Broadcom Inc. announced that it is now shipping Tomahawk® 6 – Davisson (TH6-Davisson), the company’s third-generation co-packaged optics (CPO) Ethernet switch. Designed specifically for the explosive demands of AI networking, TH6-Davisson is the industry’s first to deliver an unprecedented 102.4 terabits per second of optically enabled switching capacity. Doubling the bandwidth of any CPO switch available today, TH6-Davisson sets a new benchmark for data center performance. Built on Broadcom’s foundation of CPO innovation and field shipments, the platform doubles the bandwidth of any CPO switch available today, while delivering disruptive advances in power efficiency and traffic stability to unlock optical interconnect performance required to scale-up and scale-out the world’s most demanding AI clusters.

“TH6-Davisson, our 3rd generation CPO Ethernet switch, represents a significant advancement for AI infrastructure,” said Near Margalit, Vice President and General Manager, Optical Systems Division, Broadcom. “By enhancing link stability and energy efficiency, we’re enabling smoother, more cost-effective AI model training. We designed this platform to scale large AI clusters by delivering on the three imperatives for optical interconnect — higher model FLOPs utilization, reduced job interruptions and improved cluster reliability.”

CPO: Purpose-Built for AI Networking                                                                              

The rise of large-scale AI training and inference is driving unprecedented east-west traffic in data centers, as XPUs and GPUs exchange vast datasets across tens of thousands of servers. Traditional pluggable optics are straining under this growth — consuming more power, introducing higher latency and demanding larger system footprints. Fine-tuned over multiple generations of development, TH6-Davisson embodies Broadcom’s CPO architecture that is now fine-tuned over multiple generations of development to overcome these barriers and deliver the bandwidth, efficiency and reliability required for next-generation AI networks.                          

World-Class Energy Efficiency                                                                             

From the ground up, TH6-Davisson was architected for power efficiency. By heterogeneously integrating TSMC Compact Universal Photonic Engine (TSMC COUPETM) technology-based optical engines with advanced substrate-level multi-chip packaging, the switch dramatically reduces the need for signal conditioning and minimizes trace loss and reflections. The result is a 70 percent reduction in optical interconnect power consumption — more than 3.5x lower than traditional pluggable solutions — delivering a step change in energy efficiency for hyperscale and AI data centers.           

Improved Link Stability and Traffic Performance                                                                      

As AI training jobs scale, link stability has become a critical bottleneck, with even minor interruptions causing measurable losses in XPU and GPU utilization. TH6-
 Davisson addresses this challenge by directly integrating optical engines onto a common package with the Ethernet switch. This highly integrated design eliminates many of the sources of manufacturing and test variability inherent in pluggable transceivers. The result is significantly improved link flap performance and higher cluster reliability as validated by a TH5-Bailly link flap study. 

Bandwidth Improvement and Interoperability                                                                            

Operating at 200 Gbps per channel, TH6-Davisson doubles the line rate and overall bandwidth of Broadcom’s second-generation TH5-Bailly CPO solution. Designed for interoperability, it seamlessly interconnects with DR-based transceivers as well as NPO and CPO optical interconnects running at 200 Gbps per channel. This ensures frictionless connectivity with the industry’s most advanced NICs, XPUs, and fabric switches, enabling next-generation AI and cloud clusters to scale without compromise.

CPO Roadmap                                                                                

Leveraging the advanced CMOS nodes and continuous breakthroughs in device technology, Broadcom is now developing its fourth-generation CPO solution. The new platform will double per channel bandwidth to 400 Gbps while achieving breakthrough levels of energy efficiency, extending Broadcom’s leadership in enabling the scale and sustainability of future AI and cloud networks.                                                                          

TH6-Davisson BCM78919 Product Highlights                                                                                              

  • 102.4 Tbps Switching Capacity                                               
  • 16 × 6.4 Tbps Davisson DR Optical Engines
  • 200 Gbps per link bandwidth
  • Field-Replaceable ELSFP Laser Modules                                                          
  • Supports Scale-Up cluster size of 512 XPUs and up to 100,000+ XPUs in two-tier networks at 200 Gbps per link
  • IEEE 802.3 Compliance — Interoperable with existing 400G and 800G standards                                                         

Availability                                                                                       

Broadcom is currently sampling the TH6-Davisson BCM78919 device to its early access customers and partners. Contact your local Broadcom sales representative for samples and pricing.