OFC 2026 Outlook: How Silicon Photonics and CPO Will Reshape the Next Generation AI Interconnection System

Optical interconnection is becoming the most critical bottleneck and opportunity for AI infrastructure.

As the OFC 2026 exhibition is set to open this month in Los Angeles, the global optical communication industry chain will showcase the latest advancements in next-generation connectivity technologies. Driven by the strong demand for bandwidth in AI data centers, 800G optical modules are transitioning from pilot projects to mainstream, while 1.6T products have entered the ramp-up phase of mass production, marking a structural transformation in optical interconnection technology.

According to SemiVision Research, the rapid expansion of generative AI model scales is shifting the core bottleneck of data centers from transistor performance to interconnection bandwidth and latency. Jensen Huang, CEO of NVIDIA, proposed the "Huang's Law," which states that computing power can be continuously enhanced through process evolution and 3D packaging, but the improvement in I/O rates between chips lags behind, creating an "I/O wall." This structural contradiction is driving optical devices to migrate towards computing and switching chips to overcome constraints related to power consumption, loss, and transmission distance.

This year's OFC technical conference will be held from March 15 to 19, with the exhibition scheduled to open from March 17 to 19 at the Los Angeles Convention Center. Core topics include: the technological pathways for 1.6T and 3.2T optical modules, co-packaged optics (CPO) and new optical I/O architectures, silicon photonic heterogeneous integration, as well as the competition between pluggable solutions and CPO solutions.

Copper cables reach physical limits, optical interconnection migrates to chip side

The exponential growth of AI clusters is pushing optical interconnection technology to the core position of infrastructure architecture. Market research firm SemiVision Research points out that traditional data centers typically use copper cables for short-distance connections within racks, while pluggable optical modules are responsible for extending links between racks. However, as SerDes rates rise to 200 Gb/s per channel, the physical characteristics of copper cables are becoming a critical bottleneck.

According to Business Wire, at these rates, traditional passive copper cables can no longer reliably span a single server rack, and even face challenges with limited transmission distances within the rack. This physical limitation is driving optical devices to continuously migrate towards computing and switching chips in order to achieve a new balance between power consumption, loss, and transmission distance.

800G mainstreaming, 1.6T mass production imminent

From a product evolution perspective, the optical module market is showing a clear upgrade rhythm: 800G products are entering a strong growth phase in 2025 and accelerating penetration into the mainstream market; 1.6T products are set to begin ramping up mass production in the second half of this year.

On the specific product level, Accelink has publicly showcased and sent samples of the 1.6T OSFP224 DR8 transceiver aimed at AI data center scenarios. SemiVision Research analysis indicates that as the scale of AI clusters rapidly expands to hundreds of thousands of GPUs, optical interconnection has become one of the most critical bottlenecks in the AI infrastructure stack and is also the most valuable investment track.

The competition between CPO and pluggable solutions becomes the focus

The route dispute between Co-Packaged Optics (CPO) and pluggable solutions will become one of the core topics at OFC 2026. CPO integrates the optical engine and switching chip on the same substrate, significantly reducing power consumption and signal loss, and is seen as the long-term evolution direction for ultra-large-scale data centers; the pluggable solution, on the other hand, still dominates the current stage due to its flexibility and maintainability.

According to SemiVision Research, NVIDIA, Broadcom, and Marvell will present on manufacturability issues at this year's OFC, a topic directly related to the timeline for CPO's transition from technical validation to large-scale commercialization. Meanwhile, the evolution path from pluggable to Linear Pluggable Optics (LPO), and then to CPO, along with the resulting supply chain restructuring, will also be a focal point of the exhibition.

Silicon Photonics and Laser Technology as Key Variables in the Supply Chain

At the underlying technology level, silicon photonic heterogeneous integration—covering the integration of silicon photonics with thin-film lithium niobate (TFLN) and III-V materials—is listed as one of the key focus areas for OFC 2026. SemiVision Research points out that the supply-demand bottleneck of laser technology has become one of the key factors restricting the scale expansion of optical interconnects.

At the same time, the competition between VCSEL and MicroLED technology routes in the field of short-range optical interconnects, as well as optical interconnect solutions aimed at replacing copper cables within AI systems, will also receive widespread attention. Notably, Optical I/O (OIO), as a packaging-native optical link technology, is expected to support the decoupled deployment of AI systems and is becoming a new direction worth tracking.

Outlook: Optical Interconnects Entering a Strategic Competitive Window

As OFC 2026 convenes, it coincides with a critical juncture where the demand for optical interconnects in AI data centers is accelerating. The product iteration from 800G to 1.6T, the architectural evolution from pluggable to CPO, and the short-range replacement from copper cables to optics are all advancing along multiple technological curves, collectively building the technological foundation for this round of optical interconnect revolution.

SemiVision Research believes that as the scale of AI clusters continues to expand to hundreds of thousands of GPUs, optical interconnects have transcended their traditional supporting role and are becoming a core variable that determines the performance ceiling of AI infrastructure. The technological consensus and industry direction formed at this year's OFC will have a profound impact on the landscape of optical communication in the coming years