TrendForce pointed out that as bandwidth specifications continue to upgrade, 1.6T pluggable optical modules are expected to replace 800G in 2027.
As the scale of artificial intelligence (AI) server clusters continues to expand, the demand for accelerator bandwidth increases rapidly, and the AI network and high-speed interconnection markets are experiencing explosive growth. TrendForce, a research organization, pointed out today that the global AI network and interconnection output value is expected to climb from US$37.9 billion (approximately NT$1.2 trillion) in 2024 to US$279.8 billion (approximately NT$9.1 trillion) in 2027, with an average annual compound growth rate of 95%.
TrendForce pointed out that the growth focus of the AI network communication market is extending from back-end network switches and optical modules to accelerator high-speed interconnection and network interface cards. From the perspective of product structure, switches are still the largest source of output value, with an estimated output value of US$105 billion (approximately NT$3.4 trillion) in 2027, accounting for nearly 38% of the overall market.
In addition, network interface cards and data processors (NIC/DPU) will benefit from the increase in configurations and bandwidth upgrades of new-generation platforms, and their output value will reach US$57.4 billion (approximately NT$1.8 trillion) in 2027, an annual increase of more than 100%; the output value of pluggable optical modules will be US$84.1 billion (approximately NT$2.7 trillion), with the three major products accounting for approximately 88% of the overall market.
It is worth noting that AI chip-to-chip high-speed interconnection (Scale-up) is becoming the fastest-growing field. As GPU platforms develop towards cabinet-level architecture, chip interconnection extends from server motherboards to independent switch trays, driving a simultaneous increase in demand for switching equipment and high-speed copper cables. TrendForce estimates that scale-up’s share of the overall AI network output value will increase from 11% in 2024 to 24% in 2027. AI infrastructure investment will gradually extend from simply expanding computing capabilities to data transmission efficiency between accelerators.
In terms of technology development, TrendForce pointed out that in 2027, the interconnection within the GPU cabinet will still be dominated by copper cables. However, different AI chip platforms have diverged in their technical routes. For example, Google TPU uses inter-chip interconnection (ICI) optical interconnection with optical switches (OCS) to expand the scale of the AI cluster, reflecting that high-speed copper cables and optical interconnections will continue to develop in the future according to different architectural needs.
On the other hand, the back-end network (Scale-out) connected to servers is still the interconnection field with the largest output value. As bandwidth specifications continue to upgrade, 1.6T pluggable optical modules are expected to replace 80 0G; although the unit price of 1.6T products has gradually declined with mass production, it is still more than twice that of 800G, driving the output value of pluggable optical modules to increase by 76% annually in 2027, of which 1.6T is expected to contribute approximately 56% of revenue.
TrendForce emphasized that the increase in AI chip shipments will not be converted into Netcom's output value in equal proportions. The interconnection topology, switch configuration, and network card bandwidth adopted by each platform will affect the allocation of supply chain business opportunities. As co-packaged optics (CPO) transfers part of the optical value from pluggable modules to switches, and OCS supports large cluster optical path switching, future competition in the AI network communication market will not only depend on shipment scale, but also on the technical layout of each player in the new generation of high-speed interconnect architecture.
