Day 2 – April 23, 2026
Breakouts
BRK2-121 • Kubernetes, Cloud Runtimes • Technical
AI at Snap's scale: Building a global compute fabric with GKE custom compute classes
location_on
Jasmine A
schedule
1:30 PM - 2:15 PM
Hyperscale infrastructure hits a wall when integrating AI into global service meshes. To orchestrate obtainability-constrained accelerators, Snap and Google Cloud engineered a fundamental shift in Google Kubernetes Engine (GKE) scaling. In this session, we’ll reveal the architecture of custom ComputeClasses, a feature born from Snap’s requirements that is now a standard part of GKE. Join us to learn to architect a mesh that mixes diverse AI compute resources to overcome supply constraints and scale workloads. And leave with the blueprint for optimizing performance and capacity at planetary scale on GKE.
Read more