CICC: Focus on the field of "AI diffusion"

CICC releases research report stating that there are signs of diffusion in the recent AI market trend, focusing on price setting in the upstream "shovel-selling" sector. Since the beginning of the year, the A-share market has continued to show a clear structural trend, with the ChiNext Index and the ChiNext 50 Index in the technology growth style significantly outperforming, rising by 30.9%/51.2% respectively, with the AI industry chain as the core theme. Since the second quarter, there has been increased volatility in the AI sector, with previously leading sectors such as semiconductors, optical modules, and PCBs experiencing historically high levels of trading congestion. Recently, there have been signs of the market gradually spreading towards upstream raw materials and infrastructure sectors with relatively lower trading congestion in the AI-related field, with good performances in sub-sectors such as minor metals, electronic fabrics, and optical fibers. With the continuous strengthening background of the AI industry trend, the current market may be more likely to show style diffusion rather than systematic style switching, with some specific sectors in the upstream of the AI industry chain being the current focus. Performance and market outlook of AI industry chain segments We preliminarily divide the AI industry chain into upstream computing power infrastructure, midstream large models and platforms, and downstream applications and terminals. Overall, there is a differentiation in the performance realization capability, valuation level, and trading congestion of each segment, as follows: Upstream computing power infrastructure: Strong performance realization, with the core sectors experiencing temporary high levels of congestion. The upstream of the AI industry chain mainly includes core areas such as AI chips, storage, and optical modules. In terms of performance realization, the demand for large models for training and inference continues to grow, with global cloud vendors increasing capital expenditures, driving rapid growth in performance in the upstream sector. In the first quarter of 2026, the profits of the electronic sector increased by 50.5% year-on-year, with profits in the industry sector of science and technology chips/optical modules/PCBs increasing by 200%/149%/67% year-on-year. However, the rapid increase in asset prices in the early stage is also pushing up the valuation center of the above-mentioned sectors, and the sustainability of the market in the future needs to focus on the pace of order and performance realization. In terms of trading congestion, the trading congestion in the above-mentioned sectors is relatively high, with the overall turnover of the electronic sector accounting for nearly 35% since the beginning of the year, and the turnover rate based on free market capitalization once exceeding 10%, with optical modules/PCBs having the highest turnover rate of about 10%, and science and technology chips of about 9%. Midstream and downstream large models and applications: Long-term industry space is large, but attention is needed on the matching of current listed company performance realization and valuation. The midstream of the AI industry chain mainly includes large models, Agent platforms, and focuses on software applications and terminals downstream. In terms of performance, the commercialization of AI agents facing enterprises is accelerating, with expectations of profitability of the commercial model warming up, but compared to overseas leading companies, many relevant areas in China are still in the early stages of investment, with uncertainty in terms of performance realization. The breakthrough of DeepSeek at the beginning of 2025 led to a rise in the sector, with the open-source AI Agent framework OpenClaw going public at the beginning of this year, top large model companies such as KNOWLEDGE ATLAS and MiniMax landing on the Hong Kong stock exchange, and the AI large model sector rising again, followed by oscillations and corrections. Since the beginning of the year, the turnover rate of the Computer/WD Basic Large Model Index has averaged 5%, lower than the previous peak. Which sectors are expected to benefit from the diffusion of the AI market? AI remains the main theme in the market, with the current focus shifting towards upstream raw materials and infrastructure sectors. The AI industry chain currently maintains a high level of prosperity and will continue to be the main theme in the medium to long term, but recent trends show differentiation in the market due to asset prices and valuations. In the outlook for the second half of the year in "Stable Progress, Far-reaching", we propose that the AI industry chain in the second half of the year will need to be "carefully selected". After achieving important breakthroughs in commercialization, the sector's space has been revised upwards with improving fundamentals, but in the future, selection should be based on valuations and the intensity of supply and demand. Currently, the trading congestion is high in the upstream areas of AI, and short-term fluctuations are prone to magnification, so attention should be paid to the pace of performance realization and the ability to digest valuations. Recently, the AI market trend has spread towards sectors with relatively low congestion, mainly focusing on sub-directions such as minor metals, chemical new materials, construction materials, engineering machinery, and electrical equipment. We have integrated the views of industry analysts at CICC for a bottom-up analysis: Minor metals: Focus on sub-directions such as tungsten, tin, indium, germanium, rare earths, etc. The iterative advancement of AI chips and computing power hardware is driving the continuous release of strategic minor metal demand, combined with industry supply-side policy control and capacity constraints, widening the supply-demand gap, supporting the logic of price increase. Market performance shows that the PE ratio TTM of the Wind Minor Metals Index is around 30x, below the 40th percentile over the past decade, with a relatively low concentration of institutional holdings. Basic Chemicals: Focus on AI new materials such as optical fibers, semiconductor materials, and liquid cooling materials. The expansion of AI computing power infrastructure and hardware upgrades continue to drive demand for chemical new materials, while semiconductor materials rely on capacity expansion and domestic substitution to achieve steady volume growth; liquid cooling materials adapt to the trend of high-power computing heat dissipation upgrades, with rapidly increasing penetration rates; high-end optical fiber materials support high-speed interconnection of computing clusters. These sectors were previously part of the traditional chemical industry valuation system, with room for improvement. Construction materials: Focus on sub-directions such as electronic fabrics, glass substrates, etc. Electronic fabrics, as key raw materials for copper-clad boards and PCBs, directly benefit from the high shipments of AI hardware and upgrades to high-frequency PCBs. In recent periods, electronic fabric prices have continued to rise, gradually entering a phase of fundamental realization. Glass substrates, as emerging advanced packaging materials for AI, adapt to the demand for high computing power and low latency AI chip packaging. Engineering Machinery: Focus on sub-directions such as diamond heat dissipation, optical module equipment, AIDC equipment chain, etc. The large scale establishment of domestic AIDC computing centers and the iterative expansion of optical module production capacity are driving the full release of demand for AI-specific equipment. Orders for diamond heat sinks, optical module production and testing equipment, power generation supporting equipment, and liquid cooling supporting equipment in sub-segments have seen a significant increase, with traditional engineering machinery companies accelerating their transformation towards AI high-end computing power support, and the industry logic and valuation system are expected to undergo continuous reconstruction. Electrical Equipment: Focus on sub-directions such as gas engines, transmission and distribution electrical equipment, outdoor cabinets, server power supplies, etc. The increasing power density of AI data centers, the rapid growth in electricity demand for computing industries, and the concomitant need for electrical equipment as the core energy supply track for AI computing expansion provide long-term support for prosperity. Currently, the PE ratio TTM of the electrical equipment sector is around 36x, near the 50th percentile over the past decade. Chart: Bottom-up review of AI market areas likely to spread Source: Wind, CICC Research Department