Broadcom (AVGO, Financial), a leading chip manufacturer, has unveiled its latest network processor, the "Tomahawk Ultra," aimed at enhancing data transmission efficiency for artificial intelligence (AI) applications. This high-performance processor is capable of linking hundreds of chips for coordinated operation, positioning Broadcom to challenge AI chip market leader NVIDIA (NVDA).
According to Broadcom, the Tomahawk Ultra can manage high-speed data transfer between tens to hundreds of chips in data centers, serving as a central traffic control hub. The chip is designed to connect four times as many chips as NVIDIA's NVLink Switch, employing an enhanced Ethernet protocol instead of proprietary protocols for data transmission.
Manufactured using TSMC's (2330) 5-nanometer process, the Tomahawk Ultra has begun shipping. Originally tailored for the high-performance computing market, it has been repositioned to meet the growing computational needs of AI enterprises due to the rapid rise of generative AI. Broadcom has been a long-term partner of Alphabet's (GOOGL) Google in producing AI-specific chips, making it one of the few capable suppliers to compete with NVIDIA's GPUs.
