriate: even high-level estimates, clearly caveated, demonstrate awareness and support future reporting as provider transparency and emissions methodologies continue to evolve. While universal ESG criteria are not yet standardised, organisations are developing frameworks to assess their AI use; environmentally, they focus on energy consumption, carbon and water footprints, and whether efficient or renewable-powered systems are being used.
While precise emissions measurement remains difficult, and tackling this full scale can still seem quite overwhelming, organisations can begin by tracking a few practical indicators such as: which AI tools are used across the business, the approximate number of users, whether they use enterprise or paid AI subscriptions, the frequency or scale of usage, the key business functions supported by AI, and whether providers publish sustainability or renewable energy commitments.
Practise digital mindfulness: just as sustainable homeworking guidance encourages employees to avoid unnecessary video calls or duplicate files, the same principle applies to AI. Using smaller, tailored AI models rather than relying on large models can massively reduce the energy consumption associated with the task (UNESCO, 2025). Importantly, avoid generating content or running queries when the output isn't needed. Estimates vary considerably depending on the model, infrastructure, and type of task involved, but some studies comparing the energy use of an AI query to a traditional internet search suggest that AI-generated responses may consume several times more energy. While a conventional search engine primarily retrieves and ranks existing information, generative AI models create new responses in real time. More complex activities such as AI image or video generation are often considered significantly more resource-intensive than text-based interactions.
Consider environmental credentials in procurement decisions: where you have a genuine choice between comparable tools, environmental performance that includes a provider's renewable energy use and water efficiency commitments, is a reasonable factor to include.
Document your approach: as ESG reporting frameworks develop to encompass AI use, having a clear record of how your organisation has considered and managed this issue will demonstrate accountability and good governance. It will set you up well for incorporating more detailed reporting data in the future too.
Implement an AI policy: an effective workplace policy will help ensure employees understand your expectations around AI usage as well as putting guardrails in place to avoid overuse.
THE BIGGER OPPORTUNITY
It’s worth stepping back to recognise that AI also has significant potential to support environmental goals and that it’s not just adding to the problem. According to Deloitte's 2025 Global C-suite Sustainability Report, 81% of executives already report using AI to advance their sustainability objectives (Deloitte, 2025). The applications are wide-ranging, from optimising energy use in buildings and identifying emissions reduction opportunities in supply chains, to improving the accuracy of ESG data analysis and automating carbon footprint tracking across complex operations. AI can also help reduce greenhouse gas emissions more broadly by improving energy efficiency in industrial systems, optimising electricity grids, and enabling better integration of renewables like wind and solar. Beyond emissions reduction, AI is supporting climate adaptation by strengthening extreme weather forecasting, improving disaster preparedness, and enhancing urban resilience and environmental monitoring through tools such as satellite imagery (UN Climate Change, 2025).
The question for organisations then, is what level of AI usage, if any, is responsible and relevant to the core business needs, goals and values. Crucially, those implementing AI should consider how to use it thoughtfully and in a way that is proportionate, well-governed, and aligned with broader sustainability commitments. Organisations engaging with this now by asking the right questions of providers, building good habits around AI use, and incorporating it into their ESG thinking, will be better placed as expectations from clients, investors, and regulators continue to evolve. Importantly, they can lead the way in demonstrating how innovation and responsibility can indeed go hand in hand.
HOW ESGMARK® CAN HELP
At ESGmark® we help organisations to credibly demonstrate and improve their Environmental, Social and Governance (ESG) credentials. We do this through ESGmark® Certification, carbon footprint measurement, and sustainability support.
To learn more about our carbon footprint service, visit our page here. If you have any questions or would like to book an introductory call to discuss your project, please contact us.
Sources
Deloitte (2025). Global C-suite Sustainability Report. Available at: https://www.deloitte.com
IEA (2025). Energy and AI. International Energy Agency, Paris. Available at: https://www.iea.org/reports/energy-and-ai
Greenhouse Gas Protocol (n.d.). Scope 3 Calculation Guidance. Available at: https://ghgprotocol.org/scope-3-calculation-guidance-2
IEA (2026). Key Questions on Energy and AI. International Energy Agency. Available at: https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
Net Zero Insights (2025). How AI Growth Is Intensifying Data Center Water Consumption. Available at: https://netzeroinsights.com/resources/how-ai-intensifying-data-center-water-consumption/
UNESCO (2025). AI Large Language Models: new report shows small changes can reduce. [online] Unesco.org. Available at: https://www.unesco.org/en/articles/ai-large-language-models-new-report-shows-small-changes-can-reduce-energy-use-90.
UN Climate Change (2025). AI and Climate Action: Opportunities, Risks and Challenges for Developing Countries. Available at: https://unfccc.int/news/ai-and-climate-action-opportunities-risks-and-challenges-for-developing-countries