Broadcom has begun shipping its Tomahawk 6 switch series, a networking chip designed to support large AI clusters.

The offering is the first to deliver 102.4Tbps of switching capacity in a single chip, double the bandwidth of any Ethernet switch currently available, the vendor claims.

Built to “power the next generation of scale-up and scale-out AI networks,” according to Broadcom, the chip delivers “full system-level power efficiency and cost savings.” This includes offering 200G SerDes (serializer/deserializer – technology which is used to reduce the number of inputs and outputs on an ASIC or FPGA), or the option for 1,024 100G SerDes, allowing customers to deploy AI clusters with extended copper reach.

The Tomahawk 6 will also be available with co-packaged optics, reducing power and latency whilst improving long-term reliability, the company said.

Meanwhile, the use of AI features, dubbed cognitive routing 2.0, allows the chip to optimize network speeds by re-routing data to other connections when it detects network congestion.

Broadcom says this will enable the chip to meet the networking demands of clusters comprising up to one million XPUs, adding that multiple deployments with more than 100,000 XPUs using Tomahawk 6 for both the scale-out and scale-up interconnect are already being planned.

A single switch will be able to support up to 512 ports at 200Gbps.

Fully compatible with any NIC or XPU Ethernet endpoints, the Tomahawk 6 is also Ultra Ethernet Consortium compliant.

“Tomahawk 6 is not just an upgrade – it’s a breakthrough,” said Ram Velaga, SVP and GM in the core switching group at Broadcom. “It marks a turning point in AI infrastructure design, combining the highest bandwidth, power efficiency, and adaptive routing features for scale-up and scale-out networks into one platform. Demand from customers and partners has been unprecedented. Tomahawk 6 is poised to make a rapid and dramatic impact on the deployment of large AI clusters.”