• Technical Conference:  30 March – 03 April 2025
  • Exhibition: 01 – 03 April 2025
  • Moscone Center, San Francisco, California, USA

How Optical Interconnects to Meet AI Demand?

Thursday, 03 April, 14:30 – 15:30

Theater III

Session Description:

Since last year, there have been unprecedented demands for optical interconnect products, especially high-speed optical transceivers. With computing power doubling every 3-4 months, AI computing advances faster than before and may require our industry to evolve much quicker. While AI brings excitement and opportunities to our industry, many challenges exist. Can optical interconnects keep up with AI development? Can existing technologies move fast enough to meet AI requirements? Are disruptive technologies required? Do we need to change the way we develop new technologies and products? This panel will bring together experts from various industry segments, including end users, system vendors, optical transceivers, and chip companies, to discuss these questions. This panel intends to have end-users speak loudly; what new optical interface, like CPO, LPO, LRO, etc., will likely take off after n+1 years? What are the essential requirements and deployment issues? How should vendors improve to address those issues? With various products ready, what else must vendors address to keep end-users happy enough to issue PO? Critical questions precisely:

  • AI is exploding. Does this also drive Optics to continuously explore and break through its technical boundaries and bottlenecks, such as power consumption, cost, reliability, etc.?  
  •  Which optical technology solution will accelerate the development of large-scale AI clusters, CPOs, LPOs, pluggable modules, or beyond both?  
  •  800G has been deployed, and pluggable modules have been focused on 1.6T. What is the progress of technologies such as 200G/Lane electrical interface rate? How will these technologies play a leading role in data centers?  
  • Both CPO and LPO have their advantages but also face various problems. For example, the reliability of CPO needs to consider both testability and serviceability, while the performance of LPO depends too much on adapting the master chip. How do we solve the performance differences between different master chips and different channels?  
  • When will CPO vs LPO/LRO meaningfully impact Data Center Optical Interconnect?

Organized by: