Scaling Data in a Sovereign AI Platform
summary
A blueprint for a European sovereign AI platform that pairs Dell infrastructure with a high-efficiency Java stack: Eclipse Store/Data Grid plus JVector as an embedded vector DB and cache on Kubernetes, Slurm for HPC jobs, and GraalVM native images to start storage-cell groups in milliseconds. Claims include 80% infrastructure cost savings, support for petabyte-scale long-term memory, federated factories totaling ~200 MW for large sites, and a released version 4 of the stack ready for pilot-to-production use.
AI-written summary · about
chapters (16)
- 0:00Welcome and Historical Context.
- 2:50Defining Sovereign AI in Europe.
- 5:18Responsibility and the European AI Act.
- 7:11The Federated Factory Model.
- 11:53Infrastructure and Security Layers.
- 14:32Data Center Physicality and Liquid Cooling.
- 17:05Data Platforms and Federated Queries.
- 19:32Model Hubs and Software-Defined Infrastructure.
- 20:52Silicon Independence and Ecosystem Partners.
- 24:14Introduction to Marcus and the Java Stack.
- 27:36Eclipse Store and Data Grid Technology.
- 30:02Energy Efficiency and Monolithic Scaling Issues.
- 34:09The Principle of Storage Cells.
- 37:32GraalVM Native Images for High Performance.
- 41:05Autonomous Knowledge and Sustainability.
- 43:10Q&A.49:29
related talks
31mHow to build a sovereign European AI compute infrastructure
29mFrom Experiment to Enterprise: Architecting AI for Stability and Scale
32mSovereign AI
28mScaling agentic AI: From infrastructure to enterprise outcomes
40mAI Factory - Enabling Scalable Innovation & Intelligent Systems
21mAIaaS with Dell AI ready infrastructure
01h 26mSovereign AI Navigating Strategic Autonomy...
20mSCS: Large federated infrastructure for sovereignty