SustAInability: navigating the environmental challenges of Gen AI

While generative artificial intelligence (Gen AI) promises to transform organizations, its environmental impact remains significant… and difficult to measure. Our report Developing sustainable Gen AI sets out a roadmap for creating responsible Gen AI for sustainable business value. 

What’s everyone saying?  

• As of today, large language models (LLMs) require processing vast amounts of data and significant computational power, which consumes large quantities of electricity, water, and other resources. Training a GPT-4 model (which includes a massive 1.76 trillion parameters) requires an amount of electricity equivalent to the annual consumption of 5,000 US homes, and running an inference of 20 to 50 queries on an LLM uses about 500 milliliters of water each time. 
• By 2030, Gen AI will have produced 1.2 to 5 million metric tons of e-waste (1,000 times more than was produced in 2023!), considerably impacting its users’ scope 3 greenhouse gas (GHG) emissions and potentially jeopardizing the ESG objectives of many organizations. 
• Organizations are progressively becoming aware of this environmental impact but remain focused on their business objectives, and unsure how to assess the environmental footprint.  

What do we have to say? 

• While they are aware Gen AI is GHG-emissive, executives are not making this impact a priority, focusing instead on cost-effectiveness, productivity, and efficiency. In fact, over half of executives say driving efficiency is more important than measuring impact, and just 20% rank the environmental footprint of Gen AI among the top five factors when selecting or building Gen AI models, with performance, scalability, and cost dominating their consideration process.  
• Further, though aware of the sustainability impact of Gen AI, executives are at a loss to track this growing footprint, our research shows. While 42% have had to relook at their climate goals due to Gen AI’s growing footprint, only 38% claim to be aware of the environmental impact of the Gen AI they use.  

• With over 75% of organizations using only pre-trained models and just 4% building their own models from scratch, executives are heavily reliant on their technological partners when it comes to addressing the environmental footprint of Gen AI. As a result, a majority find it difficult to measure Gen AI’s environmental impact, due to limited transparency and lack of clear, precise data from providers.  

• This is compounded by the absence of universal standards for measuring and presenting this environmental footprint.  
• Progress is to be expected thanks to an ongoing innovation race that could well produce smaller, lighter, and more efficient Gen AI models – a win for sustainability. Additionally, Gen AI could achieve positive sustainable impact thanks to applications such as sustainable product design or waste management. The bottom line: in the future, AI has immense potential to support climate action and enable a more sustainable future.  
• Yet in the short to medium term, Gen AI’s sustainability impact will have significant consequences on organizations’ costs and scope 3 emissions. To implement Gen AI projects while anticipating their sustainability impact, organizations should seek to “look at the big picture,” analyzing both Gen AI’s business and sustainability implications simultaneously.

Who’s doing it right? 

• Capgemini is actively helping firms from a wide range of industries use generative AI to drive sustainable business value, from one of the largest automotive manufactures in the world, which is harnessing Gen AI to design, build, and virtually test more fuel-efficient engines and vehicles, to one of Europe’s largest retailers, which relies on Gen AI to provide more natural, environmentally responsible products to customers through a chatbot with personalized shopping agents, with even the last mile delivery optimized for carbon reduction. Lastly, one of the largest consumer packaged goods companies on the planet is leveraging Gen AI to reformulate and source its products from ethical and organic natural farms, simulating and optimizing every aspect of the product from the composition of the product itself to its packaging and distribution. It is removing carbon from the entire lifecycle of its value chain.  
• Additionally, Capgemini is actively supporting the development of AI models with lower carbon footprint, whether through the recent extension of our partnership with Mistral AI , which offers high-performing open models with lower CO2 footprint installations, or through our investment in Massachusetts Institute of Technology spinoff Liquid AI , which focuses on developing environmentally responsible Gen AI models.  
• Finally, in early February, we introduced a breakthrough Gen AI-driven methodology for protein engineering, which requires 99% less data than before, and can significantly improve their ability to break down PET plastic, a promising solution to tackle global plastic waste – and proof of Gen AI’s sustainability-enabling potential. 

What’s the bottom line? 

We outline how organizations can design and implement a responsible and sustainable generative AI strategy.  

• Identify the right technology that addresses their specific business needs. Businesses should conduct a thorough assessment of both the financial ROI and environmental footprint of their Gen AI projects before launch. They should consider whether they need energy-intensive LLMs in cases where they could use another technology, or smaller, prebuilt models, for a similar result.  
• Pick the right use cases for sustainable Gen AI. This entails building the business case for sustainable Gen AI (identifying and prioritizing the right use cases to tap into Gen AI, based on financial and environmental costs, benefits, associated risks, and alignment with ESG goals), evaluating Gen AI partners and models on sustainability parameters, and monitoring and reporting their Gen AI footprint.  
• Mitigate Gen AI’s environmental impact. This means implementing sustainable practices throughout Gen AI’s lifecycle, by prioritizing partners with more energy-efficient and recyclable hardware specifically designed for AI/Gen AI, using smaller, task-specific pre-trained models, considering providers that use low-carbon energy sources, and tracking and quantifying the carbon footprint of Gen AI.  
• Develop the right data and technology foundations and develop the skill sets to derive maximum benefit from Gen AI in sustainability. Building the right data foundations and developing the required skill sets are the keys to deriving maximum benefits from Gen AI. Organizations can also evaluate the potential of AI agents to create sustainable business value in ESG reporting and compliance-related areas.  
• Build an effective multidisciplinary governance structure and collaborate with various players within the ecosystem. This means implementing multidisciplinary governance models to ensure safe, transparent, ethical, and sustainable usage and design of Gen AI, setting up a governance body to implement accountability measures, and creating a framework to approve or reject Gen AI/AI pilots. 
• The environmental impact of Gen AI is a very complex topic for organizations. Working with multiple partners adds to that complexity, not to mention potential conflicts of interest may arise between partners specialized in sustainable impact analysis and others in Gen AI business outputs, forcing organizations to make complicated decisions. Organizations should therefore seek out a partner who can address the two aspects. 

Looking for more?  

• We’ve just partnered up with the École normale supérieure and the AI and Society Institute to launch a global Observatory on AI’s environmental impact. The goal: to establish a robust, shared methodology for measuring the environmental impact of AI technologies. 

And you, what are you saying?  

Has environmental impact influenced your AI adoption strategy? Are you exploring innovative ways to make your AI projects more sustainable? Head to the comments section below to share your experience with us!