The Future of Data on Kubernetes: From Database Management To AI Foundation
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 14:30· Level 1 | Hall Entrance S10 | Room D·
summary
DoK now runs >50% of workloads in production and 75% of those are advanced; expect concrete tactics not theory: use local SSD caching to cut GPU idle time, parallelize downloads for large models (Llama 70B ~13 GB, ~20 minutes from Hugging Face), store models on immutable block storage, and run vector DBs via PostgreSQL + pgvector. Pipelines use Kafka + Flink with LLM hooks, VLLM/RunAI or Triton for inference, CSI drivers for storage, and isolate DB nodes to get predictable costs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:05Panelist backgrounds
- 6:06Customer stories: AI on K8s
- 11:10Database considerations
- 14:05AIML ops at Google
- 19:01Streaming data patterns
- 21:50Future of AIML on K8s
- 26:33Cost control strategies
- 30:05Closing & Q&A
related talks
27mHarnessing Kubernetes for Scalable AI/ML Workloads
17mDevOps for Private AI: A Scalable Platform Using Kubeflow
54mEconomics and Best Practices of Running AI/ML Workloads on Kubernetes
04mOpening Remarks
40mPanel: Kubernetes and GPU Trends for AI in Financial Services
22mGenerative AI Model Data Pre-Training on Kubernetes: A Use Case Study
38mAI in Kubernetes: How to Get Started?
33mCompliant Data Management and Machine Learning on Kubernetes