Specifications
Switching Capacity
102.4 Tbps in a single chip — the industry's first at this capacity
Optics
Third-generation Co-Packaged Optics (CPO) — the first optically enabled switch to reach 102.4 Tbps
SerDes
224G SerDes lanes; the series is also offered in a 112G SerDes variant
Optical Line Rate
200 Gbps per channel
Bandwidth vs Previous Generation
Doubles the bandwidth of Broadcom's second-generation TH5-Bailly CPO solution — and of any CPO switch previously available
Standards Compliance
Compliant with Ultra Ethernet Consortium (UEC) specifications
AI Networking Features
Modern AI transport protocols, congestion signalling and high-resolution telemetry
Optical Interoperability
Interconnects with DR-based transceivers as well as LPO and CPO optical interconnects running at 200 Gbps per channel
Fabric Role
Supports both scale-up and scale-out AI network architectures over Ethernet
Prior Generation
Tomahawk 5 (TH5-Bailly) — 51.2 Tbps with second-generation co-packaged optics
Package Variants
Two physical packages exist: the 112G SerDes part needs more I/O lanes and is substantially larger than the 224G version
Availability
Tomahawk 6 switch series shipping from June 2025; the TH6-Davisson CPO variant announced as shipping on October 8, 2025
Vendor
Broadcom (NASDAQ: AVGO)
Ecosystem Note
Adopted by switch OEMs building 800G and 1.6T AI fabric platforms
Overview
Broadcom's Tomahawk 6 is the first Ethernet switch chip to deliver 102.4 Tbps of switching capacity, doubling what was previously available in a single device. The TH6-Davisson variant is the third generation of Broadcom's co-packaged optics line and the first optically enabled switch to reach that capacity, driving 200 Gbps per optical channel and doubling the bandwidth of the earlier TH5-Bailly CPO part.
For AI clusters the significance is the fabric itself. Tomahawk 6 is compliant with Ultra Ethernet Consortium specifications and carries purpose-built features for large distributed training: modern AI transport protocols, congestion signalling and high-resolution telemetry for managing fabrics that span tens of thousands of accelerators. It interoperates with DR-based transceivers and with LPO and CPO optical interconnects at 200 Gbps per channel, which keeps the optics supply chain open.
The platform is offered in two package formats — a 224G SerDes version and a physically larger 112G variant that needs more I/O lanes — and is designed to serve both scale-up and scale-out roles over Ethernet. QS Compute supplies AI cluster networking, switch silicon, DPUs and rack-scale interconnect hardware — contact us for pricing and availability.
Key Benefits
Maximum fabric bandwidth: 102.4 Tbps per switch chip. Integrated optics: third-generation co-packaged optics cut power and simplify cabling. AI-native features: UEC compliance with telemetry and congestion signalling. Open ecosystem: interoperability across DR, LPO and CPO optical interconnects.
Applications
Frontier AI training and inference clusters, scale-up and scale-out GPU fabrics, hyperscale datacentre spine and super-spine layers, high-capacity storage networking, and AI networking platforms from switch OEMs.
Request a Quote — BROADCOM TOMAHAWK 6 — DAVISSON — 102.4 TBPS ETHERNET SWITCH WITH CO-PACKAGED OPTICS
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →