Table of contents
Generative artificial intelligence is rapidly transforming the global corporate (and not only corporate) landscape, but at what environmental cost? And most importantly, how much awareness is there of the impact it will have on energy demand? Capgemini’s latest report highlights challenges and opportunities in AI sustainability
In the United States, Donald Trump has recently intensified his focus on energy production in the United States, by means of executive decrees, to remove any possible ‘environmentalist’ limits on energy production, the reason? The need to support the increase in energy demand due to the expansion of artificial intelligence.
The share of corporate emissions attributable to Gen AI is set to grow exponentially in the coming years. This is not surprising considering that training a GPT-4 model can consume between 51,772 and 62,319 MWh of electricity, equivalent to the annual consumption of more than 5,000 homes in the US. In Italian data centres, the use of Gen AI could consume 10 per cent more energy over the next two years, putting pressure on IT infrastructures.
And there is not only the energy impact: a single prompt on a language model can require up to 500 ml of water for server cooling. If the entire Italian population used Gen AI on a regular basis, water consumption could become unsustainable.
A recent report by the Capgemini Research Institute, entitled ‘Developing sustainable Gen AI’, rattles off several disturbing numbers on the extent of the energy and water effort that will be absorbed by artificial intelligence. And it takes stock of how much awareness there is in corporate organisations, which are among the main users, of the environmental impact of this technology.
Almost half (48%) of the executives surveyed are aware that the use of generative AI contributes to increased greenhouse gas emissions, but only 27% really care and compare Gen AI models according to their energy implications.
A gap between awareness and action
Despite growing awareness, only 12 per cent of companies using generative AI actually measure its environmental impact, while just 38 per cent claim to be aware of it. This gap is exacerbated by the fact that when evaluating generative AI models, the main considerations remain performance, scalability and cost, relegating sustainability to a marginal role.
“If we want AI to be a resource that can generate sustainable business value, a market-wide debate needs to take place,” said Monia Ferrari, managing director of Capgemini in Italy, emphasising the importance of defining industry standards for reporting on the environmental impact of AI.
The global context of generative AI
The adoption of generative AI has grown exponentially over the past year. According to previous Capgemini research, the percentage of organisations that have integrated this technology into their functions has quadrupled in less than a year, from 6% at the end of 2023 to 24% in October 2024.
This boom is part of a global context characterised by an increasing focus on ESG (environmental, social, governance) issues. Companies are now having to balance the competitive advantages offered by generative AI with their commitments to environmental sustainability, in a regulatory framework that is becoming increasingly stringent, especially in Europe with the AI Act.
Dependence on technology suppliers
A critical factor that emerged from the research is the strong dependence of organisations on their technology partners. With more than three quarters of companies using only pre-trained models and only 4 per cent building their own models, transparency of providers becomes crucial. Almost three quarters of executives find it difficult to measure the environmental impact of generative AI precisely because of limited transparency on the part of providers.
Towards sustainable generative AI
The report suggests a roadmap for a more responsible use of generative AI, which includes:
1. A thorough evaluation of both financial ROI and environmental impact before starting generative AI projects
2. The implementation of sustainable practices throughout the entire AI lifecycle, from hardware to energy used in data centres
3. Using generative AI to accelerate sustainability goals such as material optimisation and circular product design
Some companies are already moving in this direction: one-third of executives are already using generative AI for sustainability initiatives, and two-thirds expect a reduction in greenhouse gas emissions of more than 10 per cent in the next three to five years through sustainable initiatives based on this technology.
The need for effective governance
For almost two-thirds (62%) of the executives surveyed, there is a need to set clear rules and ensure governance that can effectively mitigate the environmental impact of generative AI. This requires multidisciplinary governance models, effective policies and industry-wide collaboration between all stakeholders in the ecosystem.
Capgemini’s report, based on interviews with 2,000 executives from organisations with annual revenues in excess of $1 billion in 15 countries and 12 industries, provides a comprehensive overview of the challenges and opportunities related to the sustainability of generative AI at a crucial time in the evolution of this technology.
The challenge for companies will be to find the right balance between accelerating technological innovation and meeting environmental commitments, in a context where transparency and accountability are becoming increasingly important for all stakeholders. (photo by BoliviaInteligente on Unsplash)
ALL RIGHTS RESERVED ©