San Francisco – Semiconductor giant Broadcom Inc. has started shipping its latest high-performance networking chip, the Tomahawk 6, which is specifically designed to meet the growing demands of artificial intelligence (AI) infrastructure. The announcement was made on Tuesday, marking a significant advancement in networking hardware aimed at accelerating the performance of next-generation AI data centers.
The Tomahawk 6 represents a major leap forward, offering twice the performance of its predecessor, according to Broadcom. The chip also features advanced traffic control capabilities that enhance network efficiency across complex data center environments.
“This speed boost means that fewer networking switches are needed to perform the same task,” said Ram Velaga, Senior Vice President and General Manager of Broadcom’s Core Switching Group. This efficiency is particularly critical as AI workloads require robust, high-speed interconnectivity between thousands of processors.
As AI applications scale at an unprecedented rate, data center operators are being challenged to link hundreds of thousands of graphics processing units (GPUs) to train and deploy large models. Specialized networking chips like those in Broadcom’s Tomahawk series are integral to making such architectures possible.
With Tomahawk 6, Broadcom engineers have designed a solution that can support massive data center clusters consisting of more than 100,000 GPUs, according to Velaga. “In a couple of years, you will start to see a million GPUs housed inside a physical building,” he noted, highlighting the scale of future AI infrastructure.
The chip uses Ethernet, a time-tested and widely adopted networking standard. This contrasts with Nvidia’s preference for InfiniBand, a competing technology used in some of its high-performance computing solutions. Velaga emphasized Ethernet’s versatility and practicality: “All of these networks can be very simply done on Ethernet, you don’t need esoteric technologies.”
Notably, the Tomahawk 6 marks the first product in the series to employ a chiplet architecture, where multiple chips are combined into a single package. This design approach—already adopted by industry players like AMD—has allowed Broadcom to significantly expand the chip’s silicon area, effectively doubling the capacity compared to previous versions.
The new chip is being manufactured using Taiwan Semiconductor Manufacturing Co.’s (TSMC) advanced 3-nanometer process, ensuring leading-edge performance and power efficiency.
As demand for AI continues to surge across sectors, the launch of Tomahawk 6 underscores Broadcom’s growing role in shaping the infrastructure backbone of the AI era. The chip is expected to play a crucial role in enabling scalable, high-bandwidth data centers that can meet the compute-intensive demands of modern machine learning applications.
