Green AI: Making Machine Learning Environmentally Sustainable
summary
Claims that cloud AI power use drives big emission rises—Microsoft reports +30% since 2020, Google +50% since 2019—and global data centers plus networks consume ~700 TWh (about Brazil). Practical levers: favor transfer learning and smaller datasets over training from scratch, use pruning, distillation and quantization to shrink models, run training with demand shifting to cut 45–99% of software carbon, try federated/edge training, speculative decoding and tools like electricitymaps, ML.Boardboard and Bedrock to benchmark and reduce per‑inference cost.
AI-written summary · about
chapters (16)
- 0:00Green AI introduction
- 1:11AI energy demands
- 5:08Data center energy myths
- 8:16Greenhouse gas science
- 11:22Engineers' carbon role
- 13:50Hardware carbon costs
- 16:15Sustainable AI planning
- 22:31Greener data collection
- 24:18Green energy training
- 27:27Model compression
- 31:44Inference optimization
- 33:47Four steps to green AI
- 37:58Summary and resources
- 38:20Q&A: AI vs aviation
- 40:28Q&A: AI future
- 42:56Q&A: Demand shifting
related talks
14mGreener Intelligence: Reducing AI’s Carbon Footprint
30mGreen AI in Cloud Native Ecosystems: Strategies for Sustainability and Efficiency
20mBuilding GenAI That Doesn't Cost the Earth
32mOptimizing AI/ML workloads for sustainability
57mThe Next Decade of Software Is About Climate - What Is the Role of ML?
01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory
44mThe Invisible Monster: The Carbon Footprint of AI
20mSustainability of AI