Wedoany.com Report on Mar 11th, A recent research report from CITIC Securities points out that with the continuous breakthroughs in global artificial intelligence technology, the rapid upgrading and iteration of GPUs and dedicated ASIC chips are driving exponential improvements in computing power performance. This, in turn, places unprecedented high demands on data transmission capabilities both within and between data centers.
The report analysis suggests that in the infrastructure construction of AI data centers, an increasing number of clients are leaning towards network hardware with larger bandwidth. The core logic behind this trend lies in the cost and efficiency optimization brought by economies of scale: larger bandwidth leads to lower transmission costs per bit of data, better power consumption performance, and smaller physical size. In other words, high-bandwidth solutions have become a key path for AI computing clusters to reduce costs and increase efficiency.
The report uses the optical module market as an example. The current sustained high growth rate of 800G optical modules clearly reflects the urgent demand for network bandwidth driven by AI computing power needs. As AI model parameters expand and training cluster scales increase, existing network architectures are facing continuous pressure to upgrade.
Based on the above assessment, CITIC Securities provides a clear outlook for the evolution path of optical modules in the coming years. The institution believes that the demand for 800G optical modules is expected to continue its high-speed growth trend in 2026, becoming the main specification for AI data center network deployment. Meanwhile, higher-bandwidth 1.6T optical modules are also expected to see a significant increase in shipment volume, gradually securing a place in hyper-scale clusters. At the more advanced technology reserve level, research and development for 3.2T optical modules has already officially commenced, paving the way for the next stage of further explosive growth in computing power demand.
