Broadcom Launches Tomahawk Ultra Chip to Challenge Nvidia in AI

CIOTech Outlook Team | Wednesday, 16 July 2025, 12:45 IST

  •  No Image

  • Broadcom launches Tomahawk Ultra chip to rival Nvidia NVLink.
  • Enables 4x more chip connections using high-speed Ethernet.
  • Built for AI scale-up computing, now shipping via TSMC.

Broadcom has introduced its newest initiative in the AI infrastructure race: a new, powerful networking chip called Tomahawk Ultra, clearly targeted at taking on Nvidia in high-performance computing and AI workloads.

The chip, announced Tuesday, is intended to be a data traffic conductor in a densely populated server rack where hundreds of processors need fast communication as they train and run large AI models. The engineering teams at Broadcom developed this processor over the course of about three years. It was originally intended for high-performance computing but was adapted to work for an acute focus on generative AI workloads.

The Tomahawk Ultra is Broadcom's answer to Nvidia's NVLink Switch, presenting a good alternative with greater scalability. As Broadcom senior vice president Ram Velaga said, the new chip can connect four times as many AI chips as Nvidia's NVLink Switch. More important, it does this over a high-speed version of Ethernet instead of a proprietary protocol, allowing data center operators flexibility and compliance with industry standards.

Also Read: BRICS Leaders Urge Data Protection From Unauthorised AI Use

In AI computing environments, "scale-up" architecture--where chips in the same physical enclosure communicate with near-zero latency--is a must. Broadcom's chip is designed to enable this by providing ultra-fast interconnects between GPUs or AI accelerators and therefore helping software developers harness huge compute capability with great efficiency.

The Tomahawk Ultra is being manufactured by TSMC (Taiwan Semiconductor Manufacturing Company) using its 5-nanometer process node and has already begun shipping. The chip is set to power the next wave of AI-optimized data centers, offering a new level of performance and interoperability.

With the launch of this service, Broadcom is solidifying its role as a leading player in AI infrastructure, already partnered with AI leaders like Alphabet’s Google to develop and expand its custom AI chips. Broadcom is not only taking the fight to Nvidia but is also opening up AI technology to be more open, scalable and efficient across a wider portion of the industry.