Broadcom has started shipping its Jericho4 Ethernet fabric router, a purpose-built platform for distributed AI and HPC infrastructure. Designed to interconnect over one million XPUs across multiple facilities, Jericho4 addresses the scaling limits of single-site deployments by delivering high-bandwidth, secure, and lossless connectivity across distances of 100 kilometers or more. The system marks a major milestone for hyperscalers seeking to extend AI fabrics beyond the power and floor space constraints of a single campus.
A single Jericho4 system supports up to 36,000 HyperPorts, each operating at 3.2 Tbps. By consolidating four 800GE links into one logical channel, Broadcom’s HyperPort technology eliminates load-balancing inefficiencies, improves link utilization by as much as 70%, and streamlines traffic flow across massive fabrics. The platform is manufactured on a 3nm process with advanced 200G PAM4 SerDes that achieve long reach without external retimers, reducing cost and energy use while improving system reliability. Deep buffering and intelligent congestion control ensure lossless RoCE operation, while integrated line-rate MACsec encryption secures traffic across data centers without performance penalties.
Broadcom frames Jericho4 as the “scale-out” complement to its recently announced Tomahawk 6 and Tomahawk Ultra switching devices. Tomahawk 6 delivers 102.4 Tbps of bandwidth for in-rack and in-cluster connectivity, while Tomahawk Ultra supports 51.2 Tbps for broader in-data center fabrics. Together, these platforms handle scale-up networking inside racks and clusters, while Jericho4 extends the AI fabric across racks and across data centers. This combination allows hyperscalers to build open and interoperable AI infrastructures that expand from a single node to regional clusters of over one million accelerators.
Jericho4 is fully compliant with Ultra Ethernet Consortium (UEC) specifications, ensuring interoperability with UEC-compliant NICs, switches, and software stacks. With this standards-based approach, Broadcom aims to strengthen Ethernet’s role as the transport of choice for AI and HPC workloads, countering proprietary fabrics while supporting congestion-free RoCE deployments at massive scale.
• Jericho4 system: 51.2 Tbps scalable router with up to 36,000 HyperPorts
• Each HyperPort: 3.2 Tbps, consolidating four 800GE links into one logical port
• Supports 100km+ lossless RoCE transport with deep buffering and congestion management
• Integrated line-rate MACsec encryption on all ports for secure inter-data center traffic
• 200G PAM4 SerDes with long reach, eliminating need for retimers and reducing power
• Complements Tomahawk 6 (102.4 Tbps) and Tomahawk Ultra (51.2 Tbps) for end-to-end AI networking
• Fully compliant with Ultra Ethernet Consortium specifications for open interoperability
“The Jericho4 family is engineered to extend AI-scale Ethernet fabrics beyond individual data centers, supporting congestion-free RoCE and 3.2 Tbps HyperPort for unprecedented interconnect efficiency,” said Ram Velaga, senior vice president and general manager of Broadcom’s Core Switching Group.
🌐 Why it Matters: The Jericho4 launch underscores Broadcom’s strategy of building an end-to-end networking silicon portfolio tailored for AI. With Tomahawk 6, Tomahawk Ultra, and Jericho4, hyperscalers can deploy Ethernet as a unified fabric for scale-up, in-data center scale-out, and cross–data center interconnect — a critical enabler as AI workloads push beyond the limits of single-site infrastructure.
🌐 We’re tracking the latest developments in networking silicon. Follow our ongoing coverage at: https://convergedigest.com/category/semiconductors/
