Environmental footprint of AI

What it is

AI is resource-intensive. Training large models consumes substantial electricity, and running them consumes more in aggregate, because every query uses computation. Data centres use water for cooling and specialised hardware with a short working life. Most organisations incur this footprint indirectly, through a cloud or model provider, which makes it easy to overlook. The risk is a gap between what the organisation says about its environmental impact and what its growing use of AI actually costs, whether through missed targets, incomplete reporting or claims that cannot be supported.

How it happens

• AI adoption is planned with no estimate of its energy use.
• The largest available model is used for tasks a much smaller one could do.
• Providers give no usable data on the footprint of their systems.
• Emissions from bought computing are left out of reporting.

Why it matters

Organisations with published climate targets or sustainability reporting duties must account for emissions in their supply chain, and AI is a growing part of it. Understating the footprint, or promoting AI as green without evidence, risks a charge of misleading claims. The EU AI Act asks providers of general-purpose models to document energy consumption. Cost follows the same curve as energy, as AI capacity and cost overrun describes.

Warning signs

• Sustainability reporting does not mention AI or cloud computing.
• No one can estimate the energy used by the organisation's AI systems.
• Model choice ignores efficiency.
• Supplier due diligence does not ask about environmental impact.