Keynote presentation exploring how traditional load balancing methods fail for LLM workloads due to the variable computational demands of different prompts, model differences, and autoregressive processing. Learn about new Kubernetes APIs specifically designed for routing LLM workloads, allowing configuration of serving objectives and priorities for different use cases. The speakers demonstrate how these APIs integrate with Gateway API and can be implemented across various Gateway API implementations for turnkey LLM routing support. Through real-world examples, see the significant efficiency improvements possible with LLM-aware load balancing strategies, especially important as model multiplexing techniques like LoRA introduce additional complexity to the serving landscape.