Research highlights negative environmental impact of GenAI
Only 12% of those using GenAI measure its carbon footprint
Many organisations are failing to track the carbon impact of their generative AI use cases. This is jeopardising ESG targets and increasing the wider environmental impact of technology.
Following on from yesterday’s announcement by the government that it wanted to “mainline AI into the veins” of the UK, research published today by Capgemini suggests that this could have a significant environmental cost.
The research also uncovers the fact that many organisations are choosing not to track the growing carbon footprint of generative AI, and this in turn is making all those “carbon neutral by 2030 pledges look increasingly meaningless.
Generative AI adoption has accelerated rapidly, with previous Capgemini research showing that while only 6% of organisations had integrated generative AI across their business functions and locations as at end-2023, that figure had risen to 24% as at October 2024.
Whilst these technologies have the potential to improve energy efficiency in the long term, right now the training of LLMs requires vast amounts of compute which is power and water hungry. The problem is exacerbated by the fact that the GPU chips required for GenAI are made from metals including copper, cobalt, tungsten, lithium, germanium, palladium, lead, chromium, cadmium, mercury, and others. The mining of these metals contributes about half of the GHG emissions associated with these chips.
Worse still, LLM training burns through hardware at a faster rate so the whole lifecycle of hardware is shorter and the materials impact greater.
A problem pinpointed by the research is that the pace of GenAI innovations has outpaced sustainability reporting. Whilst nearly half (48%) of executives believe that their use of GenAI has driven a rise in GHG emissions, only 12% of executives that use GenAI say their organisation measures the environmental footprint of their use, and only 38% claim to be aware of that environmental impact.
Similarly, as companies try to keep up with competitors, performance, scalability and cost are key considerations for GenAI model evaluation, while sustainability is only of marginal importance. Only one fifth of executives rank the environmental footprint of GenAI as a top 5 factor when selecting or building GenAI models.
Robust governance, collaboration and transparency can mitigate impact
It’s not all bad news for fans of a habitable planet. 31% of organisations have taken steps to incorporate sustainability measures into the GenAI lifecycle. For example, over half are either already using smaller models and powering GenAI infrastructure with renewable energy sources or plan to do so in the next 12 months.
However, with more than three-quarters of organisations using only pre-trained models and just 4% building their own models from scratch, executives are heavily reliant on their technology partners when it comes to addressing the environmental footprint of Gen AI.
Nearly three quarters reported finding it challenging to measure the technology’s footprint due to limited transparency from providers.
The Cap Gemini report sets out some recommendations for businesses. One is the suggestions that organisations conduct a thorough assessment both of the financial ROI and environmental footprint of their GenAI projects before launch. They should consider whether they need energy-intensive GenAI technologies in cases where they could use another technology for a similar result.
It also proposes that sustainable practices should be implemented throughout AI’s lifecycle, including hardware, model architecture, energy sources for data centres, and implementing sustainable usage policies.
The report also highlights some of the more constructive use cases for GenAI in sustainability terms such as in ESG reporting and scenario planning, material optimisation for key industries, or sustainable/circular product design. One-third of executives are already using GenAI for sustainability initiatives and two thirds say they expect a reduction of more than 10% in GHG emissions in the next 3-5 years as an output of GenAI-led sustainable business initiatives.
However, given how few businesses are even measuring the carbon impact of their GenAI use, this assumption looks, at best, optimistic. Almost two-thirds (62%) of executives believe that robust guardrails and governance can effectively mitigate GenAI’s environmental impact. Again, this looks optimistic, but governance models and effective policies and industry-wide collaboration between stakeholders across the GenAI ecosystem are vital if we are to avoid the worst-case emissions scenarios.
“If we want Gen AI to be a force for sustainable business value, there needs to be a market discussion around data collaboration, drawing up industry-wide standards around how we account for the environmental footprint of AI, so business leaders are equipped to make more informed, responsible business decisions, and mitigate these impacts,” said Cyril Garcia, Capgemini’s Head of Global Sustainability Services and Corporate Responsibility and Group Executive Board Member.
“AI has the potential to accelerate business objectives and sustainability initiatives. We are proposing here practical steps to follow for business leaders to fully harness technologies such as GenAI and deliver a positive impact for organisations, society and the planet.”