FIRESIDE CHAT: The Future of AI Infra: The Opportunities and Challenges That Await Us On Our Journey
summary
Meta describes building next‑generation data centers with liquid cooling because AI chips draw >5x the power of CPU servers, and it already runs a purpose‑built internal silicon program and an AI Research Supercluster to train models that run for weeks across thousands of GPUs. PyTorch (now in a foundation) and PyTorch 2.0 double training performance, while generative requests cost ~1000x more than ranking—so custom hardware, software stacks and inference tiers are deployed and working at scale.
AI-written summary · about
chapters (13)
- 0:00Introduction and panel
- 0:36Infrastructure team roles
- 2:33Scaling AI at Meta
- 4:23Next generation data centers
- 6:08Integrated hardware design
- 7:41Evolution of AI software
- 9:28AI research and development
- 11:17Impact of generative AI
- 13:18Software challenges in AI
- 15:17Open science and philosophy
- 16:45Future AI trends and data
- 18:32Cross-team collaboration
- 20:18The next decade of AI
related talks
12mKeynote: AI Infrastructure
40mAI Factory - Enabling Scalable Innovation & Intelligent Systems
16mUsing AI and Computer Technologies to Enhance IT Infrastructure
05mAI and ML are changing traditional business
22mCurrent State of AI
20mOperating Frontier Intelligence at Scale
29mUnlocking the Future of AI: Trends, Infrastructure, and Tokenomics
06mThe AI Compute Crisis & Web3 Opportunity