The new Co-Packaged Optics (CPO) platform achieves 200G/lane and marks a key milestone for next-generation networks in large-scale AI-focused data centers.
Broadcom has announced the launch of its third generation of CPO (Co-Packaged Optics) technology with a transmission speed of 200G per lane, during a key presentation from Palo Alto, California. This advance not only enhances energy efficiency and bandwidth density of optical interconnections but also reaffirms the readiness of the CPO ecosystem to power future AI model inference and training networks at scale.
🌐 From 100G to 200G/lane: The leap that defines the next generation
After years of development, Broadcom has demonstrated the maturity of its second-generation CPO technology with 100G/lane, integrating critical innovations in thermal design, OSAT processes, fiber routing, and automated testing. Now, with the introduction of the third generation, the ecosystem expands to new heights with 200G/lane, an essential requirement for interconnect networks that support thousands of nodes in distributed AI architectures.
“With our third-generation CPO solutions of 200G per lane, we are once again setting the standard for next-generation optical interconnections,” said Near Margalit, Vice President of Broadcom’s Optical Systems Division.
🔧 Expanding CPO Ecosystem
The launch of this technology is accompanied by milestones announced by key partners:
- Corning collaborates on advanced optical connectivity, already in production for the TH5-Bailly platform.
- Delta Electronics launches 51.2 Tbps CPO Ethernet switches in compact air and liquid-cooled formats.
- Foxconn Interconnect Technology announces the production of LGA sockets and critical optical components to integrate CPO engines.
- Micas Networks reports over 30% energy savings compared to traditional optical modules, with its TH5-Bailly switch.
- Twinstar Technologies celebrates volume shipments of high-density CPO fiber cables.
This consolidated ecosystem is a sign of progress towards industrial-scale CPO integration, crucial for maintaining energy efficiency in high-performance AI networks.
🚀 Ready for foundation models and generative AI
Broadcom’s third-generation CPO technology is designed for internal interconnections in networks with over 512 nodes, addressing the bandwidth, latency, and power consumption challenges posed by large foundation models. Additionally, Broadcom is already planning its fourth generation of 400G/lane, underscoring its leadership in optical network innovation.
These solutions are optimized to ensure reliability comparable to copper connections, minimizing disruptions and ensuring a low cost per processed token, a key metric in large-scale AI operations.
🧠 Long-term vision and commitment to open standards
Broadcom emphasizes its commitment to standardization and interoperability, key elements for the mass adoption of CPO in hyperscale architectures. Its roadmap includes ongoing collaboration with technology partners to optimize the design and manufacture of energy-efficient and scalable optical interconnections.
📌 In summary:
- Broadcom launches its third generation of CPO technology with 200G/lane, leading the way towards next-generation AI networks.
- The partner ecosystem, including Corning, Delta, Foxconn, and Micas Networks, has already validated solutions in mass production.
- CPO promises greater density, lower energy consumption, and critical reliability in hyperscale environments.
- The roadmap includes a fourth generation of 400G/lane and a commitment to open standards and scalability.
via: Broadcom