CICC: The "Differences" in AI Investment between China and the United States

In the past one or two years, under the dilemma of weak traditional demand, global growth might have faced greater pressure if not for the rise of AI. For example, the $1 trillion investment in technology hardware and software in the United States contributed one-third of the GDP in 2025 (Chart 1), not to mention the potential boost in factor productivity for future growth (Chart 2).

AI's contribution to the stock market is also significant. Since the release of ChatGPT at the end of 2022, the US stock market's Mag7[1] contributed 45ppt of the 84% return of the S&P 500 index, accounting for more than half (Chart 3). Since the release of DeepSeek in early 2025, the seven leading technology stocks in Hong Kong[2] once contributed 14ppt of the 37% return of the Hang Seng Index, also accounting for 40% (Chart 4). Not only in China and the US, but also in South Korea, Japan, and Taiwan, which are leading the global market in 2025, are also key links in the AI industry (Chart 5).

Chart 1: In the average annualized quarter-on-quarter growth of 2.5% in real GDP for the first three quarters of 2025, US technology hardware and software contributed 0.8ppt

Source: Haver, CICC Research Department

Chart 2: Since 2023, labor productivity in the US non-farm business sector has increased by 7.2%

Source: Haver, CICC Research Department

Chart 3: The US stock market's Mag7 accounted for 45ppt of the 84% return of the S&P 500 index, accounting for more than half

Source: FactSet, CICC Research Department

Chart 4: Seven leading technology stocks in Hong Kong accounted for 14 percentage points of the highest return of the Hang Seng Index at 37%, making up 40%.

Source: FactSet, CICC Research Department

Chart 5: Since the beginning of the year, the AI style in the global market has once again led.

Source: FactSet, Bloomberg, CICC Research Department

However, alongside the enthusiasm for AI is the persistent concern about bubbles. After three years of rapid advancement, few people now question the prospects of the AI industry itself, but there are still worries about the potential gap between the speed of realization and investment, just as the tech bubble of the 1990s laid a solid foundation for the booming development of mobile internet in the 21st century, yet did not prevent a large amount of excess investment from being squeezed out in the form of a bubble burst in 2000. Therefore, the bubble itself is not necessarily a bad thing, and it has also driven industrial development; discussing whether it will turn into a bubble is not very meaningful, what is more important is to confirm the stage we are in.

In this process, investment plays an important role, as the source and direction of funds directly determine the behavior of funds and the orientation of investments. As the "two poles" of the global AI industry landscape, due to differences in computing power infrastructure, chips, and models, there are significant differences between China and the United States in terms of funding sources and investment directions, so tracing these differences helps us understand the differences in development paths and provides insights into different investment directions.

Chart: The proportion of foundational layer investments in both China and the U.S. is around 87-88%, while the proportion of the technological layer is 12-13%.

Source: CICC Research Department

Chart: There is an industrial chain linkage between China and the United States at the basic level, with a self-sufficient split in chips, and mapping at the application level.

Note: This chart is only a rough schematic and does not represent a complete industrial chain.

Source: CICC Research Department

China-U.S. AI Industry Landscape: The U.S. has a first-mover advantage, while China is rapidly catching up; the U.S. "lacks electricity," and China "lacks chips," with limited model differences.

The cornerstone of the artificial intelligence industry lies in computing power infrastructure, models, talent reserves, and financial support from the capital market. In the early stages of development, the U.S. had significant advantages in computing power infrastructure, models, advanced talent development, and data quality. However, since the release of DeepSeek in early 2025, China has made breakthrough progress in the model layer, especially in the effectiveness of open-source models (Chart 6), and has begun to narrow the gap with the U.S. in multiple fields.

Chart 6: China has made breakthrough progress in the model layer, especially in the effectiveness of open-source models.

Source: The Atom Project, CICC Research Department

► Computing Power Infrastructure: It encompasses physical infrastructure centered around data centers, digital infrastructure represented by chip research and development, and the mobilization and delivery of computing power in the form of cloud computing. According to the China Academy of Information and Communications Technology's 2025 "Cloud Computing Blue Book" [3], it cited Gartner's data indicating that the global cloud computing market size will reach USD 692.9 billion in 2024, with North America dominating with a 54.3% market share, while China holds 16.8%, and the share is expected to further rise to 18.3% in 2025 (Chart 7).

