We are actively reaching out to refresh and refine our opinions, and we invite you to contact us if you have something to add to the conversation. Cignal AI will soon publish a detailed Active Insight report on CPO. But first, an editorial, and a genuine wish for feedback.
CPO: Technology of the Future for the Past Decade
It’s often noted that people tend to overestimate the impact of disruptive technologies in the near-term and underestimate them in the long-term impact. Co-Packaged Optics (CPO) are a perfect example of this. After a decade of silicon photonics progress and several years of no-volume hype, it now looks like real applications are on the horizon.
Despite impressive switch technology demonstrations by Broadcom, Intel, and Cisco (the latter two of which have abandoned the products), the future of CPO was uncertain in both viability and magnitude. As a market research firm, Cignal AI has not been able to put together a credible and defensible quantitative forecast for this technology – so we didn’t offer one. Other firms did, but we can’t agree with their justifications.
This doesn’t mean we ignored CPO. We have stayed abreast of the impressive R&D pursued by companies large and small – (see Optical Component Startup Tracker) but have remained on the sidelines when it came to building a forecast. All the work done with COBO, NPO, hybrid silicon photonics was deeply interesting from an engineering perspective, but ultimately, we felt there were no compelling advantages over conventional optical modules that would result in near term volume. We were right.
Investors: Distant Early Warning
Cignal AI held a series of 1:1 and group meetings with investors during January 2025. We look forward to conversations like these, as investors’ interests are often orthogonal to those of people in the industry. We enjoy being challenged by investors to explain and defend our positions and always come away from these exchanges with enhanced perspective.
For these investor meetings, we prepared a short set of slides to highlight what we expected to be the issues of interest amongst the investment community: Pluggable Coherent and IPoDWDM, Hyperscaler WDM spending and the impact of AI, 1.6T pluggables and AI cluster demand, the Infinera/Nokia merger, and the explosion in demand for pluggable datacenter optics.
We were wrong. Once the meetings began, it was clear that investors were intensely interested in co-packaged optics. The catalyst for this curiosity was a series of articles of questionable accuracy (here and here) from Taiwan indicating that TSMCs efforts to bring CPO to its process and packaging flow was imminent. Specifically, the articles claimed TSMC was integrating 1.6T optical interfaces into its chips and volume production would begin in 2H25.
This timeline was ridiculous, but this fact didn’t mitigate longer term investor concerns. It triggered an intense interest into the state of CPO and whether the imminent demise of pluggable optics was upon us.
These conversations made us realize that, despite our firm’s skepticism, now was a good time for us to focus on the state of CPO R&D and perhaps put a stake in the ground as to when the market would materialize. After analyzing this for the past few weeks, a much clearer, concise picture is emerging than what we’ve seen to date.
Turning the Corner: TSMC, Broadcom, Nvidia
Necessity is the mother of invention, and the networking requirements of AI are now pushing the envelope of copper interconnect within a rack (scale up) and perhaps, eventually, challenging the economics of pluggable optics used in networking switches (scale out). While the copper interconnect of the Nvidia NVL72 is a true marvel of engineering, making the jump to 400G/lane presents what looks like an insurmountable challenge. This technical challenge appears to be the forcing function for CPO.
TSMC is making major investments in enabling CPO, in conjunction with two critical partners – Broadcom and Nvidia. TSMC is approaching CPO not as a product, but as a critical step of its manufacturing process – specifically a version of its complex packaging solutions. It seeks to incorporate novel optics and mechanical products from outside suppliers to offer near-term and long-term CPO solutions. Meanwhile, optical suppliers and startups including Marvell, Lightmatter, and others are partnering with OSATs (outsourced semiconductor assembly and test entities) like Amkor, Sanmina and others to provide a similar path to CPO.
So, after several years as a solution looking for a problem, CPO finally has a killer app with AI. R&D and investment activity targeting attractive applications is underway, but even if the significant technical and manufacturing challenges are resolved – material volume is still a few years out.
What This Means
No doubt, the hype about CPO is about to take off. As foreshadowed in our recent investor meetings, people are likely to overreact in the short term. Many of you probably remember OFC 2016, at which investors fretted over the advent of silicon photonics and the impending destruction of the existing optical business. Silicon photonics was indeed a critical new technology and was widely adopted, but it didn’t jeopardize the optical business. And the prime mover and perceived leader for production silicon photonics at the time, Intel, was never able to capitalize on its success.
Next month brings the arrival of both Nvidia’s GTC conference as well as OFC ‘25. It is likely that Nvidia will provide a concrete roadmap to CPO solutions at GTC, most likely as part of its next generation Rubin platform. It’s also likely that Nvidia, following in Google’s footsteps (see The Optical Circuit Switching Market) will highlight the more immediate use of OCSs to combat the cost and power of electrical switches. OFC comes two weeks later, and after Nvidia’s announcements CPO and OCS are likely to dominate the event’s conversations.