Broadcom Launches New Tomahawk Ultra Networking Chip

Broadcom has unveiled a new networking processor aimed at accelerating artificial intelligence (AI) data processing, which requires linking together hundreds of chips that must operate in close coordination, Reuters reported via Max Cherney.

The new chip is called the Tomahawk Ultra. I Image: Broadcom

The new chip, called the Tomahawk Ultra, is the latest hardware release in Broadcom’s ongoing competition with AI heavyweight Nvidia.

Broadcom helps Google build its in-house AI chips, which are considered by many developers and industry experts as one of the few serious alternatives to Nvidia’s dominant graphics processing units (GPUs).

Designed to act as a high-speed traffic controller for data traveling between dozens or even hundreds of chips in a single server rack, the Tomahawk Ultra directly competes with Nvidia’s NVLink Switch chip.

According to Broadcom Senior Vice President Ram Velaga, the Tomahawk Ultra can link together four times as many chips as Nvidia’s offering.

Unlike Nvidia’s proprietary protocol, Broadcom’s chip uses a high-speed version of Ethernet. Both companies’ technologies are central to what’s known as "scale-up" computing—interconnecting chips placed physically close together to maximize performance.

This proximity is essential for delivering the computing power that advanced AI workloads demand.

Broadcom’s new chip will be manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC) using its five-nanometer process.

The processor is already shipping. Velaga said it took about three years to develop, initially for high-performance computing but later adapted for AI applications as demand surged with the rise of generative AI.