Chart 7: The global cloud computing market size will reach USD 692.9 billion in 2024, with North America accounting for 54.3% and China holding 16.8%.

![](https://mmbiz-qpic.wscn.net/sz_mmbiz_png/fzHRVN3sYs9TlqfNYOQZlRIKyw1YU8gWx4Zj764ibdmibOa7PMtyCHG8vLibX9bGkicQU0qujaGVUOE9S61oXdWh1w/640? Data source: China Academy of Information and Communications Technology, Gartner, CICC Research Department

1) The United States has a first-mover advantage in infrastructure, but China is continuously closing the gap supported by its power supply advantage, making the U.S. more "power deficient." Although the U.S. currently has far more servers and data centers than China (Chart 8), China's computing power density is currently higher. For example, the number of data centers in the U.S. is more than 8 times that of China (4,165 vs. 500), but its capacity is only 1.7 times that of China (U.S. 53.7GW vs. China 31.9GW). More importantly, large-scale deployment of data centers requires power support. China's power generation in 2024 will exceed that of the U.S. by more than double (Chart 9). The electricity consumption of existing data centers in the U.S. already accounts for 4.4% of its total electricity consumption, while China's is only 1.1%[4]. The executive order signed by Trump in January 2025 explicitly stated[5] that "new large AI infrastructure must be accompanied by the construction of new clean energy power generation facilities to avoid occupying residential electricity demand."

Chart 8: The U.S. currently far exceeds China in the number of servers, physical data centers, and cloud computing facilities

Data source: Federal Reserve, Statista, IEA, Hawkins et.al, CICC Research Department

Chart 9: China's power generation in 2024 will exceed that of the U.S. by more than double

Data source: Wind, CICC Research Department

2) The U.S. dominates in chip research and development, while China's domestic substitution process is accelerating, but there is still a gap in advanced process technology. According to the Semiconductor Industry Association (SIA) statistics[6], the sales of the U.S. semiconductor industry will reach $318 billion in 2024, accounting for 50.4% of global revenue, while the sales shares of mainland China and Taiwan are 4.5% and 6.5%, respectively (Chart 10). Although the share is still significantly lower than that of the U.S., China's chip scale is growing rapidly, with total shipments of AI chips exceeding 2.7 million units in 2024, and the shipment volume of domestic chip brands increasing by 310% year-on-year, exceeding 820,000 units[7] However, apart from being a "substitute" in mature processes, continuous breakthroughs and innovations in advanced processes are even more important.

Chart 10: The sales of the U.S. semiconductor industry will reach $318 billion in 2024, accounting for 50.4% of global revenue.

Source: SIA, CICC Research Department

► Model: The U.S. still leads in overall quantity and quality, but China has already gained an advantage in the open-source model field. According to Epoch AI's statistics, among the current 976 well-known models, the U.S. has four times as many as China in quantity (632 vs. 156), but in the Artificial Analysis Intelligence Index scoring [8], Chinese large models such as Zhipu GLM, DeepSeek, and Kimi closely follow U.S. large models like ChatGPT, Claude, and Google Gemini in the top ten (Chart 11). China also leads in the download and usage of open-source models; according to Atom Project's statistics [9], the cumulative downloads of Chinese open-source models surpassed those of the U.S. in August 2025 (Chart 12), and by December 2025, over 62% of model derivatives were based on Chinese large models (Chart 13). To some extent, China has leveraged its model advantages to compensate for its shortcomings in chips.

Chart 11: Comparison of major model scores at home and abroad

Source: Artificial Analysis Intelligence Index, CICC Research Department

Chart 12: The cumulative downloads of Chinese open-source models surpassed those of the U.S. in August 2025


Source: The ATOM Project, CICC Research Department

Chart 13: By December 2025, over 62% of model derivatives will be based on China's large models, surpassing the 32% in the United States.

