Click to listen to the full contentRecently, the new generation large modelGemini 3 has significantly improved its capabilities, coupled with the United States launching the AI “Genesis Project” which further expands computational infrastructure, leading to a collective strengthening of optical modules, with the optical module (CPO) index reaching a new historical high.
We believe that the breakthroughs of Gemini 3 at the large model level also demonstrate the value of ASIC chips represented by TPUs (Tensor Processing Units) and search engines, strengthening the fundamentals of cloud and search businesses, and proving the significance of the “hardware-technology-ecosystem” full-stack capability. As the AI industry trend is once again recognized by the market, considering that large model training still requires higher bandwidth, GPUs, and optical modules, the optical module market may maintain high prosperity in the next one to two years.
Previously, tech giants needed to useGPU chips to train large models, while Gemini 3 can smoothly reference data sources from search engines, showing significant advantages with TPUs, and ASIC chips demonstrate strong competitiveness. To reduce dependence on GPUs, many tech companies are beginning to shift towards ASIC chips. According to media reports, overseas internet companies are considering investing billions of dollars to transition from GPUs to TPUs fordata center construction.
We believe that, on one hand, regardless of the type of chip used, the use ofoptical modulesis inevitable. Due to the rapid growth in bandwidth demand for AI training and inference networks, the market previously expected that the 1.6T optical module in 2026 would be significantly revised upwards, coupled with good performance in the third quarter reports, the potential of the optical module-related sector is gradually strengthening.
On the other hand, the increase in ASIC market share is expected to drive the growth of optical modules. Although the performance of ASICs can only reach 70%-80% of the same generation GPUs, the decrease in costs is more significant, with the cost per token compressible to about 50% of that of GPUs. The usage of optical modules is independent of the performance of the main chip and only relates to the number of chips. Currently, the computational infrastructure and AI applications are forming a positive cycle, and under the dual promotion of policy and industry, the demand for AI chips is increasing day by day. This means that the cost share of optical modules in cloud vendors’ capital expenditures is expected to increase significantly.
Risk Warning: This material is provided by ICBC Credit Suisse Fund Management Co., Ltd. as customer service information and is not promotional material for funds. The views are for reference only, are time-sensitive, and do not constitute investment advice or profit guarantees. Investors should not replace their independent judgment with this information or make decisions based solely on it. The fund manager manages and utilizes fund assets in accordance with the principles of diligence, honesty, and prudence, but does not guarantee that the fund will definitely make a profit or guarantee minimum returns. Investors should carefully read the “Fund Contract”, “Prospectus”, “Fund Product Information Summary”, and other legal documents of the fund, fully understand the risk-return characteristics, fee structure, and charging standards of various sales channels, and make prudent investment choices based on their own investment objectives, investment horizons, investment experience, asset status, etc., and after considering the suitability opinions of sales institutions. Funds carry risks, and investments should be made cautiously.
Welcome to share
