1.6 Tbps Ethernet Switch for Next-Generation AI Infrastructure

As AI workloads continue to scale, high-performance networking has become essential to keep thousands of GPUs efficiently connected. QuantaMesh TB064-IXP is designed to provide high-bandwidth, low-latency Ethernet connectivity for AI clusters, enabling efficient communication between GPU servers, compute nodes, storage, and network infrastructure.

Built for next-generation AI data centers, QuantaMesh TB064-IXP delivers high switching capacity and high-density connectivity to support demanding scale-out and scale up network architectures. Its optimized hardware and open networking capabilities provide the performance, scalability, and flexibility required for large-scale AI workloads.

Co-Packaged Optics (CPO) Architecture

Enabling the Next Generation of High-Speed AI Networking

As AI workloads continue to scale, the increasing number of GPUs and growing network bandwidth requirements are placing greater demands on conventional optical interconnect architectures. Higher data rates introduce challenges in signal integrity, power consumption, thermal management, and electrical reach.

Co-Packaged Optics (CPO) addresses these challenges by integrating optical engines directly alongside the switch ASIC, creating a shorter and more efficient path between electrical and optical signals.

Integrated Optical Architecture

Co-Packaged Optics (CPO) integrates optical engines directly alongside the switch ASIC to minimize high-speed electrical paths. By reducing electrical reach, CPO improves signal integrity and power efficiency while enabling high-bandwidth optical connectivity for next-generation AI networks. Combined with high-density fiber interfaces and advanced thermal solutions, CPO provides a scalable architecture for large-scale AI fabrics and hyperscale data centers.

Key Features

• 102.4 Tbps Switching Capacity
  

  Delivers 102.4 Tbps of switching capacity to support next-generation AI networking and large-scale GPU clusters.

• High-Density 64 × MMC-16 Fiber Interfaces
  

  Supports 64 × MMC-16 fiber interfaces, enabling high port density while simplifying optical cable management.

• Co-Packaged Optics (CPO) Architecture
  

  Integrates optical engines directly with the switch ASIC, shortening high-speed electrical paths to improve signal integrity, reduce latency, and lower power consumption.

• Efficient Liquid Cooling
  

  Uses liquid-cooled cold plates to directly dissipate heat from the switch ASIC and other high-power components, enabling efficient thermal management for high-performance AI networking.

• ORv3 Infrastructure Compatibility
  

  Designed to be compatible with Open Rack v3 (ORv3) infrastructure and supports 48V busbar power delivery for efficient rack-level power distribution.