Source: The ATOM Project, CICC Research Department

► Talent Reserve: China's attractiveness to top talent is gradually increasing, with patents exceeding those of the United States. The core driving force behind the continuous development of AI technology is talent. According to the "Global Artificial Intelligence Research Situation Report (2015-2024)" released at the Global Digital Economy Conference in July 2025, AI researchers from the United States and China account for 57.7% of the global total, with the United States leading globally with over 63,000 researchers. The number of researchers in China has increased from less than 10,000 in 2015 to 52,000 in 2024. The rapid growth in the number of talents has strengthened China's research capabilities in the AI field, and by 2022, the number of AI patents held by Chinese researchers had reached three times that of the United States (Chart 14).

Chart 14: By 2022, the number of AI patents held by Chinese researchers had reached three times that of the United States.

Source: Stanford HAI Index Report, CICC Research Department

In summary, the United States started earlier in the field of artificial intelligence, holding a first-mover advantage in computing power and models, and has produced leading companies globally. In contrast, China, benefiting from policy support, a large domestic market, and a talent dividend, is gradually narrowing the gap with the United States in the field of artificial intelligence. The bottleneck for the United States lies in more fundamental infrastructure such as electricity, while China's bottleneck is in the research and development of advanced process chips, with differences in models, especially the limitations of open-source models. This basic pattern also establishes the "differences" in investment orientation between China and the United States.

China-U.S. AI Investment Landscape: Investment intensity is close between China and the U.S., but China is stronger when considering infrastructure; the macroeconomic pull effect is more evident in the U.S

► Technology Equipment Investment: China's nominal investment scale is 60% of that of the United States, but the proportion of GDP is comparable, both at 3.3-3.4%. If we define narrow AI investment as technology hardware + software equipment under GDP, the United States is expected to reach approximately USD 1.05 trillion by 2025, accounting for 3.4% of nominal GDP, an increase of 0.5ppt since 2023. China's equivalent technology hardware + software annualized scale is about USD 650 billion (4.6 trillion yuan), which is equivalent to 60% of the United States, accounting for 3.3% of nominal GDP, similar to that of the United States (Chart 15).

Chart 15: The proportion of U.S. technology investment to nominal GDP in 2025 is 3.4%, close to China's 3.3%

Source: Haver, CICC Research Department

► Investment including infrastructure: China's proportion of GDP is close to 6%, higher than the United States' 4.6%. The AI industry chain is not limited to technology equipment; considering the construction of data centers and power facilities, as well as R&D investment in related industries, the spillover demand for AI in the United States may bring an investment increment of USD 400 billion by 2025, raising the broad AI investment scale to USD 1.4 trillion, accounting for 4.6% of nominal GDP (Chart 16). However, the GDP itemization between China and the United States is not consistent, so we use the increase in computing power scale to measure the positive pull of AI investment on GDP.

Chart 16: The scale of U.S. AI-related investment may rise to USD 1.4 trillion by 2025, accounting for 4.6% of nominal GDP

Source: Haver, CICC Research Department

The China Academy of Information and Communications Technology mentioned in the "Research Report on the Development of Computing Power Economy (2025)" that [11] "According to empirical analysis, a 1% increase in computing power scale corresponds to a 0.425‰ increase in China's GDP." According to IDC data, China's computing power scale is expected to increase by 43% year-on-year by 2025, corresponding to an increase of 2.5 trillion yuan in GDP (accounting for 1.8% of the overall nominal GDP), and China's broad AI investment scale may rise to 5-6% (Chart 17) Chart 17: China's computing power scale will increase by 43% in 2025, corresponding to AI investment scale potentially rising to 5-6%

Source: Haver, China Academy of Information and Communications Technology, IDC, CICC Research Department

► Economic Contribution: The U.S. information technology sector contributes 30% to GDP growth, while China contributes 10%. When measuring the macroeconomic impact of the AI industry in China and the U.S., to avoid the influence of equipment imports and capital goods prices, we still observe the actual value created by the technology industry domestically using the industry value-added (GVA) approach. If we only compare the contribution of the information technology sector to GDP, the U.S. contributed 0.6 percentage points (34% of the total) to the 1.6% real GDP growth in the first half of 2025, while China's information technology industry contributed 0.55 percentage points to the overall cumulative year-on-year growth of 5.2% in the first three quarters of 2025, with a contribution ratio of 10.6%, slightly up from 9.6% in 2023 (Chart 18).

