Istio dropped two big releases at KubeCon + CloudNativeCon Europe 2026: Ambient Multicluster in beta and the Gateway API Inference Extension in beta.
Ambient Multicluster Beta
Istio’s ambient mode now supports multicluster deployments, giving you cross-cluster traffic routing without sidecars. The key capabilities include:
- Dynamic cross-cluster failover where requests automatically redirect to another cluster when a service failure is detected
- ztunnel-to-ztunnel mTLS for zero-trust security across cluster boundaries
- Simplified operational model compared to sidecar-based multicluster deployments
This eliminates the sidecar tax for cross-cluster traffic, a pain point that’s plagued multi-cluster Kubernetes teams for years.
Gateway API Inference Extension
Istio also integrated the Gateway API Inference Extension, which effectively turns any Istio gateway into an inference gateway. The extension adds:
- Model-aware routing so you can route requests based on the target model
- Per-request criticality levels to prioritize inference requests
- Load balancing based on real-time model server metrics for better GPU utilization without manual tuning
This means platform teams can use their existing Istio service mesh infrastructure to route to LLM backends with model-aware intelligence, without deploying a separate inference proxy.
The Bigger Picture
Ambient multicluster removes the sidecar overhead for cross-cluster networking while keeping zero-trust security intact. The inference extension is equally useful: instead of bolting on a separate inference gateway, your existing mesh handles model routing natively. Two fewer things to deploy and operate.