From the NVIDIA edition of October 21, 2026
UK and France plan federated learning across national AI supercomputers (Isambard-AI, Jean Zay, Adastra) so sensitive data stays in-country; announced at GTC Berlin with NVIDIA technical support
Bristol Centre for Supercomputing (University of Bristol); GENCI (France); IDRIS; CINES; University of Cambridge; NVIDIA · Public research computing (national AI supercomputers) · United Kingdom and France
- Publisher
- University of Bristol, Bristol Centre for Supercomputing
- Original publication
- 2026-10-21
- Source retrieved
- 2026-10-22
- Event date
- 2026-10-21
What happened
BriCS and GENCI announced a federated learning plan under the UK-France ConcordAI partnership to connect Isambard-AI (NVIDIA Grace Hopper, part of the government-funded AI Research Resource) with France's Jean Zay (NVIDIA) and Adastra (other hardware), so researchers in both countries can train larger models while each keeps control of its own machines and data. Each site trains locally and shares only model updates, so sensitive data in fields such as healthcare, defence or aerospace need not leave its country. Teams at Bristol, Cambridge, IDRIS and CINES, with NVIDIA's support using the open-source NVIDIA FLARE, have been testing with increasing model sizes and checking results match single-machine training; direct connection is targeted by the end of 2026. The release also notes Juelich and GENCI interest in linking JUPITER AI Factory and AI France Factory infrastructure, and GENCI plans to connect its supercomputers with French hyperscalers for startups and industry.
Why it matters
Directly relevant to public research computing: two publicly funded national systems are sharing capacity across borders without moving data, including across mixed vendor hardware. The model transfers to U.S. state or university consortia (for example, multi-institution health research) that face data-residency or data-use restrictions, with the caveat that it is a plan with testing under way, not a production service.
Evidence and measured results
Joint announcement by the public operators, framed as a plan. It reports testing activity but no results, model sizes, performance overheads or governance agreements. NVIDIA's role is technical support and software.
Limitations and uncertainty
Announcement of a plan; no published test results, overhead figures or allocation rules; completion target end of 2026.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-22; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Researchers need more compute than one centre holds but cannot move sensitive data across jurisdictions.
- Stakeholders
- Research computing directors at each site
- Data governance or IRB officers
- National or state research funders
- Discovery questions
- Which datasets cannot leave their institution or state?
- Do partner sites run different hardware?
- What network capacity links the sites?
- Value hypothesis
- Interpretation: federated training could expand usable compute for restricted data; no performance data yet.
- Potential engagement
- Federated learning feasibility study across two or more research computing centres.
- Unsupported claims
- Parity with centralised training (not yet reported)
- Production readiness by a specific date
- Elimination of all privacy risk
Pre-sales engineering
Role takeaway
- Fit
- Multi-site research with data-residency constraints.
- Architecture and integration
- NVIDIA FLARE coordinating training across national supercomputers; local data, shared model updates.
- Prerequisites
- Inter-site connectivity
- Aligned software stacks
- Agreed model and evaluation protocol
- Constraints
- Heterogeneous hardware
- Separate schedulers and allocation policies
- Target completion end of 2026
- Security
- Secure aggregation and protection of model updates should be assessed; not described in the source.
- Proof of value
- Compare federated and single-site model accuracy
- Measure training overhead from coordination
- Test with progressively larger models
Delivery
Role takeaway
- Work
- Set up cross-site coordination, software alignment, joint testing and governance agreements.
- Dependencies
- Bilateral or consortium agreement
- Network links
- FLARE deployment at each site
- Ownership
- Each centre for its system; joint steering group for the federation.
- Skills and adoption
- Research software engineers need federated learning skills at each site.
- Governance checkpoints
- Data governance approval per dataset
- Equivalence test sign-off
- Security review of update exchange
- Acceptance criteria
- Proposed: federated model within an agreed accuracy margin of centralised baseline
- Proposed: no raw data transfer confirmed by audit
- Proposed: coordination overhead reported per run
- Risks
- Scheduling conflicts between centres
- Software drift across sites
- Model-update leakage of sensitive information
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Federated training over heterogeneous hardware (NVIDIA and non-NVIDIA) using NVIDIA FLARE; only model updates cross sites.
Governance
Who approves, reviews and stays accountable for outcomes?
Each country retains control of machines and data; cross-border collaboration rests on a bilateral partnership rather than data-sharing agreements for raw data.
Security and privacy
What data, permissions and controls need testing?
Keeps raw sensitive data in-country; the release does not address risks from shared model updates (such as inference about training data).
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Aims to widen access to larger compute pools for researchers and startups in both countries.
Procurement
What should contracts, pricing and exit terms secure?
Demonstrates interoperability across vendors via open-source software, which may reduce lock-in for federated use cases.
Operating model
Which teams own the service once it runs?
Two national operators coordinate scheduling and software stacks for joint training while retaining separate operations.
Publication history
- 2026-10-21NVIDIA · Issue 469 resources
Stable resource ID: bristol-brics-genci-federated-learning-isambard-jean-zay-adastra-20261021