Breakouts
BRK2-124 • Kubernetes, Cloud Runtimes • Technical
How Anthropic uses GKE and TPUs to scale resilient AI
location_on
South Seas H
schedule
5:15 PM - 6:00 PM
Hardware failures and resource fragmentation shouldn’t stall AI innovation. Join this session to discover how Anthropic uses Google Kubernetes Engine (GKE) architecture to accelerate workload scheduling with modular compute cubes. Dive deep into how to architect a new GKE Dynamic slice management layer to recover from failures faster. Learn to reshape capacity on the fly and isolate faulty units without the slow delete-and-recreate cycle. And take away a blueprint for smarter bin-packing to maximize fleet utilization and performance at scale.
Read more