Enterprises Must Make Smarter Choices to Curb GenAI Emissions
A recent report reveals that training modern models consumes energy equivalent to the annual usage of 5,000 US homes, while running these models in operation demands equal or greater energy.
Notably, GenAI’s share of organisational carbon footprints is predicted to rise from 2.6% to 4.8% by 2026. Additionally, it estimates that one large language model query uses ten times the energy of a Google search and that the technology could generate up to five million metric tons of e-waste by 2030.
Capgemini suggests companies adopt smaller, less resource-intensive models, leverage renewable energy, and improve transparency with AI vendors. The report also highlights the importance of identifying cases where simpler AI tools could suffice, reducing both emissions and costs. Sustainability, they stress, must be embedded into AI strategies.