Broadcom Ships Next-Generation AI Chip to Revolutionize Data Center Connectivity
Broadcom has started shipping its latest networking chip, Jericho4, designed specifically to boost performance in hyperscale AI GPU clusters. This chip dramatically accelerates AI model training efficiency by enhancing latency and bandwidth for data center connectivity. Engineered to interconnect over one million processors, including GPUs and CPUs distributed across multiple data centers up to 60 miles apart, Jericho4 enables scalable, distributed AI infrastructure without performance loss.
Jericho4 Delivers Unmatched Performance and Scalability for AI Infrastructure
The Jericho4 chip supports 51.2 terabits per second of lossless Ethernet with deep buffering and smart congestion management, ensuring stable, lossless data transport across geographically dispersed data centers. Manufactured using TSMC’s advanced 3-nanometer process, it provides enhanced power efficiency with a 40% reduction in power per bit, lowering operational costs for hyperscalers like Microsoft and Amazon as they expand AI infrastructure. Its design eliminates the need for additional retimers, further reducing power use and improving system reliability.
Designed to meet the demands of growing AI workloads, Jericho4 integrates extensive security features, including data encryption, critical for the protection of sensitive information transmitted between data centers. The chip’s ability to link smaller, distributed data centers supports a trend toward localized infrastructure that reduces latency and improves efficiency by bringing computing power closer to users.
Jericho4 Joins Broadcom’s Portfolio of High-Performance Networking Solutions
This new chip complements Broadcom’s existing Tomahawk 6 and Tomahawk Ultra switch chips, completing a robust portfolio tailored for AI and high-performance computing needs. While Tomahawk 6 focuses on extremely high bandwidth offering up to 102.4 terabits per second, Jericho4 is optimized for scalable distributed AI networks over regional distances, supporting massive AI workloads and future-proofing data center architectures against growing demand.
Broadcom’s stock has risen in response to this announcement, reflecting strong investor confidence in the company’s leadership in AI infrastructure technology.
In summary, the Jericho4 chip from Broadcom sets a new standard for AI data center networking by improving speed, scalability, power efficiency, and security, empowering hyperscale AI GPU clusters to train larger models faster across distributed environments while reducing costs and infrastructure constraints.