Broadcom's Jericho4 Chip: Revolutionizing AI Infrastructure with Distributed Data Center Connectivity

Broadcom targets distributed AI training with Jericho4 ASIC

Forget building massive super clusters. Cobble them together from existing datacenters instead Broadcom on Monday unveiled a new switch which could allow AI model developers to train models on GPUs spread across multiple datacenters up to 100 kilometers apart. The switch could help pave the way for an alternative to the massive facilities currently being built to power the AI boom, allowing companies to stitch together distant and less power-hungry datacenters. Codenamed Jericho4, Broadcom claims 51.2 Tb/s of aggregate bandwidth across the ASIC's switch and fabric ports. But while the chip can serve double duty as a scale-out or scale-up network switch, Broadcom has far higher radix and lower latency options with its Tomahawk 6 or Ultra accelerators. Instead, Broadcom has positioned the chip for datacenter-to-datacenter interconnect (DCI). "If you're running a training cluster and you want to grow beyond the capacity of a single building, we're the only valid solution out there," Amir Sheffer, an associate product line manager at Broadcom told El Reg. Each Jericho can be configured with up to eight of what Broadcom calls "hyper ports" - four 800GbE links that behave like one big 3.2Tb/s port. Compared to simply using ECMP link aggregation to bind a bunch of 800GbE ports together, Broadcom says its hyper ports can achieve 70 percent higher link utilization. The silicon-slinger said users can scale Jericho4 into configurations of up to 36,000 hyper ports, which should be enough to connect two datacenters at a blistering 115.2 petabits per second. That's enough bandwidth to connect 144,000 GPUs, each at 800Gbps, to an equal number in a neighboring datacenter without running into bottlenecks. Historically, datacenter operators have employed some degree of over-subscription in their DCI deployments, whether it be 4:1 or 8:1, and this is likely to continue to be the case, Sheffer said. As great as alleviating the power constraints associated with large scale AI training workloads by distributing those workloads across multiple datacenters might sound, bandwidth isn't the only factor. Latency also comes into play. While Jericho4's deep HBM-backed buffers and congestion management tech can help with tail latency caused by packet loss, they can't change the fact that light only travels so quickly through glass fiber. Over a 100-kilometer span the round trip latency works out to a nearly one millisecond - and that's before you take into consideration the latency incurred by the transceivers and protocol overheads. However, progress is being made to mitigate the impact of latency on distributed training workloads. Back in late January, Google's DeepMind team published a paper titled "Streaming DiLoCo with overlapping communication," in which the web giant detailed an approach to low-communication training. The basic idea was to create distributed work groups that don't have to talk to one another all that often. By using quantization, and strategically scheduling communication between the datacenters, researchers suggest, many of the bandwidth and latency challenges can be overcome. Jericho4, which is currently available for large customers to sample so they can start designing appliances, arrives as hyperscalers and cloud providers break ground on massive multi-gigawatt datacenter campuses. These clusters are so large that in many cases they require new power plants to be built to support them. Meta, for instance, contracted Entergy to construct three combined-cycle combustion turbine generators totaling 2.2 gigawatts of capacity to fuel its Richland Parish megacluster. With Jericho4, Broadcom has presented an alternative. Rather than build one great big datacenter campus, AI outfits could build multiple smaller datacenters and pool their resources. ®

Broadcom Chip to Help Power AI by Linking Up Smaller Data Centers

The age of ever-bigger AI data centers is leaving cloud companies with a question: What do with older, smaller facilities? Broadcom Inc. is offering a possible answer, at least for operations located with some proximity. A new version of its Jericho networking chip announced Monday can transfer larger volumes of data at higher speeds. That means customers can link up several smaller data centers to create one big system for developing or running AI models, said Ram Velaga, senior vice president and general manager of the company's core switching group.

Broadcom launches Jericho chip to advance AI data center networks

SAN FRANCISCO, Aug 4 (Reuters) - Broadcom's silicon division launched its next-generation Jericho networking chip on Monday, which is designed to connect data centers over 60 miles (96.5 km) apart and speed artificial intelligence computation. The company's Jericho4 introduces and improves several features that increase the amount of networking traffic speeding across large networks that operate inside and between data centers. Building and deploying artificial intelligence has become more computationally intensive and requires stringing together thousands of graphics processors (GPUs). Cloud computing companies such as Microsoft (MSFT.O), opens new tab and Amazon (AMZN.O), opens new tab require faster, more sophisticated networking chips to ensure data moves efficiently. Security when transferring data beyond the physical walls of a data center is crucial for cloud companies because of the potential attacks that could intercept it ahead of reaching its destination. Broadcom's engineers designed the Jericho chips to be deployed at a massive scale, and a single system can encompass roughly 4,500 chips, according to Ram Velaga, senior vice president and general manager of Broadcom's Core Switching Group. To help mitigate issues around network congestion, the Jericho4 chips use the same high-bandwidth memory (HBM) designers such as Nvidia (NVDA.O), opens new tab and AMD (AMD.O), opens new tab use for their AI processors. It's necessary because of the volume of data that needs to be stuffed into memory at any given moment of operation. "The switch is actually holding that traffic (in memory) till the congestion frees up," Velaga said. "It means you need to have a lot of memory on the chip." The longer the distance the data must travel from the chip to its destination, the more memory designers must include in the chip as well. In addition to performance improvements, the Jericho4 also beefs up security by encrypting data. Broadcom opted to use TSMC's (2330.TW), opens new tab three nanometer process for the Jericho4. Reporting by Max A. Cherney; editing by Diane Craft Our Standards: The Thomson Reuters Trust Principles., opens new tab * Suggested Topics: * Artificial Intelligence Max A. Cherney Thomson Reuters Max A. Cherney is a correspondent for Reuters based in San Francisco, where he reports on the semiconductor industry and artificial intelligence. He joined Reuters in 2023 and has previously worked for Barron's magazine and its sister publication, MarketWatch. Cherney graduated from Trent University with a degree in history.

