Sep.1 2026, The AI boom is changing the optical communications industry in a way that goes well beyond the race from 800G to 1.6T.
As AI models and computing clusters continue to scale, optical connectivity is becoming a system-level constraint. Power consumption, connection density, latency and fiber capacity are emerging alongside the traditional challenge of increasing bandwidth.
That was one of the clearest messages from the 7th Liqiang Forum on Photonic Integrated Circuits, held in Xi’an, China, from August 21 to 27. The forum brought together more than 1,000 experts and representatives from more than 280 organizations, according to the organizers.
The discussions suggest that the next wave of optical innovation will not be defined by a single faster transceiver. Instead, it will involve changes across the optical link—from photonic integration and packaging to fiber and network architecture.
Fiber is becoming part of the solution
One of the most interesting observations at the forum came from Yangtze Optical Fibre and Cable executive Rui Chun Wang, who outlined six major challenges for AI optical interconnect: power, density, bandwidth, latency, return loss and fiber bend performance. His central point was that faster optical modules alone cannot overcome these physical constraints.
This is driving interest in new fiber technologies.
Hollow-core fiber, for example, can reduce propagation latency because light travels faster through an air core than through conventional glass. Multicore fiber offers another way to increase capacity by adding spatial dimensions rather than continuously pushing the data rate of individual channels. Other approaches, including bend-insensitive and polarization-maintaining fibers, address the density and packaging requirements of AI systems.
The implication is significant: fiber is no longer simply the passive medium connecting optical modules. It is becoming an active part of the AI networking architecture.
Coherent optics moves into shorter links
The same architectural shift is bringing coherent optics into applications far shorter than traditional telecom networks.
At the forum, FisiLink, a Fiberhome company, discussed coherent technology for AI computing centers and links of only a few kilometers. The attraction is not simply longer reach. It is the ability to carry more capacity over fewer fibers using wavelength-division multiplexing.
As AI campuses grow, the number of fibers required for direct-detection connections between buildings can become a physical and economic problem. Coherent optics could therefore become attractive even where distance itself is not the primary challenge.
This represents an important change in the competitive landscape: coherent and direct-detection technologies may increasingly compete based on total network cost and fiber efficiency, rather than transmission distance alone.
From CPO to manufacturing
Inside the data center, meanwhile, NPO and CPO remain major areas of development. The Liqiang Forum highlighted silicon photonics, heterogeneous integration and advanced packaging, alongside emerging materials and optical technologies.
The focus is also shifting from laboratory performance toward manufacturing.
Shaanxi’s photonic industry ecosystem, for example, is developing wafer-level silicon-photonics and silicon-nitride process platforms intended to support design, prototyping and eventual small-volume production. The broader challenge is to build the EDA, lithography, wafer processing and testing infrastructure needed to turn photonic integration into a scalable manufacturing industry.
The next optical bottleneck is the whole link
The evolution of AI networking is therefore changing the definition of an optical interconnect.
The first phase was largely a race in transceiver speed: 400G, 800G and now 1.6T. The next phase is broader.
Photonic integration can reduce electrical distances inside the system. NPO and CPO can bring optics closer to switching and computing. New fiber technologies can address latency and spatial capacity. Coherent optics can reduce fiber requirements between AI facilities.
All of these developments point in the same direction: more data, less power, less space and fewer physical connections.
That may be the most important signal from the Liqiang Forum. In the AI era, the optical industry is no longer optimizing an optical module in isolation. It is beginning to optimize the entire optical path as part of the computing architecture.