Panels Archive

Building Gigawatt-Scale Infrastructure for AI


Moderator: James Hongyi Zeng (Meta Platforms, Inc.)
Panelists: Maria Garzaran (Advanced Micro Devices (AMD)), Greg Steinbrecher (OpenAI), Satoshi Matsuoka (RIKEN Center for Computational Science (R-CCS)), Dean Hildebrand (Google LLC), Pavan Balaji (Meta Platforms, Inc.)

Abstract: The explosive scale of generative AI models demands highly specialized infrastructure. This panel explores the critical challenges of building and scaling dedicated AI clusters tailored specifically to the modern model lifecycle: massive-scale pre-training, nuanced post-training, and high-throughput, capacity-demanding inference. As models scale to trillions of parameters, infrastructure must deliver unprecedented compute density, ultra-low latency interconnects, data processing, reliability requirements, parallelisms, and continuous I/O throughput. Our panelists will dissect the unique hardware and software bottlenecks inherent to each phase. We will debate the most effective strategies for mitigating hardware failures at scale, optimizing distributed storage pipelines, and tackling power constraints. This interactive 90-minute session will equip attendees with actionable insights into architecting the resilient, high-performance systems required to power the next generation of generative AI breakthroughs.

Back to Panel Archive Listing