Broadcom is now shipping switch chips specially designed to speed data transmission for AI workloads. These AI boosting data center switch chips, the Tomahawk 6 series, are yet another move by Broadcom to distinguish itself as a primary player in the burgeoning enterprise AI sector.

Switches are an unsung but vital aspect of data center networks that enable servers to rapidly transmit data back and forth—and this Tomahawk version is especially fast. Broadcom says a single Tomahawk 6 chip is capable of performing the signal tasks that once required 6 of this product line’s earlier version.

Also boosting processor capacity: the Tomahawk 6 is the first chip in the Tomahawk line to combine several chips into a single package, a design that doubles the silicon surface used in the chip. Broadcom is fabricating the chip on Taiwan Semiconductor Manufacturing Co’s three nanometer process.

The company touts the Tomahawk 6 as delivering the first ever switch chip with 102.4 Terabits/sec of switching capacity in a single chip, which is fully double the bandwidth of any Ethernet switch sold today. This new version of Tomahawk supports 100G/200G SerDes and—for systems that require it—co-packaged optics (CPO). The CPO feature resolves bottlenecks in bandwidth and signal integrity, and also consumes less power than traditional pluggable optics. Most important, CPO offers a technology that scales past 100 Tbps switches, which lays the groundwork for the optical I/O fabrics that are essential for AI workloads.

And AI is the point here. Broadcom leaves no doubt about its focus on AI, promoting the Tomahawk as built to with “AI-optimized features” that will power “the next generation of scale-up and scale-out AI networks.” Pressing the point still further, the Tomahawk’s supports 100G/200G SerDes and includes “the industry’s most comprehensive set of AI routing features and interconnect options, designed to meet the demands of AI clusters” with huge numbers of GPUs.

“Tomahawk 6 is not just an upgrade – it’s a breakthrough,” said Ram Velaga, senior vice president and general manager, Core Switching Group, Broadcom, claiming that demand from customers and partners has been “unprecedented.” Wall Street seemed to agree that Tomahawk is big news, with Broadcom’s stock closing up 3.27% on the day, hitting a 52-week high.

The massive scaling that the Tomahawk 6 enables is indeed impressive. The chip can handle the data transmission requirements of massive data facilities that are evolving to hold in the range of 100,000 to 1 million GPUs. Broadcom claims that multiple deployments are already on the drawing board with the Tomahawk 6, including more than 100,000 XPUs for both the scale-out and scale-up interconnect.

Broadcom, while not one of the so-called “Magnificent Seven” tech companies that have played the highest profile role in AI’s rise, has clearly played an ever-increasing role in AI in recent years. The company’s market cap hit $1 trillion in late 2024 due to exponential growth in AI revenues and profits from its VMware acquisition. Industry observers see Broadcom as emerging into the top tier of AI vendors based on its line of AI-capable semiconductors; particularly noteworthy are the custom AI chips it built in partnership with Google.

Also powering Broadcom’s emergence as a top player is its partnership with NVIDIA. Among the various complementary roles, Broadcom’s Tomahawk and Jericho series switches are used to connect thousands of NVIDIA GPUs in AI clusters.