The Jericho3-AI Ethernet switch from Broadcom Inc. - 800G ports push AI clusters harder

Reviewed: ad hoc news B2B & Pro desk. Edited and checked on 2026-06-27, 17:21. Details in the imprint.

The Jericho3-AI Ethernet switch from Broadcom Inc. sits in a cold data hall, fans humming, link LEDs flickering like a quiet heartbeat in front of rows of liquid-cooled GPU racks. One rack tech brushes the metal chassis and feels barely any vibration despite the torrent of packets rushing through its 800G ports.

What Jericho3-AI is built for

Jericho3-AI is Broadcom's latest high-radix Ethernet switch silicon aimed explicitly at artificial intelligence and machine learning clusters, not generic campus networks. It is designed to connect tens of thousands of GPUs with lossless-like behavior over Ethernet fabrics, competing directly with InfiniBand-based approaches. Broadcom official product page

The chip focuses on congestion control and packet spraying techniques so that large AI training jobs see predictable tail latency, even when the network is heavily utilized. David Li, a senior product director at Broadcom, described Jericho3-AI as a "purpose-built engine for AI Ethernet fabrics" in the launch material, underlining the narrow focus of the design. Broadcom press release on Jericho3-AI

Ports, scale and power budget

At the hardware level, Jericho3-AI supports up to 800G Ethernet ports, enabling dense top-of-rack or spine switches that can feed modern accelerators such as H100 or custom ASICs without becoming the bottleneck. The silicon targets fabrics with more than 32,000 GPUs connected, pushing Ethernet deeper into territory that was once considered InfiniBand-only. The Next Platform analysis of Jericho3-AI

The chip's power envelope stays within the constraints of standard high-density switch designs, which matters when operators stack dozens of units in a single aisle. Sliding a Jericho3-AI-based switch into a rack, the installer hears the fans ramp briefly, then settle into a steady, controlled whirr once the thermal profile stabilizes under load.

Ethernet versus InfiniBand for AI

Jericho3-AI arrives in a market where high-end AI clusters are split between traditional InfiniBand and newer cloud-scale Ethernet fabrics. Broadcom positions the chip as a way to get InfiniBand-like congestion management and fairness on Ethernet, which many hyperscalers already use for everything else in their data centers.

The silicon implements advanced, fine-grained traffic steering that keeps short AI synchronization messages flowing smoothly, even alongside bulky parameter updates and data streams. For operators, this means less jitter in training times from one run to the next and more predictable utilization when they schedule jobs in already busy clusters.

How it fits into Broadcom's lineup

Jericho3-AI sits alongside Broadcom's standard Jericho3 switches and Tomahawk line, but its feature set is narrowly tuned for collective operations common in AI workloads. The company highlights that it can be paired with its own network adapters and optical modules to build end-to-end Ethernet fabrics, which simplifies qualification for large cloud operators.

In Broadcom's launch material, vice president Ram Velaga emphasized that customers asked for Ethernet-based alternatives that still delivered low tail latency and efficient job completion times on massive GPU farms. His comments underline that Jericho3-AI is not a speculative science project but a response to requests from existing hyperscale clients.

Deployment, management and who buys it

In practice, Jericho3-AI will not be bought by small enterprises but by hyperscalers, big cloud providers and large AI research institutions. These buyers typically integrate the chip into custom switch designs or work with OEM partners who produce systems tuned to their rack power and cooling profiles.

Once deployed, operators interact with Jericho3-AI primarily through the switch operating systems and fabric controllers that sit above it. They can tune parameters like queue depths, spraying policies and congestion response thresholds, often via automation, to match the behavior of their specific training workflows.

Stock context and market angle

Jericho3-AI is part of Broadcom's strategy to remain a central supplier for high-performance networking as AI spending shifts from pilot projects to large-scale infrastructure. For investors, the product underscores how much of Broadcom's growth thesis now rests on providing plumbing for compute-intensive workloads rather than consumer-facing brands.

Broadcom shares (ISIN US11135F1012) trade on NASDAQ, and on 2026-06-27 the Broadcom share price is quoted around 365.02 US dollars according to a recent market snapshot.

This article was AI-assisted and editorially reviewed. Product information without guarantee; prices and availability may change at short notice. No investment advice, no buy or sell recommendation. Stock-market transactions involve risks up to total loss.