Chart 18: The U.S. information technology sector contributes 30% to overall growth, while China's information technology sector contributes 10% to overall growth

Source: Haver, Wind, CICC Research Department

Funding Source "Differences": U.S. led by the private sector, China driven by both government and private sectors

Although the overall investment scale in China and the U.S. is similar, there are differences in the current development speed and direction of AI industry infrastructure, chip research and development, and model applications. One important reason behind this is the different funding sources for AI investments in China and the U.S., which determine the attributes and behaviors of the funds, such as return expectations, time tolerance, and investment directions. Funding sources are divided into private sector and government sector, with the private sector further divided into funds from publicly listed companies and venture capital.

**Overall, U.S. AI investment is primarily led by the private sector (USD 552 billion), with limited direct government investment (USD 11 billion); while China's private sector investment (USD 90 billion) is only one-sixth of that of the U.S., the intensity of direct government investment and guiding funds is stronger (USD 75 billion) ** Specifically,

► Private Sector: The investment scale in the United States is larger (USD 552 billion), nearly 6 times that of China (USD 90 billion). 1) At the leading company level, the investment scale in the U.S. is nearly 5 times that of China. In the U.S. stock market, we selected the capital expenditures of large-scale cloud service providers directed at the foundational layer, as well as the R&D expenses for chips and large models to measure the investment scale across the entire industry chain (details below), which has already surpassed USD 400 billion by 2025. In the Chinese market, we estimate the overall scale to be around USD 84 billion (Chart 19). 2) In terms of venture capital, the scale in the U.S. is 25 times that of China. According to PitchBook statistics [12], the venture capital in the AI sector in mainland China for 2025 is USD 6 billion, while the venture capital financing related to AI in the U.S. reaches as high as USD 175 billion; even after avoiding double counting and completely excluding expenditures on model layers such as OpenAI and Anthropic, the financing amount still reaches USD 152 billion (Chart 20).

Chart 19: The investment scale of leading companies in the foundational and technological layers in the U.S. is nearly 5 times that of China by 2025

Source: FactSet, CICC Research Department

Chart 20: Venture capital scale in the AI sector in China and the U.S.

Source: FactSet, CICC Research Department

► Government Funding: China's investment scale is larger (USD 7.5 billion), about 7 times that of the U.S. (USD 1.1 billion). The direct funding from the U.S. government is much weaker compared to its private sector. The U.S. government's budget for R&D spending on AI technology has increased from USD 8.2 billion in the fiscal year 2021 to USD 11 billion in the fiscal year 2025. Although the Stargate project was officially announced by Trump in January 2025, the core funding sources are OpenAI and SoftBank, and part of the budget overlaps with the capital expenditures of Oracle and Microsoft, so it is not included in the government investment amount The scale of direct investment by the Chinese government may exceed 500 billion yuan (75 billion USD), with large-scale national-level direct investment in the AI field coming from the National Integrated Circuit Industry Investment Fund Phase III (344 billion yuan) and the National Artificial Intelligence Industry Investment Fund (60.06 billion yuan) as its special sub-fund. In addition, the Ministry of Finance is leading the establishment of a National Venture Capital Guidance Fund with an investment of 100 billion yuan, along with the capital expenditures of the three major telecom operators, roughly estimating government investment to be over 500 billion yuan (75 billion USD).

Investment Direction "Differences": The U.S. invests more in data centers and supporting infrastructure, while China invests more in chips and models

From the perspective of investment direction, the AI industry can be divided into three main sectors: 1) Basic Layer focuses on hardware computing power, covering core hardware such as AI chips, servers, optical modules, as well as data center energy and supporting infrastructure including liquid cooling equipment and power equipment; 2) Technical Layer focuses on technological innovations such as large models and algorithm frameworks; 3) Application Layer serves as the carrier for technology implementation, including vertical solutions in various industries.

