Keynote Welcome
summary
Meta describes building an LLM-based outage tool Sevmate that identifies the top five suspect code check-ins to cut outage time, and reports deploying two 24K clusters (Rocky and InfiniBand), scaling to 100K+ and planning a million-plus, gigawatt-plus fleet. Infrastructure examples: a 13.5 million km outside-plant fiber pull, modular 30 MW “tents,” and a 50,000 km cable Project Waterorth; a Manhattan-scale cluster Hyperion targets 5 GW, while GPU scale-up moves from 8 to 72 toward 144+ and forces new fabric and transport designs.
AI-written summary · about
chapters (7)
- 0:00AI transformation at Meta
- 2:26Infrastructure for AI scale
- 4:42AI models and workflows
- 7:30Inference and user experience
- 9:18Physical global infrastructure
- 13:33Performance and optimization
- 17:08Resilience and future outlook
