News
Updates, product notes and practical coverage for AI platforms and operators.
Google Cloud has outlined an open-source Kubernetes stack spanning reinforcement learning, distributed inference, agent execution and sandboxing. OpenRL, llm-d 1.0, KubeRay, Agent Substrate, gVisor and TorchTPU are intended to help platform teams run specialised models and long-lived agents securely on shared, governed GKE infrastructure. It targets production-scale enterprise use.
NVIDIA has detailed an AI Cluster Runtime recipe that packages Dynamo with pinned Kubernetes infrastructure, validation checks and two cache-aware routing options. The approach separates runtime preparation from model deployment, covering GPU discovery and allocation, scheduling, node health, metrics and ingress so platform teams can reproduce and verify an inference foundation before workloads arrive.
NVIDIA Mission Control 2.5 adds an autonomous recovery engine for Kubernetes-based AI factories. It classifies node health, isolates failing capacity, coordinates remediation and returns recovered nodes to scheduling while tracking stranded or at-risk resources. The release targets long-running distributed training where hardware, networking and software faults make manual intervention too slow at cluster scale.
Google and the llm-d community have released version 1.0 of the open Kubernetes-native inference stack. It adds stable APIs, cache-aware routing, multi-tenant flow control, disaggregated upgrades and asynchronous queues across TPU and GPU fleets. Published tests report large time-to-first-token, throughput and cache-hit improvements for long-context and bursty agent workloads.