High-performance serverless infrastructure for AI. Run inference, sandboxes, and intensive workloads on CPUs and GPUs