As AI workloads exceed the power and space limits of a single data center, the challenge of interconnecting compute resources across multiple facilities is becoming critical. Broadcom has begun shipping Jericho4, its latest Ethernet fabric router designed for distributed AI infrastructure.
In this interview, Jim Carroll of Converge Digest speaks with Hasan Siraj of Broadcom’s Core Switching Group about how Jericho4 introduces the innovative 3.2 Tbps HyperPort, delivers lossless RoCE over 100 km, and integrates line-rate MACsec encryption to secure traffic between data centers.
🔹 Learn how Jericho4 fits into Broadcom’s Scale-Up Ethernet (SUE) vision alongside Tomahawk 6 and Tomahawk Ultra, and why Ultra Ethernet Consortium (UEC) compliance is key for interoperability at AI scale.
⏱️ Timestamps
• 00:00 – Introduction: AI workloads outgrowing single data centers
• 00:31 – Broadcom announces Jericho4, featuring the 3.2T HyperPort
• 01:00 – The interconnect challenges of scaling AI across hyperscale data centers
• 02:10 – What is the 3.2T HyperPort and why is it needed?
• 02:51 – Solving utilization inefficiencies of 800G links with HyperPort
• 03:55 – Handling large (“elephant”) flows and achieving 70% better link utilization
• 04:22 – Delivering lossless RoCE over 100 km: the role of deep buffers
• 05:21 – Integrated MACsec at line rate: securing distributed data center interconnects
• 06:09 – Closing thoughts on Jericho’s role in distributed AI infrastructure
👉 Tomahawk 6 link
“Watch our previous interview on Tomahawk 6—Broadcom’s first 100 Tbps Ethernet switch designed for AI and HPC—right here: https://youtu.be/bLzFU57utow” 
👉 Tomahawk Ultra link
“Also check out our conversation about Tomahawk Ultra, the ultra‑low latency 51.2 Tb/s Ethernet switch engineered for tightly coupled GPU and HPC environments: https://youtu.be/h4RQbh4931M”
Have a tech update that you want to brief us on? Contact [email protected]!
https://ngi.fyi/tech-broadcom-hasan
