Sovereignty Is an Architecture Decision, Not a Compliance Checkbox
Wednesday, June 24, 2026 3:00 PM to 3:20 PM · 20 min. (Europe/Berlin)
Hall H, Booth L01 - Ground Floor
HPC Solutions Forum
AI FactoriesHPC in the Cloud and HPC ContainersIndustrial Use Cases of HPC, ML and QCSovereignty in AIStorage Technologies and Architectures
Information
Most organisations treat data sovereignty as a compliance requirement to be satisfied after the architecture is designed. DCAI built it the other way around — and the difference turns out to matter enormously for what workloads are even legally possible.
In this session, Chris Ott (WEKA) and Ali Syed (DCAI) walk through the specific decisions that made Gefion one of Europe's first fully sovereign, end-to-end encrypted AI supercomputing platforms: how the data layer was designed to eliminate GPU stalls, how multi-tenant isolation was achieved without a throughput penalty, and why certain government contracts only exist because of encryption choices made at the infrastructure design stage.
This is not a roadmap session. It is a postmortem on a production system — what worked, what the trade-offs were, and what the on-prem versus cloud question actually looks like when sovereignty is a hard constraint, not a preference.
In this session, Chris Ott (WEKA) and Ali Syed (DCAI) walk through the specific decisions that made Gefion one of Europe's first fully sovereign, end-to-end encrypted AI supercomputing platforms: how the data layer was designed to eliminate GPU stalls, how multi-tenant isolation was achieved without a throughput penalty, and why certain government contracts only exist because of encryption choices made at the infrastructure design stage.
This is not a roadmap session. It is a postmortem on a production system — what worked, what the trade-offs were, and what the on-prem versus cloud question actually looks like when sovereignty is a hard constraint, not a preference.
HPC Solutions Forum Questions
How are high-performance data platforms evolving beyond individual technologies like parallel file systems and burst buffers?What is most important: maximizing performance in a given power envelope, minimizing power costs, or being green? Do you have to choose?Discuss your solution in terms of benefits for specific use cases, rather than general horizontal terms like HPC, AI, performance, or scalability.
Format
on-site
