Keynote: An Optimized Linux Stack for GenAI Workloads
Talkat KubeCon + CloudNativeCon China 2025 · on Tuesday, 2025-06-10· at 09:43· Level 16 | Grand Ballroom I·
summary
Argues that container sidecars break performance and portability for GenAI and shows a practical stack: Flatcar Linux + WasmEdge running portable Wasm inference to avoid CUDA/driver mismatches. Demonstrates a GenAI app running across GPUs and CPUs, cites deployments on 600,000 machines and a customer processing 500,000 videos per day, compares a single 50 MB wasm binary to an 8 GB Python image and targets ~100 ms latency for interactive pipelines.
AI-written summary · about
chapters (7)
- 0:00AI as cloud workloads
- 1:04AI breaks cloud-native portability
- 2:48Sidecar model limitations for AI
- 4:55HTTP APIs fail for real-time AI
- 6:18Portability with WASM for AI
- 8:02Embedding AI API servers
- 9:25Future of WASM API integration
related talks
25mLeveraging Wasm for Portable AI Inference Across GPUs, CPUs, OS & Cloud-Nativ...
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
32mKeynote: Powering Generative AI with Kubernetes
29mSupercharge Your Kubernetes AI Deployments with Wasm
11mKeynote: Open Source Stack Empowers the AI Era
36mApplied AI on the edge
33mReady, Set, Go: WASM-Powered Containers Taking Your AI To New He...
33mWebAssembly based AI as a Service with Kubernetes