Broadcom unveils Jericho4 at OCP Taipei, pushes Ethernet as backbone of next-gen AI infrastructure

Broadcom has unveiled Jericho4, its latest AI fabric router chip designed to connect geographically dispersed data centers, reinforcing the company's belief that Ethernet -- not proprietary interconnects -- will power the next wave of large-scale AI and machine learning systems. The announcement was made during the 2025 OCP APAC Summit in Taipei, where Ram Velaga, Broadcom's Senior Vice President and General Manager, outlined a three-tiered approach to scaling AI infrastructure -- from intra-rack communication to global data center clusters. "AI is still at the one-percentile mark of what's possible," Velaga said on stage, arguing that Ethernet's ubiquity, cost-efficiency, and openness make it the best choice for scaling AI workloads across every layer of the stack. "The best way to build a large distributed computing system is to do it on Ethernet." At the heart of the Taipei announcement was Jericho4, a 3nm multi-die AI fabric router chip designed specifically for inter-data center connectivity. It supports line-rate, full-speed encryption, very deep buffering, and high-bandwidth memory (HBM) to facilitate the movement of AI workloads between facilities separated by more than 100 kilometers. As AI model sizes grow, Broadcom sees a need to link multiple 50-60 MW data centers to function as unified compute clusters. Jericho4 fills that gap -- marking a departure from the company's Tomahawk product line, which focuses on intra-data center switching. Scaling AI with Ethernet: Rack to region Velaga's presentation also reiterated Broadcom's broader Ethernet strategy for AI: Within racks (Scale-Up Ethernet): Broadcom's Scale-Up Ethernet (SUE) specification enables low-latency communication between XPUs (GPUs, TPUs, etc.) and HBM within a rack or a few racks. The company's Tomahawk Ultra switch achieves under 400 nanoseconds of XPU-to-XPU latency, with about 250 nanoseconds inside the switch, helping scale up domains to hundreds or even thousands of XPUs. Across data centers (Scale-Out Ethernet): For larger clusters within a single data center, Broadcom introduced the Tomahawk 6, delivering 100 terabits per second of bandwidth. Velaga said it can reduce optical components by 67%, cut power usage, and lower network complexity for installations like 128,000-GPU clusters. The company's pitch centers on Ethernet as a scalable, vendor-neutral alternative to proprietary AI interconnects. Broadcom continues to invest in open standards, contributing the SUE specification to the Open Compute Project (OCP) community to foster multi-vendor innovation across hardware and software layers.

Broadcom launches Jericho chip to advance AI data center networks - The Economic Times

Building and deploying artificial intelligence has become more computationally intensive and requires stringing together thousands of graphics processors (GPUs). Cloud computing companies such as Microsoft and Amazon require faster, more sophisticated networking chips to ensure data moves efficiently.Broadcom's silicon division launched its next-generation Jericho networking chip on Monday, which is designed to connect data centers over 60 miles (96.5 km) apart and speed artificial intelligence computation. The company's Jericho4 introduces and improves several features that increase the amount of networking traffic speeding across large networks that operate inside and between data centers. Building and deploying artificial intelligence has become more computationally intensive and requires stringing together thousands of graphics processors (GPUs). Cloud computing companies such as Microsoft and Amazon require faster, more sophisticated networking chips to ensure data moves efficiently. Security when transferring data beyond the physical walls of a data center is crucial for cloud companies because of the potential attacks that could intercept it ahead of reaching its destination. Broadcom's engineers designed the Jericho chips to be deployed at a massive scale, and a single system can encompass roughly 4,500 chips, according to Ram Velaga, senior vice president and general manager of Broadcom's Core Switching Group. To help mitigate issues around network congestion, the Jericho4 chips use the same high-bandwidth memory (HBM) designers such as Nvidia and AMD use for their AI processors. It's necessary because of the volume of data that needs to be stuffed into memory at any given moment of operation. "The switch is actually holding that traffic (in memory) till the congestion frees up," Velaga said. "It means you need to have a lot of memory on the chip." The longer the distance the data must travel from the chip to its destination, the more memory designers must include in the chip as well. In addition to performance improvements, the Jericho4 also beefs up security by encrypting data. Broadcom opted to use TSMC's three nanometer process for the Jericho4.