AI has become increasingly embedded in how businesses operate. From data analysis and extraction to content creation and process automation, its applications have expanded rapidly. But as AI use grows, so does the infrastructure and energy required to support it.
This was one of the topics discussed at the recent Carbon Accounting Alliance Conference, including the work being done by the Sustainable AI Group to better understand the environmental impact of AI. Their CLEER Dashboard provides a way of estimating the energy and carbon impacts associated with different AI models and workloads, helping to address some of the transparency challenges around AI use.
The conference raised an interesting question: as AI becomes more embedded within business, how can we make sure we are using it more responsibly?
AI is becoming part of everyday business
AI has grown rapidly since the emergence of generative AI in 2022 and is increasingly being integrated into everyday business activities. It can process large amounts of information, support data analysis, automate repetitive tasks and improve the efficiency of existing processes.
The question is therefore becoming less about whether organisations will use AI, and more about how they can use it responsibly and effectively.
The hidden infrastructure behind AI
The capabilities AI has can feel almost limitless. You type a prompt and receive an answer in seconds. Due to the intangible nature of the interaction, it can be easy to forget about the physical infrastructure sitting behind it.
AI relies on data centres, electricity, cooling systems, servers, semiconductors and other physical infrastructure. Those systems require resources and eventually contribute to electronic waste. Organisations using AI extensively therefore shouldn’t treat it as an impact-free service.
AI’s growing energy demand
The scale of this infrastructure is growing quickly. The International Energy Agency projects that electricity generation required for supplying data centres will rise from around 460 TWh in 2024 to more than 1,000 TWh in 2030, with data centres accounting for around 3% of global electricity generation. AI is not responsible for all that demand but is contributing to its growth.
The environmental impact also varies depending on where and how AI workloads are run. Data centres differ considerably in their efficiency and electricity sources, meaning the same type of digital activity does not necessarily have the same environmental impact everywhere.
For example, cooling alone can represent roughly 7% of electricity consumption in an efficient hyperscale data centre, and more than 30% in a less efficient enterprise data centre.
Responsible use does not mean avoiding AI
None of this means businesses should stop using AI. Instead, responsible AI use can start with simply asking whether we need AI for a task.
Before deploying AI, businesses can consider what purpose it serves, which model is appropriate and whether the same task could be achieved effectively using conventional software. Different AI models and workloads can have very different impacts, so using the right tool for the task rather than using AI for AI’s sake can make a meaningful difference.
This is particularly relevant as AI becomes embedded into routine business processes. The Sustainable AI Group’s research suggests that, for the same task, model choice can result in more than a 30-fold difference in energy use. It also found that a typical agentic session used considerably more energy than a typical chat session.
Therefore, the question is about which AI, for what purpose, and how much of it do we need?
Understand, Measure, Reduce, Demand
One framework raised at the conference was to think about responsible AI use through four stages: understand, measure, reduce, demand.
Understand: Build a clearer picture of how AI is being used within your organisation. What models are being deployed? What options are available? And where possible, what do you know about the infrastructure supporting them?
Measure: Start tracking usage and gathering the information available to estimate its impact. Token usage can provide one input, although measuring AI’s impact is still an evolving area and not perfectly precise.
Reduce: Identify necessary use and consider whether smaller or more energy-efficient models could complete the same task. The aim shouldn’t necessarily be to use less AI at all costs, but to avoid consuming resources without creating meaningful value.
Demand: Ask more of AI providers. Businesses can push for greater transparency around energy use, infrastructure and environmental performance. Internally, they can also continue to simply ask: do we need AI here?
Starting with better questions
Businesses don’t need to understand every technical element of AI infrastructure before they start acting. What matters is beginning to ask better questions about where it is being used, the value it creates, and the impact associated with it.
That could mean understanding where AI tools are already being used across the organisation, considering whether the model being used is proportionate to the task, and building environmental impact into decisions about AI adoption alongside factors such as cost, security and performance.
The goal isn’t to make every AI interaction perfectly measurable. It is to move towards more informed and intentional use.
What does this mean for carbon accounting?
There is an interesting tension here with the carbon accounting industry. AI has the potential to support activities such as data extraction, categorisation, cleaning and analysis. Used well, it can make some elements of carbon accounting quicker and more efficient.
However, the industry should also consider the environmental impact of the tools being used to achieve those efficiencies. If AI is going to help us measure environmental impact, we need to get better at understanding and measuring the impact of AI itself.
Making AI use more intentional
Responsible AI use doesn’t mean less innovation. AI itself could contribute to environmental solutions and help businesses make better use of increasingly complex data. Responsible use should focus on creating as much value as possible while minimising unnecessary resource use.
Measurement is still imperfect, transparency remains limited and technology is moving quickly. The CLEER methodology itself recognises this challenge, particularly for models where the underlying infrastructure and energy consumption are not directly visible to users.
But imperfect information shouldn’t become a reason for inaction. We don’t need to know everything about AI’s environmental footprint before we evaluate how we use it.
Ultimately, responsible AI is about becoming more intentional. Understand where AI is being used, measure what you can, reduce unnecessary consumption and demand greater transparency from the organisations providing the technology.
As AI becomes a bigger part of everyday business, asking those questions now will help organisations ensure the technology delivers genuine value without losing sight of the environmental impact sitting behind it.