Environment Variables Sustainable AI
Sep 26, 2025
Boris Gamazaychikov, Head of AI Sustainability at Salesforce, discusses the critical intersection of artificial intelligence and environmental responsibility. He highlights the varying energy impacts of AI models and the launch of the groundbreaking AI Energy Score for benchmarking. Boris explains how specific models can lead to massive efficiency gains, the necessity of aligning AI workloads with renewable energy, and the importance of transparency in AI development. He encourages embracing AI while advocating for smarter, sustainable strategies in the tech landscape.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Has A Huge Energy Spectrum
- AI covers a huge spectrum of models whose energy impacts range from bicycle-level to jet-plane-level differences.
- The AI Energy Score standardizes comparisons so users can pick the right model for the right task.
Real World Efficiency Examples
- Boris described seeing orders-of-magnitude energy differences between models, citing a 60,000x delta he has shared widely.
- He also recounted Salesforce using fine-tuned models in a trust layer that improved efficiency by about 99%.
Three Pillars For Sustainable AI
- Manage AI demand: apply AI only where it adds value and pick smaller domain-specific models where possible.
- Combine demand management with efficiency and clean energy supply to make AI sustainable.
