Fixing the AI infra scale problem by stuffing 1M sandboxes in a single server (Unikraft presentation)

KEY POINTS

1. Felipe Huici presents how Unikraft uses isolation primitives, Linux kernel optimizations, and snapshotting techniques to enable millisecond cold boots, stateful scale-to-zero, and extreme density for sandboxing AI workloads.
2. He reports being able to maintain sub-10ms performance at scale and describes integration with Kubernetes and hardware-level security.
3. This matters because techniques that cut cold-start latency and increase sandbox density can materially reduce cost and improve responsiveness for large-scale AI serving and experimentation.