Keynote: Rules of the Road for Shared GPUs: AI Inference Scheduling at Wa... M. Muralikrishnan (ASL)

Don't miss out! Join us at our next KubeCon + CloudNativeCon events in Mumbai, India (18-19 June, 2026), Yokohama, Japan (29-30 July, 2026), and Shanghai, China (8-9 September, 2026). Connect with our current graduated, incubating, and sandbox projects as the community gathers to further the education and advancement of cloud native computing. Learn more at https://kubecon.io Keynote: Rules of the Road for Shared GPUs: AI Inference Scheduling at Wayve - Mukund Muralikrishnan, Staff Engineer, Wayve As AI inference workloads grow in both scale and diversity, predictable access to GPUs becomes as important as raw throughput, especially in large, multi-tenant Kubernetes clusters. At Wayve, Kubernetes underpins a wide range of inference workloads, from latency-sensitive evaluation and validation to large-scale synthetic data generation supporting the development of an end-to-end self-driving system. These workloads run side by side, have very different priorities, and all compete for the same GPU capacity. In this keynote, we will share how we manage scheduling and resources for multi-tenant AI inference on Kubernetes. We will explain why default Kubernetes scheduling falls short, and how we use Kueue, a Kubernetes-native queueing and admission control solution, to operate shared GPU clusters reliably at scale. This approach gives teams predictable GPU allocations, improves cluster utilisation, and reduces operational noise. We will close by briefly showing how frameworks like Ray fit into this model as Wayve scales its AI Driver platform.