The Zhitong Finance App learned that according to TrendForce Jibang Consulting's latest AI Infra (AI infrastructure) industry research, as AI server clusters continue to expand and accelerator bandwidth demand rises rapidly, it is estimated that global AI networking & interconnect (AI Networking & Interconnect) output value will grow from US$37.9 billion in 2024 to US$279.8 billion in 2027, with a compound annual growth rate of 95% from 2024 to 2027. The focus of market growth is expanding from back-end network switches and optical modules to accelerator high-speed interconnections and network cards. In the future, the shipping mix and network topology of various AI chip platforms will be the key to influencing market size and supply chain value distribution.
Looking at the product structure, switches are still the largest source of production value, which is expected to reach US$105 billion in 2027, accounting for nearly 38%. Network interface cards and data processors (NICs/DPUs) benefit from the increased configuration volume and bandwidth upgrades of next-generation platforms, and the output value will increase by more than 100% per year in 2027 to 57.4 billion US dollars. Pluggable optical modules are driven by high-speed specification upgrades, with an estimated output value of US$84.1 billion in 2027. Together, the above three types of products will account for about 88% of the overall AI network and interconnection output value.
In terms of interconnection analysis, the output value of high-speed interconnection (scale-up) between AI chips has grown the fastest. As GPU platforms evolved towards cabinet-level architectures, chip interconnections gradually extended from server motherboards to independent switch trays, driving increased demand for switching equipment and copper cables. According to the AI cluster deployment scale, the share of scale-up in AI network output value will increase from 11% in 2024 to 24% in 2027. In 2027, the interconnection within the GPU cabinet is still mainly copper cables, and Google (Google) TPU has expanded the cluster size through ICI optical interconnect with optical circuit switches (OCS).
The back-end network (scale-out) connected to servers is the interconnection field with the largest production value, and the bandwidth upgrade will drive the main revenue of pluggable optical modules to shift from 800G to 1.6T in 2027. Even though the unit price of 1.6T products continues to decline with mass production, it is still more than double that of 800G, so the shipping mix gradually favors higher rate products, which can still drive an increase in overall output value. Jibang Consulting expects the annual revenue growth rate of pluggable optical modules to reach 76% in 2027, which is higher than the growth in shipment volume. It is estimated that 1.6T will contribute about 56% of revenue.
It is worth noting that the increase in AI chip shipments will not be converted in equal proportion to network communication output value. The interconnection topology, switch configuration, and network card bandwidth used by each platform will all affect the amount of network communication corresponding to a single accelerator. In addition, new optical connection technology and topology architectures are also simultaneously boosting overall production value. For example, the introduction of co-packaged optics (CPO) and all-optical switches (OCS) each supports different interconnection requirements. The CPO co-encapsulates the optical engine and switching chip, so that part of the optical value is transferred from a pluggable module to the switch; OCS supports large-scale cluster interconnection through optical path switching. Overall, the value allocation of AI networks and interconnect supply chains will continue to adjust as platform combinations and optical interconnect architectures evolve.