
Broadcom has launched its latest networking marvel, the Jericho4 chip, designed to supercharge data transfer across sprawling AI-driven data centers—even those located over 60 miles apart. With AI workloads growing ever more demanding, the new chip is built for scale, capable of powering systems with up to 4,500 interconnected chips. Featuring high-bandwidth memory (HBM) to reduce congestion and TSMC’s cutting-edge 3nm process, Jericho4 offers both speed and security with advanced encryption for data moving outside the physical confines of cloud facilities.
As tech giants like Amazon and Microsoft race to scale their AI infrastructure, Broadcom’s Jericho4 offers a crucial backbone for seamless GPU communication and massive data throughput. With memory capacity tailored to buffer traffic during network congestion, the chip helps ensure uninterrupted AI operations across vast networks. In short: Jericho4 is the invisible muscle powering tomorrow’s intelligent cloud—and it’s built to go the distance.

Broadcom’s 3.2T HyperPort technology consolidates four 800GE links into a single logical port — eliminating load balancing inefficiencies, boosting utilization by up to 70%, and streamlining traffic flow across large fabrics.
Thanks to deep buffering and intelligent congestion control, Jericho4 ensures lossless RoCE across 100km+ enabling truly distributed AI infrastructure unconstrained by power and space limitations at a single location.
Jericho4 supports MACsec encryption on every port at full speed to protect data moving between data centers, delivering strong security without compromising performance — even at the highest traffic loads.
Manufactured on a 3nm process, Jericho4 features Broadcom’s advanced 200G PAM4 SerDes with industry-leading reach. This eliminates the need for extra components like retimers, resulting in lower power usage, reduced cost, and higher system reliability.
Jericho4 is fully compliant with specifications developed by the Ultra Ethernet Consortium (UEC), ensuring interoperability across open, standards-based Ethernet AI fabrics. This allows seamless integration with a broad ecosystem of UEC-compliant NICs, switches, and software stacks