Broadcom Ships Tomahawk 6: World’s First 102.4 Tbps Switch

Core Insights - Broadcom Inc. has launched the Tomahawk 6 switch series, which features the world's first 102.4 Terabits/sec switching capacity in a single chip, doubling the bandwidth of existing Ethernet switches [1][2] - The Tomahawk 6 is designed to support the next generation of scale-up and scale-out AI networks, offering energy efficiency and AI-optimized features [1][2] - The demand for Tomahawk 6 from customers and partners has been unprecedented, indicating a significant market opportunity for Broadcom [2] Product Features - Tomahawk 6 includes support for 100G/200G SerDes and co-packaged optics (CPO), providing flexibility and efficiency for AI clusters with over one million XPUs [1][3] - The switch offers advanced routing features, including Cognitive Routing 2.0, which enables global load balancing and adaptive flow control tailored for modern AI workloads [5][11] - It supports both scale-up and scale-out networking, allowing operators to utilize a unified technology stack across their AI infrastructure [6][11] Market Impact - The introduction of Tomahawk 6 is expected to alleviate network bottlenecks as AI clusters scale from tens to thousands of accelerators, providing a clear path for hyperscalers to enhance their AI infrastructure [2][7] - Multiple deployments are planned with over 100,000 XPUs using Tomahawk 6, highlighting its significance in the AI networking landscape [7] - Broadcom's commitment to open standards through the Scale Up Ethernet (SUE) Framework aims to foster an open ecosystem for AI infrastructure [9][10] Industry Collaboration - Broadcom is collaborating with various industry leaders, including AMD and Arista Networks, to enhance AI infrastructure capabilities using Tomahawk 6 [18][19] - The switch is compliant with Ultra Ethernet Consortium specifications, ensuring compatibility with modern AI transports and network topologies [11][22] - Industry analysts recognize Tomahawk 6 as a game changer for hyperscale data centers, enabling the development of scalable and efficient AI networking solutions [23][24]