Building leaner and greener AI

A novel tool compresses AI models to make them faster, cheaper, and more sustainable

Spotted: The rapid expansion of artificial intelligence is revolutionising industries, but it comes with a hefty price tag – both financially and environmentally. AI training is expected to drive the bulk of the electricity demand, at 40 terawatt hours (TWh) in 2026 and 402 TWh by 2030, many companies find themselves grappling with rising costs and squeezed resources. Addressing these challenges, a new innovation offers a way to streamline AI model development while drastically cutting costs and emissions.

Pruna AI’s optimisation engine simplifies and accelerates the deployment of AI models by compressing them without requiring extensive reengineering. Using techniques like pruning, quantisation, and caching, the platform reduces the size of models, making them faster and more efficient on any hardware setup. This enables machine learning teams to bypass the painstaking and time-consuming process of fine-tuning models for each deployment, saving both time and resources.

What sets Pruna AI apart is its dual focus on efficiency and sustainability. Compressed models not only reduce computational costs but also significantly lower energy usage, cutting carbon emissions by up to 91 per cent. By making AI models leaner, the platform also extends the lifespan of hardware, reducing the need for frequent replacements. As Bertrand Charpentier, Pruna AI’s co-founder, explains, “Our goal is to help engineers develop AI in the fastest, cheapest, and most sustainable way possible.”

Still in its early stages, Pruna AI has already attracted $6.5 million in seed funding led by EQT Ventures, alongside other notable investors. The company plans to expand its team and expedite its product rollout. Pruna AI’s inclusion in the European AI programme by Meta, Hugging Face, and Scaleway highlights its potential to democratise cutting-edge AI tools for businesses of all sizes.

Written By: Oscar Williams