The proposed Networking for AI Research Group (NET4AI) brings together researchers and the Internet community to investigate the networking foundations of distributed AI systems spanning multiple stakeholders. It provides a forum for long-term architectural research beyond the immediate requirements addressed by IETF working groups.
NET4AI explores Internet support for autonomous agents, decentralized AI systems, and communication architectures for AI ecosystems. Its scope extends beyond individual datacenters to the challenges of operating AI workloads across networks and administrative domains.
Generative and agentic AI have prompted new protocol discussions in the IETF and growing research interest in networking for training, inference, and large language model serving. NET4AI complements these activities by examining longer-term questions about Internet architecture, decentralization, and support for networked AI systems.
NET4AI will assess and develop experimental tools, including simulators and testbeds, and collect and share datasets for reproducible experiments. The group will encourage research on networked AI systems, new architectures such as key-value-cache-centric and mixture-of-experts systems, inter-datacenter communication, and security. It will also explore principled approaches to communication among agents.
The proposed charter and current group information are available on the IETF Datatracker.
Membership in the NET4AI is open to all interested parties.
NET4AI is chaired by Antoine Fressancourt and Dirk Kutscher.
The NET4AI mailing list is net4ai@irtf.org. To subscribe or access the list archives, visit the mailman page.
Documents and meeting materials for the NET4AI can be found on the IETF datatracker.