Istio announced beta support for Ambient Multicluster alongside a beta Gateway API Inference Extension at KubeCon, extending its lower-overhead ambient service mesh mode across multiple Kubernetes clusters and adding traffic-routing primitives designed specifically for AI model inference rather than generic HTTP services. Ambient mode was Istio's answer to the long-standing complaint that sidecar-based service meshes add too much per-pod resource overhead and operational complexity for what many teams actually need; extending that lighter architecture across clusters, rather than confining it to a single cluster, matters for any organization running Kubernetes across multiple regions or cloud accounts for redundancy or data residency reasons. The Gateway API Inference Extension is the more novel piece: it brings inference-aware routing, meaning the mesh can make traffic decisions based on model-serving-specific signals like queue depth or GPU utilization on backend inference pods, using the same Kubernetes-native Gateway API resources teams already use for regular ingress, rather than requiring a bolt-on, inference-specific load balancer. That matters because as more teams self-host LLM or embedding model inference on Kubernetes rather than calling a hosted API, the traffic patterns look nothing like a typical stateless web service: requests have wildly variable latency and cost, and naive round-robin or least-connections routing can badly misallocate expensive GPU capacity. Istio also added experimental support for agentgateway as a data-plane component, aimed at AI agent traffic specifically. For platform teams building internal AI infrastructure, this is a sign that the service mesh ecosystem is starting to treat inference workloads as a first-class traffic pattern rather than forcing them through tooling designed for conventional microservices.