Broadcom has officially started shipping its groundbreaking Tomahawk 6 switch chip series — the first Ethernet switch chip to hit 102.4 Terabits per second (Tbps). That’s double the bandwidth of any other switch on the market today. Built to fuel massive, next-gen AI networks, Tomahawk 6 brings unmatched performance, efficiency, and flexibility for hyperscale data centers.

Made for the Future of AI Networking

As AI clusters scale from hundreds to hundreds of thousands of processors, networking has become a serious bottleneck. Tomahawk 6 breaks through that barrier, supporting networks with over one million XPUs (AI accelerators). It also supports both 100G and 200G SerDes, plus co-packaged optics (CPO) — all aimed at reducing latency, boosting reliability, and simplifying system design.

“This isn’t just an upgrade — it’s a revolution,” says Broadcom’s Ram Velaga. “Tomahawk 6 will dramatically change how AI infrastructure is built.”

Flexible and Efficient Connectivity

Tomahawk 6 is loaded with features that help cloud providers build smarter, leaner AI networks:

• Industry-leading 200G SerDes for long-range copper connectivity

• Support for 1,024 100G SerDes on a single chip

• Optional co-packaged optics to reduce power, latency, and link failures

• Lowest total cost of ownership thanks to efficient hardware and Broadcom’s ecosystem

AI-Optimized Architecture

With Cognitive Routing 2.0, Tomahawk 6 intelligently handles congestion, failures, and changing traffic — perfect for complex AI workloads like reinforcement learning and large language model training. Whether you’re scaling up (inside a server) or out (across data centers), it’s built to handle it all.

A Unified Ethernet Backbone for AI

The chip supports open Ethernet standards for both scale-up and scale-out infrastructure, helping data centers avoid vendor lock-in. Broadcom is also sharing its new Scale Up Ethernet (SUE) Framework with the open-source community, pushing innovation across the board.

“With Ethernet, operators can dynamically optimize resources without relying on closed systems,” says Bloomberg analyst Kunjan Sobhani.

Built for Scale, Backed by Ecosystem

Tomahawk 6 is just one part of Broadcom’s complete AI infrastructure lineup — including Thor NICs, Jericho switches, Agera retimers, and Sian DSPs — all designed to work together in ultra-efficient, AI-optimized environments.

---

Tomahawk 6 Highlights:

• ? 102.4 Tbps switching bandwidth — in a single chip

• ? Cognitive Routing 2.0 for dynamic, AI-smart traffic control

• ⚡ Support for 100G/200G PAM4 SerDes with long-reach copper

• ? Option for co-packaged optics for low-power, low-latency setups

• ? Scales to 100,000+ XPUs in 2-tier networks

• ? Fully compliant with the Ultra Ethernet Consortium

• ? Works with any NIC or XPU endpoint — no vendor lock-in

• ? Compatible with a variety of network topologies: Clos, torus, rail-only, and more

---

Broadcom’s Tomahawk 6 isn’t just pushing the limits — it’s redefining them. The age of ultra-scalable, open, Ethernet-based AI infrastructure has officially arrived.