Broadcom has officially launched its next-generation networking chip designed specifically for artificial intelligence applications, marking a significant advancement in AI infrastructure technology.
The Tomahawk 6 switch chip, which began shipping to customers over the weekend and will be broadly available in July, delivers an industry-first 102.4 terabits per second of switching capacity in a single chip. This represents double the bandwidth of any Ethernet switch currently available on the market, according to Ram Velaga, Senior Vice President and General Manager of Broadcom's Core Switching Group.
The chip's unprecedented performance makes it particularly well-suited for powering large AI clusters. With its advanced AI features dubbed Cognitive Routing 2.0, the Tomahawk 6 can detect network congestion and dynamically reroute data to other connections, helping avoid performance bottlenecks that typically plague AI workloads. This capability is crucial for AI training processes that require coordinating work across multiple graphics processing units (GPUs).
Broadcom's new chip supports both scale-up and scale-out AI network architectures. In two-tier scale-out networks, the Tomahawk 6 can link together more than 100,000 processors, with Broadcom reporting that multiple customers already plan to integrate the chip into AI clusters of this scale. The company states the chip is designed to eventually meet the demands of AI clusters with more than one million XPUs (AI accelerators).
The launch comes at a strategic time for Broadcom, which has seen explosive growth in its AI chip business. The company generated $12.2 billion in revenue from AI chips in fiscal 2024, representing a 220% increase from the previous year. Broadcom CEO Hock Tan has projected that the company's serviceable addressable market in custom AI processors and networking chips could range between $60 billion and $90 billion by fiscal 2027.
The Tomahawk 6 represents Broadcom's commitment to Ethernet technology as the networking standard for AI infrastructure, competing with Nvidia's InfiniBand technology. "All of these networks can be very simply done on Ethernet, you don't need esoteric technologies," Velaga stated, highlighting Broadcom's strategy to provide open, standards-based networking solutions for the rapidly expanding AI market.