Broadcom Tomahawk 6 Ushers in the Terabit Ethernet Era for AI Data Centers
Broadcom aims to dominate next-gen AI networks with 102.4 Tbps switch, 1M+ XPU scalability, and co-packaged optics.
4 Min Read June 4, 2025
Broadcom unveils Tomahawk 6 with 102.4 Tbps bandwidth, CPO support, and AI-optimized routing. Learn how it’s shaping the future of AI networks.
Broadcom has just set a new benchmark in data center networking with the launch of its Tomahawk® 6 switch series, the world’s first single-chip Ethernet switch delivering a staggering 102.4 Tbps of bandwidth. Doubling the switching capacity of existing solutions, Tomahawk 6 is clearly engineered for the future of AI-scale data infrastructure.
As workloads for AI training and inference balloon in complexity and scale, so too must the networks behind them. The Tomahawk 6 aims to meet these demands with industry-first support for 100G/200G SerDes, co-packaged optics (CPO), and a comprehensive set of AI-native routing capabilities.
Optimized for AI: Cognitive Routing at Massive Scale
The Cognitive Routing 2.0 engine embedded in Tomahawk 6 is designed for real-time load balancing and congestion mitigation in hyperscale AI environments, supporting up to one million XPUs across scale-up and scale-out clusters. This opens the door for highly adaptable, fungible XPU fabrics, allowing cloud operators to dynamically reconfigure AI compute resources as needed.
Co-Packaged Optics (CPO): A Power and Efficiency Play
Broadcom isn’t just scaling performance, it’s reinventing how data centers manage power and connectivity. Tomahawk 6 offers the industry’s first implementation of 1,024 100G SerDes on a single chip, providing copper connectivity at extended reach. For optical systems, the chip’s CPO variant significantly cuts power consumption and latency, reducing total cost of ownership (TCO) and improving long-term reliability.
Built for the AI Ethernet Era
Tomahawk 6 isn’t an isolated chip, it’s a pillar in Broadcom’s larger end-to-end Ethernet AI platform, working in tandem with Jericho switches, Thor NICs, and Sian optical DSPs. It’s Ultra Ethernet Consortium-compliant, supporting modern congestion signaling and flexible topologies such as Clos, torus, and scale-up configurations.
Why It Matters
With the release of Tomahawk 6, Broadcom is no longer just competing, it's redefining the Ethernet standard for AI networking. As hyperscalers demand unified fabrics and lower power envelopes, Broadcom is positioning itself at the center of this transition. And with NVIDIA’s photonic Spectrum ASICs on the horizon, the race for the AI data center backend is officially on.
Want deeper insight into Broadcom’s Tomahawk 6 architecture and its competitive positioning in AI infrastructure?
Read the full TechInsights Platform blog — free to access now.
Already registered to the TechInsights Platform? Sign in or Request Access to unlock in-depth analysis, architecture comparisons, and strategic forecasting for AI-centric semiconductor innovation.
