Gazettabyte is asking industry figures for their thoughts after attending ECOC 2025 in Copenhagen. Here are the first contributions from LightWave Logic's Yves LeMaitre, Maxim Kuschnerov of Huawei, and LightCounting's Daryl Inniss.
The Little Mermaid, Copenhagen
Yves LeMaitre, CEO of LightWave Logic
The optical centre of gravity has shifted towards AI networking; everything else is becoming an afterthought. Even data centre interconnect/ ZR coherent optics, a major topic at OFC2025, is relegated to a secondary topic.
The achievement of 400G/lane is happening faster than everyone thought. The race to chiplets, co-packaged optics, integration and the co-packaging of Electronic and photonic ICs (EICs/PICs) is what will define the winners of tomorrow. Winning in the transceiver world might feed you today, but you'd better adjust quickly to the new AI world order.
As for what I learned, our little and comfortable world of photonics is being rocked by the semiconductor giants stepping in with their oversized wallets and investments. Silicon foundries and major semi-players are not just influencing; they are now driving the photonics roadmap. The shock of culture and mutual discovery between somewhat segregated semi and optical players was on full display at ECOC.
I know the optical royalty was attending ECOC at the Bella Center, but did we need that much presence from the Danish police? Or was something else happening? Regardless, we had a blast in our tiny Lightwave Logic electro-optics polymer bubble and away from the drones. What a great show!
Maxim Kuschnerov, Senior Director R&D, Optical Technology Laboratory at Huawei
ECOC 2025 was not the stage for major announcements — those now tend to happen at larger AI infrastructure conferences. On the surface, walking around the exhibition, it was hard to tell that we’re in the midst of one of the greatest technological breakthroughs of our time. Everything looked the same, and all the familiar faces were still there — just one year older. But looking deeper, fascinating technical discussions unfolded. The conference sessions featured several notable highlights, such as Linfiber’s 0.052dB/km hollow-core fibre, imec’s 110GHz GeSi EAM, Microsoft’s 11,154km hollow-core fibre experiment, and the use of submarine optical fibres for earthquake precursor monitoring.
One particularly hot topic was GPU scale-up networking. After 18 months of debate around cabled GPU racks, the industry now agrees that we need cheap, dense optics in the scale-up domain — and we need them soon.
The silicon photonics versus VCSEL debate was prominent in the workshops, with both technologies appearing ready for neat-psckaged optics and co-packaged optics (NPO/CPO) and even optical input-output applications. Vendors demonstrated impressive high-temperature tolerance and reliability of VCSELs for silicon interposer use cases. In the classical Ethernet scale-out domain, discussions continued around thin-film lithium niobate and indium-phosphide, both of which seem likely to be ready for 448G/lane optical deployment.
The seemingly eternal coherent versus PAM-4 battle is now stabilising around the 2km reach for the 1600CL (coherent-lite) interface, with some optimists claiming that this time it will be different. Unlike the limited appeal of 800LR coherent for 10km applications, it will be different this time. The incursion of optical circuit switches into the data centre, along with the fundamental difficulties of intensity-modulation direct-detect in achieving a 2km reach, will carve out a large enough application space for coherent-lite 400G/lane optics. An interesting question here is how to reduce the 'coherent tax' relative to intensity-modulation direct-detect. Marvell, for example, proposed lowering the digital-to-analogue converter and analogue-to-digital converter (DAC/ADC) baud rate to PAM4 levels — a move that could help converge toward a largely homogeneous ecosystem of electronic and optical building blocks.
Lastly, we should remember that today’s massive AI boom is built on optical technologies that were designed and standardised before large-scale large-language models were a thing. The next generation of photonics will be heavily shaped by the demands of today’s extreme compute scaling. Adoption cycles and hesitation around so-called “unconventional” technologies — such as co-packaged optics, Optical I/O, or panel-sized optical interposers — will likely vanish like morning mist once optics becomes the sole enabler for highly dense GPU compute networks.
Daryl Inniss, Principal Market Analyst, LightCounting Market Research
Reflecting on ECOC 2025, two things stand out that make a profound commercial impact on the industry. These represent both technology advancements and market shifts.
Dr Nakajima’s ECOC plenary talk, Next generation optical fibre technology: Expectations and applications, set the stage by proposing hollow-core fibre (HCF) and multi-core fibre (MCF) as the fifth optical fibre generation. Dr Nakajima views the fifth generation as having a positive environmental impact and outlines the attributes of each fibre type.
However, widespread adoption of these fibres will require a significant change in the optical communications infrastructure ecosystem, and operating procedures for the fibres to become cost-effective. Importantly, the environmental contribution must be a part of the business case analysis for accurate evaluation.
History must be taken into consideration when thinking about hollow-core fibre and multi-core fibre. There are examples of new fibres that promise better performance than silica-based ones, but never hit mainstream status due to the inability to scale manufacturing. Fluoride-based fibre is one such example. Linfiber presented encouraging progress of hollow-core fibre manufacturing with a design that has the potential to produce higher volumes.
Its Interstitial-Tube Assisted Double-Nested Anti-Resonant Nodeless Fibre presented in the post-deadline paper session is designed to overcome the fibre-draw scaling problem. Interstitial tubes are inserted to control the gap between the main capillaries, counteracting the natural contraction that occurs during draw. Moreover, Linfiber demonstrated two firsts: the longest fibre on a continuous draw and the lowest loss.
The second aspect of ECOC's significant commercial impact is the number of vendors, their status, and the fibre attachment to the chip solutions they’re developing. These include Corning, Teramount, Senko, and GlobalFoundries, among others. Co-package optics is a new market opportunity, and its reliability, high level of performance, and diverse ecosystem are essential for success.
What surprised me was the prevalence of vendors using pluggable coherent transceivers in routers. This is booming due to the adoption of geographically dispersed AI training sites. And 800-gigabit is emerging as the data rate of choice.
What did not surprise is the number of suppliers and technologies competing to become the next-generation high-speed modulator.