Broadcom Unveils Jericho4: Next-Generation Ethernet Router for AI and HPC
Broadcom is expanding its footprint in the world of AI and high-performance computing with the launch of Jericho4, the newest member of its Ethernet fabric routing family. Designed to link vast numbers of accelerators across multiple data centers, Jericho4 is intended to meet the connectivity demands of today’s large-scale distributed AI workloads.
Unlike its predecessor Jericho3, which focused on intra–data center interconnects, Jericho4 is built to manage long-haul connections spanning more than 100 kilometers, providing the backbone for regional and multi-site AI deployments.
A Networking Platform for AI-Scale Growth
According to Broadcom’s technical specifications, Jericho4 offers the ability to connect over one million accelerators, or XPUs, across geographically dispersed facilities. The router integrates high-bandwidth memory (HBM) for packet buffering, enabling 160 times more buffering capacity compared to traditional on-chip memory. This level of scalability helps eliminate packet loss, even during periods of intense congestion.
The new platform works alongside Broadcom’s Tomahawk 6 and Tomahawk Ultra devices, forming what the company describes as a comprehensive portfolio for AI and HPC networking. Collectively, these technologies are designed to handle traffic within racks, between racks, and now across entire data centers.
Bandwidth, Security, and Scalability
Jericho4 delivers 3.2 Tbps HyperPorts with built-in deep buffering, native MACsec encryption at line rate, and support for RDMA over Converged Ethernet (RoCE) to ensure congestion-free performance. A single deployment can scale up to 36,000 HyperPorts, giving hyperscalers and enterprises a tool to build out massive GPU or accelerator farms without bottlenecks.
Built on TSMC’s 3-nanometer process technology, Jericho4 increases transistor density, allowing for more ports, integrated memory, and improved power efficiency. These advances are particularly important for distributed AI training workloads that demand secure, high-throughput communication between GPUs and storage systems.
Analysts See Strategic Impact
Industry analysts highlight Jericho4’s potential to redefine data center design strategies. Omdia’s chief analyst, Lian Jie Su, notes that HBM integration not only raises performance but also lowers I/O power consumption, enabling faster data processing compared to conventional buffering. While the use of HBM can increase overall costs, Su argues that the efficiency gains could offset expenses by maximizing utilization across geographically distributed facilities.
Other analysts emphasize Jericho4’s role in long-haul interconnects. By enabling compute and storage to be separated over city-scale distances, cloud providers could design more resilient and flexible regional AI zones, reducing dependency on a single mega data center.
TechInsights’ Manish Rawat views Jericho4 as a step toward so-called smart switches. By embedding HBM for localized congestion management and incorporating strong on-chip encryption, Jericho4 reduces the need for complex external signaling and enhances security. This combination positions the router not just as a switching solution, but as a cornerstone for next-generation AI fabrics, including hyper-mesh architectures and GPU superclusters.
Broader Implications for Cloud and AI
Broadcom’s influence in data center networking has already been established through its Tomahawk line. With the addition of Jericho4, the company strengthens its ability to serve hyperscale customers such as AWS, Microsoft Azure, and Google Cloud, which are building ever-larger AI training and inference clusters.
The introduction of Jericho4 means these providers can distribute AI workloads more evenly across multiple locations rather than concentrating them in a single facility. This shift opens the door to city- and region-wide AI clusters, offering greater resilience, security, and efficiency.
A Key Piece of Broadcom’s AI Puzzle
Jericho4’s arrival demonstrates Broadcom’s strategy to deliver an end-to-end networking solution tailored for the age of AI. With its combination of lossless performance, security-first design, and long-haul scalability, the new Ethernet router provides the critical infrastructure needed to keep pace with the explosive growth of distributed computing.
By pairing Jericho4 with its Tomahawk Ultra and Tomahawk 6 chips, Broadcom is positioning itself as a leading enabler of the next wave of hyperscale AI infrastructure—where data centers are no longer just standalone facilities, but interconnected hubs working together across vast distances.