TEL：+86 158 1857 3751 
Driven by the rapid development of large language model training, inference, and commercial applications, global cloud service providers and major internet companies have significantly invested in building AI data centers. As the critical "data arteries" enabling high-speed interconnections within data centers, demand for AI optical modules has surged accordingly. The global AI optical module market grew from RMB 600 million (USD 90 million) in 2020 to RMB 6 billion (USD 900 million) in 2024, achieving a compound annual growth rate (CAGR) of 82.2%.
Looking ahead, propelled by continuous iterations of next-generation high-speed products (such as 1.6T and 3.2T) and the gradual commercial adoption of new low-power architectures, the global AI optical module market is expected to expand further. By 2029, the global AI optical module market is projected to reach USD 22.4 billion, with a CAGR of 30.1% from 2024 to 2029. This growth is primarily fueled by three core drivers: explosive growth in AI computing power demand, global data center upgrades, and cloud service expansion, as well as technological iterations and commercialization breakthroughs in optical modules.
Current Technological Landscape
While mainstream optical modules in global data centers still remain at the 1.6T generation, Huagong Tech has completed prototype validation of its 3.2T CPO (Co-Packaged Optics) optical engine, which is equivalent to expanding the Beijing-Shanghai high-speed railway from a single lane to an eight-lane bidirectional thoroughfare. The 1.6T optical module (with a rate reaching 1.6 trillion bits per second) has become the core enabler of next-generation AI computing infrastructure, with NVIDIA's Blackwell series GPUs listing it as standard equipment. This has directly driven market demand from an initial 2 million units to 4-8 million units.
In terms of technological pathways, there are two mainstream solutions for 1.6T optical modules: 16×100G and 8×200G. Industry consensus suggests that 8×200G will become the preferred choice for large-scale commercial adoption due to its high integration and superior energy efficiency. The Linearly Driven Pluggable (LPO) technology, favored by NVIDIA for its low-power characteristics, has demonstrated 30% energy savings compared to traditional DSP modules in tests, making it highly attractive for hyperscale data centers.
Silicon photonics technology integrates electronic/photonic devices onto a single chip through CMOS processes, achieving high integration and cost-reduction potential, which has become a core driver for the 1.6T era. Accon Photonics will demonstrate its new low-power 1.6T OSFP224 DR8 optical module at OFC 2025, featuring a 3nm process DSP chip that significantly reduces module power consumption.
