Introduction to the AI Hypercomputer with Google Cloud

Discover Google Cloud's AI Hypercomputer in this 12-minute talk from AI Infrastructure Field Day. Learn how Google Cloud is investing in infrastructure to simplify AI workload deployment through optimized software and purpose-built hardware. Sean Derrington, Product Manager for Storage at Google Cloud, unveils Google Cloud Managed Luster, a scalable parallel file system developed in partnership with DDN and Exascaler that delivers petabyte-scale storage with sub-millisecond latency and up to a terabyte per second throughput. Explore Anywhere Cache, which enables caching up to a petabyte of data closer to accelerators, and Rapid Storage, which provides up to 20 million QPS per bucket with 6 terabytes per second throughput. The presentation also covers computing advancements including new A4 and A4 Ultra GPU machines, the seventh-generation TPU (Ironwood) with 42.5 exaflops of compute capacity across 9,200 chips, and networking improvements through Cloud WAN and GKE inference for enhanced AI training. Recorded in Santa Clara on April 22, 2025, this talk provides valuable insights into Google Cloud's comprehensive AI infrastructure strategy.