Presentation: Fixing the AI Infra Scale Problem by Stuffing 1M Sandboxes in a Single Server

Felipe Huici of Unikraft has detailed a method for deploying up to one million sandboxed environments on a single server to handle AI workloads. By utilizing kernel optimizations and snapshotting techniques, this approach achieves sub-10ms cold start times while maintaining high density. This innovation addresses infrastructure bottlenecks by allowing more efficient scaling and isolation for resource-intensive applications.