This week at KubeCon + CloudNativeCon North America, Rita Zhang, Principal Software Engineer at CoreWeave, received the Lifetime Achievement award for her work over the past decade shaping and supporting the Kubernetes and Cloud Native Computing Foundation (CNCF) communities. Her work reflects how much Kubernetes matters to us at CoreWeave. Today, we’re taking that further: we’ve deepened our commitment to the CNCF by becoming a Platinum member, announced at the event.
Kubernetes powers CoreWeave. We built CoreWeave Cloud on it, and it’s how we run some of the largest AI models in the world at production scale. It runs everything from the control planes behind our customers’ clusters to the bare-metal GPU nodes, with no hypervisor in between.
CoreWeave Kubernetes Service (CKS), our managed Kubernetes environment purpose-built for building, training, and deploying AI applications, is where many of our customers run their workloads. With Slurm on Kubernetes (SUNK), their training jobs and Kubernetes services share the same pool of GPUs. We love what the Kubernetes community builds, from the projects we rely on to the people who maintain them, and we want to do more to support it.
That’s why we are committing to a Platinum membership: to better support the community in the AI-native cloud space. It builds on work we’re already doing. Our engineers, Rita among them, contribute directly to Kubernetes, Kueue, and CloudNativePG, and we take part in the Certified Kubernetes AI Conformance Program. We’re also a founding contributor to llm-d alongside Red Hat, IBM, Google, and NVIDIA, and llm-d is now a CNCF Sandbox project.
Come meet the people behind that work this week. You’ll find them in sessions on Kubernetes authorization, batch scheduling, and topology-aware scheduling, in a hands-on inference tutorial, and at booth 147. If there’s an open source project you’d like to see us contribute to next, tell them at the booth. Here’s the full lineup.
Meet our engineers in breakouts and a hands-on tutorial
Kubernetes authorization with Rita Zhang
Rita Zhang, Principal Software Engineer at CoreWeave, is a Kubernetes Steering Committee member, SIG Auth co-chair, and one of the leads of the Working Group on AI Conformance. On November 10 at 5:30 PM MST in Salt Palace room 251 A-C, Rita’s session, “Authorized? It Depends.”, will cover conditional authorization and how the API server now proves its identity to admission webhooks.
Scheduling AI workloads with Amy Chen
Amy Chen, Senior Software Engineer at CoreWeave, is giving two talks on scaling batch workloads on Kubernetes. On November 11 at 4:10 PM MST in the Grand Ballroom GI, Amy and JJ Jordan of Cohere will present “Quota Isn’t Enough: Topology Aware Scheduling on Modern GPU Clusters,” to explain how Kueue’s Topology Aware Scheduling (TAS) accounts for NVLink domains and rack boundaries, and share the production lessons that came with it.
In addition, on November 12 at 11:50 AM MST in Salt Palace room 250 D-F, Amy and other WG Batch maintainers will present “WG-Batch Updates: What’s New and What Is Next?”, covering what’s new in Job, Kueue, and JobSet, and the working group’s plans through 2027.
Hands-on inference pipeline tutorial with Suraj Deshmukh
Suraj Deshmukh, Senior Software Engineer at CoreWeave, is running “Minting Your Own Tokens: The Server Side of the Token Economy,” on November 11 from 2:20 PM - 3:40 PM MST in Regency Ballroom A at the Hyatt Regency. Attendees will start with a simple CPU deployment, then climb the ladder from single-GPU to tensor-parallel and multi-node pipeline parallelism with LeaderWorkerSet, before demystifying disaggregated prefill/decode and learning to read metrics that matter, like Time-to-First-Token.
Visit us at booth 147
We’re at booth 147 in the Expo Hall all week, and here’s what you’ll find:
- Live demos of training and inference at scale, agent development, and CoreWeave Forge.
- Conversations with our software engineers, product managers, and cloud architects about batch scheduling, AI conformance, and GPU cluster design.
- A chance to tell us which open source projects you’d like us to contribute to next.
Stop by, grab some swag, and come talk about running AI on Kubernetes. If you’re unable to join us live in Salt Lake City, you can watch keynotes and breakout sessions on the CNCF YouTube page.
Get started with CoreWeave
Whether or not you’re in Salt Lake City, here’s how to start building on the same stack our engineers are presenting:
- CoreWeave Kubernetes Service (CKS) runs managed Kubernetes on bare-metal GPUs and is integrated with Kueue for batch and training job queueing. It’s the foundation the rest builds on.
- Inference comes in three options. Serverless Inference is fully managed and gives you APIs for open-source models. Dedicated Inference is built for steady production load. Inference on CKS lets you bring your own runtime, such as vLLM, SGLang, llm-d, or NVIDIA Dynamo.
- CoreWeave Sandboxes give agents that run code an isolated environment, so agent workflows and inference run side by side.
- CoreWeave Forge connects your build, train, and deploy loop, and it works with any model, framework, or cloud.
