SNAI 2026 · Barcelona, Spain · 7–9 December 2026
SNAI 2026 features three keynote talks from leading researchers in networking, AI infrastructure, and operations. Two are confirmed; the third will be announced in August 2026.
Self-driving networks have been a stated goal for nearly a decade, and learned components now appear in production load balancers, congestion controllers, and traffic engineering systems. Yet end-to-end network autonomy remains elusive. This talk examines the gap between learned components and fully autonomous networks, drawing on recent field measurements and on the limits of current closed-loop control. Sørensen argues that the next generation of self-driving networks must be designed around explicit, verifiable intent rather than around end-to-end optimisation, and shows early evidence from industrial deployments.
Ingrid Sørensen is Professor of Computer Systems at ETH Zürich, where she leads the Networked Systems Group. Her work spans programmable networks, network verification, and the systems foundations of self-driving networks. She is a Fellow of the ACM and serves on the steering committee of several networking venues.
The fabric that connects thousands of accelerators has become as load-bearing as the accelerators themselves. This talk examines the design space of training-cluster networks at the trillion-parameter scale, drawing on operational experience from one of the largest training deployments in production. Costa walks through the tradeoffs between topology, congestion control, collective communication primitives, and failure handling, and presents a set of design principles for the next generation of training fabrics. The talk is intended for researchers and practitioners who design, deploy, or operate large-scale training infrastructure.
Rafael Costa is a Distinguished Engineer working on the networking systems that sustain large-scale AI training. He has led the design of multiple generations of training-cluster fabrics and has been a long-standing contributor to RDMA and collective communication research.
The third SNAI 2026 keynote will be announced in August 2026. Past editions have featured keynotes spanning programmable data planes, federated learning under bandwidth scarcity, and ten years of learned networking. Subscribe to the low-traffic announcement list to be notified.