In a recent article, Capgemini Invent highlights how organisations accelerating from AI pilots to enterprise-wide adoption are encountering rising energy use, growing infrastructure demands and increased regulatory scrutiny. While larger models are often deployed by default, they can exceed what specific use cases require, adding unnecessary cost, complexity and environmental impact.
A more targeted approach is emerging, with leading organisations rethinking the models themselves rather than scaling infrastructure indefinitely. By rightsizing large language models, enterprises can reduce compute requirements and emissions while maintaining strong performance, making this shift increasingly important for sustainable, cost-effective AI at scale.
Read more on their website.