ECOC 2026 provided a useful reality check related to the importance of optical communications within the AI infrastructure build-out supercycle. Before the event, there were three questions on my mind. First, is 1.6T moving toward genuine commercial deployment? Second, which architectures matter as AI clusters scale beyond traditional pluggable optics? Third, can the optical supply chain and manufacturing keep pace to feed the insatiable demand for AI infrastructure?
ECOC provided some meaningful answers, and I left with the impression that no single optical architecture has emerged as the clear winner. Rather, the segment is moving toward a more heterogeneous infrastructure model where pluggables, LRO, LPO, NPO, CPO, optical switching, and advanced fiber technologies will each serve different parts of the AI networking stack.
With that broader context set, here are my three big takeaways from ECOC 2026.
1.6T Moving From Roadmap to Reality
My first takeaway is that 1.6T is moving from the roadmap to reality. The Ethernet Alliance exhibited 1.6T interoperability at ECOC, leveraging a multi-vendor ecosystem covering switches, modules, cables, and test equipment. In combining 224G SerDes technology and 1.6T OSFP connectivity, the demonstration offered a useful example of how the ecosystem is transforming the standard into commercially deployable infrastructure.
Cisco showcased its associated advancements across optical networking, routing, coherent optics, and the power of its Silicon One custom silicon by leading discussions related to OIF CMIS and the broader evolution of Ethernet for AI. Optical modules must evolve to support the greater network traffic demands of AI, and they will grow in complexity with expected higher lane counts, configuration flexibility needs, and expanded telemetry and security capabilities. At 1.6T, interoperability and management become as important as raw bandwidth. AI data centers will contain thousands of optical links, and operators cannot manage those links through bespoke configurations and manual processes.
My Point of View: Cisco’s efforts point to why vertical integration is essential as optical complexity rises. Cisco and other total solution providers will facilitate the requisite deployment and operational management of heterogeneous networks, in addition to providing speedier optics.
NewPhotonics introduced its laser-integrated 6.4T NPO optical engine and showcased its broader portfolio of 1.6T PIC solutions at ECOC 2026, including FRO, LRO and LPO+ architectures. NewPhotonics and Tower Semiconductor also announced high-volume shipments of laser-integrated, serviceable optical engine PICs supporting 800G and 1.6T applications. The company’s announcements clearly point to how the optical industry is broadly scaling manufacturing and deployment.
My Point of View: The level of NewPhotonics’ integration is impressive, and there is tremendous value tied to providing improved power management and higher reliability. NPO has great promise as a transition bridge to CPO, and it may end up having a longer lifecycle than many expect.
OIF plays an important role in facilitating standardization and ecosystem efforts in optical networking. At ECOC 2026, it brought 39 member companies together for a live multi-vendor interoperability demonstration covering 448G and 224G electrical interfaces, CMIS, energy-efficient interfaces, co-packaging, and optical systems. I personally spent time with OIF to better understand how the industry is moving from individual specifications toward interoperability across the complete AI connectivity stack. OIF’s work on 448G is particularly significant as the industry looks beyond the 224G electrical interfaces supporting the current generation of 1.6T architectures.
My Point of View: OIF’s importance as the “glue” that binds broader industry collaboration can not be understated. Standards and interoperability are critical to reducing integration risk, preserving vendor choice, and turning promising technologies into deployable infrastructure that fosters innovation.
The Optical Architecture Debate Is About Deployment
My second takeaway is that ECOC 2026 did not produce a definitive winner between pluggable optics, LPO, LRO, NPO, and CPO. Instead, the discussion moved toward the practicality of where each architecture makes the best economic and technical sense. Scale-up, scale-out, and scale-across networks have different distance, bandwidth, latency, thermal, and serviceability requirements. A technology that works exceptionally well inside a rack does not necessarily equate to the best solution within a data center or across clusters of data centers.
Corning provided a good example of how this architecture transition is expanding the addressable market for AI infrastructure. At ECOC 2026, the company introduced its Contour Form Micro Cable and showcased multicore fiber, dense MMC connectivity, and its existing passive CPO and NPO portfolio. Corning’s micro-cable technology is designed to increase fiber density within existing pathways to improve performance and scalability without growing the form factor. Corning also highlighted a multicore fiber approach that can deliver four times the capacity per fiber within the standard 125-micron cladding footprint by moving fiber closer to the chip through CPO and NPO architectures.
My Point of View: Corning continues to demonstrate the importance of AI optical opportunity beyond transceivers. As data rates rise – fiber density, cable routing, connectors, and fiber-to-chip connectivity become system-level constraints. Companies like Corning that solve these physical constraints have an opportunity to reap the rewards as AI data center deployments scale.
Lightera offers another solution to physical constraint challenges, but further out in the network. The company showcased multicore fiber technology through the OIF interoperability program and highlighted research demonstrating 52.8 Tb/s aggregate transmission over 303.4 kilometers of uncoupled two-core multicore fiber without intermediate optical repeaters. Lightera, the new global brand created by Furukawa Electric for its optical fiber cable business, was recognized for its efforts at ECOC 2026. It received the event’s Fiber Infrastructure Innovation Award for its Snap-Beam branded connector, designed to provide compact, detachable optical connectivity for photonic integrated circuits and CPO applications. Multicore fiber addresses capacity and density at longer distances, while advanced connectors address density closer to the chip.
My Point of View: Advancing bandwidth alone is not enough. There is a need for scalable AI infrastructure that incorporates deeper synchronized physical-layer capabilities. Companies, including Lightera, use it to scale and do so more efficiently.
Optical Switching and Photonics Are Moving Closer to the AI Fabric
My third takeaway is photonics’ growing role beyond traditional optical transport. ECOC 2026 showcased a growing industry trend in optical circuit switching and photonic architectures designed specifically around AI workloads. The fundamental optical networking challenge is about moving enormous quantities of data that feed training and inferencing efficiently between compute resources.
Salience Labs showcased its 64-port silicon photonic all-optical circuit switch, positioning optical circuit switching as a technology for both scale-out and scale-up AI architectures. Optical switching can create direct optical paths between compute resources, and the potential benefits include driving lower latency, lower power consumption, and more efficient use of expensive compute infrastructure. The commercial challenge is making optical switching reliable, manageable, and programmable enough for hyperscale environments.
My Point of View: Salience Labs is a company that I will continue to watch closely because optical circuit switching represents a different way of thinking about AI networking. Photonics is decidedly faster, and it can also become an active part of the network fabric itself.
Final Thoughts
In attending my first ECOC event, I left with a sense of awe related to the continued innovation in optical networking that is keeping pace with the demands of AI applications and workloads. The speed of modern AI innovation requires commensurate speed in connectivity, but not every technology investigation will become a commercial success. The opportunity is understanding where value can be created, and lasers, PICs, DSPs, fiber, connectors, packaging, testing, manufacturing, and the software and management layers connecting all of these technologies matter.
ECOC 2026 went far in my opinion to highlight what is possible for the future of optical networking. It will be interesting to gauge what eventually materializes in the future.


