CPO and ELSFP – 1Q26 Update

Since Cignal AI’s 1Q25 report on CPO was published a year ago, the technology has advanced rapidly towards commercial deployment. TSMC’s Coupe process has jumpstarted efforts by Nvidia and Broadcom to have 200G/lane CPO platforms deployed by the end of 2026. Lumentum announced its largest purchase order ever for the ELSFPs that drive CPO, and Coherent announced a PO for its high-power laser for the same application. Ciena purchased Nubis, a company focused on open CPO and NPO. Both Lumentum and Coherent recently stated that they expect to ship into hardware designed for the scale-up network by 2027. All told, there is a lot of noise in the market, but no products shipping in volume yet. The question is, how far away are we from that inflection point?

CPO is still inevitable, but is it now imminent? The next twelve months should provide an answer.

CPO proponents have a list of reasons why CPO is inevitable, while its opponents have an equivalent list of why it is unnecessary. To date, reasons not to deploy CPO have outweighed the reasons to deploy, but that is changing as bandwidth density and power requirements in rapidly scaling AI networks are growing beyond what traditional pluggable optics can serve.

CPO’s technical challenges have not changed, and practical alternative solutions are available in most applications. Business and operational issues need to be addressed. Even Nvidia, which leads the charge for commercial CPO deployment, is hedging its bets with pluggable versions of its switches. CPO will emerge in small demonstration deployments this year, gaining the real-world experience that is needed before large-scale deployments start.

The key questions with CPO are its TCO relative to pluggables, and the effect on pluggable module shipments. Some of the most vertically integrated vendors, such as Coherent and Lumentum, are already pivoting to supplying other component parts (e.g. ELSFPs) into CPO. Others, such as Innolight and Eoptolink, are assemblers with little to offer in a closed CPO system and therefore are pushing for high density pluggables like XPO, or a more open CPO environment in which their manufacturing expertise can add value. The companies that stand to benefit the most are packaging experts like TSMC and perhaps OSATs like Celestica, Sanmina, and Fabrinet. CPO should be considered more of a process than a product, and it’s the process leaders that will benefit the most.

Key Takeaways

• CPO now seems imminent, with major effort and money being spent by Nvidia and Broadcom in partnership with TSMC.
• CPO’s downsides have kept the technology from mass deployment for years. Now that Nvidia is assuming the risk to develop and deploy CPO at scale, many of those concerns (reliability, cost, etc.) could be eliminated allowing the significant benefits (power, size, maintenance, etc.) to dominate the discussion.
• CPO’s main applications are within the scale-up and scale-out AI and compute networks. Scale-up is where there is pain (bandwidth, density, power), but that network is currently served by copper, which beats optics in power and cost and will therefore be hard to dislodge. Scale-out is already 100% optical and looks like a good place to start CPO deployments, but there are existing solutions still working well (pluggables) and improved solutions that can be adopted (LPO/LRO with XPO). Nvidia has started its efforts with scale out due to lower risk, and therefore that is where deployment will begin. Scale-up will follow in 2-3 years.
• CPO’s impact on pluggable optics will be minimal at first, as AI bandwidth requirements have grown so large that both have plenty of headroom. Long term, CPO could have an impact on the growth of the next generation of pluggable optics (400G/lane and higher). In the scale-up network, the impact is nonexistent, as that will be an entirely new market for optics.
• The number of CPO ports will be small in 2026, with growth starting in 2027 and taking off in 2029/2030 as scale-up applications come online. Over 25 million CPO ports could, conservatively, be deployed annually by 2030.

Table of Contents

• Introduction
• Key Takeaways
• Defining CPO
• CPO Technology
  • Silicon Photonics
  • ELSFP
  • Chip-to-Chip Interconnect
  • Co-Packaging
• Applications for CPO
  • Scale-Out
  • Scale-up
• The Good and the Bad of CPO
• The Current Status of CPO
• CPO Forecast
  • ELSFP Forecast
  • CPO Ports Forecast
• Conclusions