At the 27th China International Optoelectronics Exhibition (CIOE), held from September 9 to 11 at the Shenzhen Convention and Exhibition Center, optical connectivity driven by artificial intelligence took center stage. Exhibitors presenting modules, chips, fibers, connectors, and test equipment reported capacity shortages and limited supplies. Some orders have already been placed for 2027, as large clusters for training and inference require more high-speed channels between servers, switches, and GPUs.
For AI-focused data centers, the primary outgoing product in 2026 remains 800G optical modules. Meanwhile, leading manufacturers are rolling out 1.6T modules for mass production and commercial validation. Comments from the exhibition indicate a phased transition: 800G still accounts for the bulk of shipments, 1.6T is beginning to increase output in limited quantities, and broader growth depends on how quickly server and switch platforms adopt corresponding ports.
Higher speeds also influence the choice of architectures—such as LPO (Linear Pluggable Optics), NPO (Near-Packet Optics), and CPO (Co-Packaged Optics). These architectures represent a trade-off between scale-out and scale-up clustering, balancing power, density, and reach. Standardized pluggable devices still dominate many scale-out connections, while NPO and CPO are attracting more attention within dense scale-up environments where low latency and high bandwidth density are valued.
The pressure extends up the supply chain. Optical chips, MPO and FAU assemblies, specialized fiber, and high-speed electrical chips are facing increased demand. Thin-film lithium niobate (TFLN) technology is moving from lab demonstrations to small-batch production, but it still requires process maturity and customer qualification. General 1.6T architectures, such as 8×200G, still rely on scarce DSP silicon and drivers, where global specialists maintain a significant market share despite relatively strong Chinese modular manufacturing capacity.
Test equipment is another limiting factor: every speed jump requires higher throughput and more complex end-to-end testing of devices, modules, and systems. This shift opens opportunities for Chinese instrument manufacturers, who have historically lagged behind foreign counterparts in oscilloscopes, clock signal restorers, and related production testers, although the share of high-end equipment still largely belongs to international suppliers.
Industry reviews conducted around the exhibition also note that scaling up 1.6T will be a topic for 2026, while volumes of 3.2T and above are mostly in the sampling stage. For buyers planning AI clusters, the reality is a race for bandwidth and qualification, not the launch of a single product. Demand for optics is growing right now; 800G still fills most supply slots, 1.6T is being supplied by manufacturers in early commercial batches, and the bottleneck remains the supply from modules through chips to test equipment next year.
