2026-11-16 –, Saal B
Inference is quietly overtaking training as the dominant AI workload, and unlike training — concentrated in a handful of hyperscale halls — inference wants to sit close to users, pulling AI capacity into distributed sites across Europe.
The industry has spent the last two years re-engineering Ethernet (via the Ultra Ethernet Consortium and platforms like NVIDIA Spectrum-X) to match InfiniBand's performance for exactly this kind of workload, and now even ships Ethernet fabrics purpose-built to stitch multiple physical data centers into one logical AI factory.
That's a solved problem when one company owns every site. This talk asks what happens when it isn't — when distributed inference has to span operators, jurisdictions, and ownership boundaries — and what a federated, policy-aware Ethernet fabric for that world would need to look like.
Ethernet is rapidly becoming the preferred back-end network for AI infrastructure. Technologies such as RoCE, Ultra Ethernet, NVIDIA Spectrum-X, and emerging “scale-across” architectures promise predictable, high-performance connectivity within and between AI data centers.
These architectures generally make one important assumption: a single organization controls the servers, switches, telemetry, congestion policies, and inter-site connectivity. Under that model, several physical data centers can be engineered to behave as one logical AI environment.
Distributed European AI infrastructure may not have that luxury. Inference capacity will increasingly be deployed across hyperscalers, regional clouds, colocation facilities, metro-edge sites, and specialist GPU providers. The networks connecting those sites may also belong to different fiber and connectivity operators. Each participant controls only part of the infrastructure, exposes limited operational information, and may operate under different commercial and jurisdictional requirements.
This creates a question that faster Ethernet alone cannot answer: what happens when an AI fabric reaches an operator boundary?
The talk begins by separating two ideas that are frequently conflated. Distributed AI services can steer requests between independent sites without extending a tightly coupled GPU fabric between them. Some workloads, however, including model distribution, distributed inference, resilience between GPU sites, and certain forms of model parallelism, place much stronger demands on inter-site connectivity. We will examine which workloads genuinely benefit from network-level federation and which should remain distributed at the application layer.
We will then look at what changes when traffic crosses a network-to-network interface. Can congestion-control assumptions survive between independently operated domains? What latency, loss, and bandwidth guarantees are technically credible? How can operators exchange useful telemetry without exposing internal topology? Who selects the end-to-end path, who responds to failures, and how can customers verify that the delivered service matches what was promised?
Sovereignty adds another dimension to this problem. A sovereign AI service is not defined solely by the location of its GPUs. Customers may also care about where their traffic travels, who owns and operates the infrastructure, which jurisdictions can exercise control over it, where encryption keys are held, and whether those claims can be independently verified. Consequently, path selection may need to consider geography, jurisdiction, operator identity, and operational control alongside conventional performance metrics.
The talk proposes a candidate architecture based on coordinated path computation with domain-local enforcement. Each operator retains control of its own network while exposing an abstract set of capabilities such as latency class, available bandwidth, path diversity, geographic scope, and jurisdictional attributes to a higher-level coordinator. The coordinator requests an outcome rather than configuring another operator’s network, while end-to-end measurement verifies whether the resulting service meets its performance and sovereignty objectives.
We will test this architecture against the timing and congestion requirements of GPU networking and identify where conventional carrier federation begins to fail. The goal is not to claim that a continent-wide lossless Ethernet fabric is immediately practical. It is to determine which multi-operator AI connectivity services are feasible, which workloads they can support, and which technical assumptions still require new standards or experimentation.
Attendees will leave with:
- A clear distinction between distributed inference services and distributed GPU fabrics.
- An understanding of what breaks when AI traffic crosses an operator boundary.
- A practical definition of sovereignty covering compute, connectivity, control, and verification.
- A candidate model for policy-aware, multi-operator AI connectivity.
- A framework for identifying which AI workloads can realistically operate across federated networks.
The commercial opportunity for operators may not be “one enormous Ethernet fabric.” It may instead be something more achievable: verifiable, policy-aware connectivity between AI-capable sites, with measurable performance, geographic, jurisdictional, and operational guarantees.
Leading MaiaEdge's expansion into Europe.
I spent over a decade as Head of Network Engineering / CTO at QSC AG and Plusnet GmbH.
QSC was one of the first alternative DSL network operators in Germany, and we helped build the independent broadband infrastructure that broke Deutsche Telekom's monopoly on German internet access.
At that time, I also Co-authored the L2-BSA broadband standards with the national Next-Generation-Access Forum.
I went on to become CDO and later CEO of ecotel communication ag, and spent roughly a decade active in telecom regulatory work, helping shape the German market both nationally and at EU level.
At MaiaEdge, I work closely with Europe's Datacenter operators, fiber operators, carriers, and Internet Exchange Points on federated, sovereignty-aware connectivity.