Sponsor Talk

Mohan Kalkunte

Vice President of Architecture & Technology, Core Switching Group, Broadcom

Title: TBD

Abstract:

The rapid scaling of AI models is shifting system bottlenecks from compute to interconnect, making network performance a primary determinant of training efficiency and inference latency. Training stresses bandwidth density and collective communication efficiency, while large-scale inference introduces stringent latency, fanout, and concurrency requirements. These pressures demand a unified networking approach spanning scale-up, scale-out, and geo-distributed environments.

This presentation examines architectural and transport requirements for next-generation AI interconnects and presents an Ethernet-based fabric addressing these challenges. At the scale-up level, tightly coupled XPU clusters require low-latency, high-bandwidth communication with efficient support for fine-grained synchronization and memory-semantic operations. At the scale-out level, flat, high-radix topologies must deliver high bisection bandwidth while maintaining predictable performance under incast and congestion.

We highlight recent industry initiatives, including Scale-Up Ethernet Transport (SUE-T) and Ethernet Scale-Up Networking (ESUN), which extend Ethernet into traditionally specialized scale-up domains. These efforts, combined with advances in congestion management, credit-based flow control, adaptive routing, and link-layer reliability, enable deterministic performance across diverse AI workloads.

Extending beyond a single data center, we discuss requirements for geo-distributed AI fabrics, including transport efficiency, resiliency, and consistency over long-haul links.

Together, these innovations position Ethernet as a unified, open interconnect capable of scaling to million-XPU systems, offering a practical and efficient alternative to proprietary AI fabrics.

Biography:

Mohan Kalkunte is the Vice President of Architecture & Technology in the Core Switch Products group at Broadcom. He leads the architecture for Broadcom’s network switching and NIC products, including Trident, Tomahawk, Jericho, and Thor, across Enterprise, Data Center,and Service Provider markets.

With over 35 years of experience, Mohan holds a Ph.D. in Engineering from The Ohio State University. He began his career at AT&T Bell Laboratories, later working at AMD, Nortel Networks, and Maverick Networks, which was acquired by Broadcom in 1999. Mohan holds over 150 patents and was elected Broadcom Fellow in 2009, IEEE Fellow in 2013 and to National Academy of Engineering in 2025 for his contributions to Ethernet Switching. He co-authored Gigabit Ethernet: Migrating to High-Bandwidth LANs and pioneered key innovations in Ethernet switch technology.