► First, from the perspective of private sector investment, 1) Basic Layer focuses on the capital expenditure of cloud vendors in China and the U.S. directed towards infrastructure (U.S.: Amazon, Microsoft, Google, Meta, Oracle, CoreWeave; China: Baidu, Alibaba, Tencent, ByteDance, etc.), as well as the expenses major chip manufacturers allocate for chip research and development (U.S.: NVIDIA, AMD, Broadcom, and Qualcomm; China: Haiguang Information, Cambricon, Moore Threads, Muxi, Huawei, and self-developed chips from Baidu and Alibaba); 2) Technical Layer focuses on the R&D investments of leading large model companies (U.S.: OpenAI, Anthropic, xAI, and Google’s large models; China: Baidu, Alibaba, Tencent, ByteDance, Zhizhu, Minimax, etc.); 3) Application Layer covers various industries, making it difficult to achieve precise separation and statistics.

Overall, if we do not consider application layer investments, by 2025, the investment scale of leading companies in the U.S. in the basic and technical layers will be 5 times that of leading companies in China (40 billion USD vs. 8.4 billion USD), and the expectation for 2026 may further expand.

1) U.S.: Of the 40 billion USD, 88% is directed towards the basic layer (mainly data centers and supporting infrastructure at 83%, with chips accounting for 5%), and 12% is directed towards the technical layer (models). The scale of U.S. basic layer investment is expected to be around 350 billion USD in 2025, an increase of 2.4 times compared to 2022. Among this, the investment scale in data centers and supporting infrastructure is 334 billion USD (83%); chip R&D accounts for a relatively small proportion of the overall investment (5%), and the scale has remained stable at around 15-20 billion USD. The investment scale in the technical layer has rapidly increased since 2022, rising from 4.2 billion USD to 48 billion USD by 2025, with its proportion of the overall investment scale also rising to 12% (Chart 21) Chart 21: 90% of the $400 billion from leading American companies is directed towards the foundational layer, while 10% is directed towards the technological layer.

Source: FactSet, CICC Research Department

2) China: 78% of the $84 billion is directed towards the foundational layer (70% for data centers and infrastructure, with a relatively larger share of 8% for chips), and 22% is directed towards the technological layer (models). The investment scale in China's foundational layer is expected to be around $65 billion by 2025, with $59 billion (70%) allocated to data centers and supporting infrastructure, and $6.4 billion (7.5%) for chip research and development. The investment scale in the technological layer is $19 billion (22%) (Chart 22).

Chart 22: Nearly 80% of the $84 billion from leading Chinese companies is directed towards the foundational layer, while 20% is directed towards the technological layer.

Source: FactSet, CICC Research Department

► Next, looking at government investments, 1) U.S. government funding focuses on foundational research in the technological layer and cutting-edge applications, with a budget of $11 billion primarily directed towards non-commercial foundational research in artificial intelligence, such as new algorithms and AI+. 2) Chinese government funding focuses on foundational layer chip research and "hard technology," such as the National Integrated Circuit Industry Investment Fund Phase III (344 billion yuan), which clearly allocates 70% of its funds to domestic equipment and materials, and 30% to advanced packaging and AI storage, emphasizing heavy asset and ultra-long cycle industries like semiconductor manufacturing, with a duration of 15 years; the National Venture Capital Guidance Fund, with a duration of 20 years, is currently the longest-lasting "patient capital" in the country, adhering to the principle of "invest early, invest small, invest long-term, invest in hard technology" [13].

In summary, private and government investments show that the foundational layer investment proportions in both China and the U.S. are around 87-88%, while the technological layer accounts for 12-13% (Chart 25). The scale of foundational layer investment in the U.S. is $350 billion (from leading companies), while the technological layer investment is approximately $53.5 billion (with $48 billion from leading companies and 50% of the $11 billion government funding), accounting for 87% and 13% respectively (Chart 23); China's foundational layer investment scale is USD 140 billion (leading companies USD 65 billion + three major operators USD 11.4 billion + USD 63.4 billion government funds), while the technological layer investment is approximately USD 19 billion (funded by leading companies), accounting for 88% and 12% respectively (Chart 24).

Chart 23: The overall foundational and technological layer investment proportions in the United States are 87% and 13% respectively.

Source: FactSet, CICC Research Department

Chart 24: The overall foundational and technological layer investment proportions in China are 88% and 12% respectively.

Source: FactSet, CICC Research Department

Chart 25: The foundational layer investment proportions in China and the U.S. are both around 87-88%, while the technological layer accounts for 12-13%.

Source: CICC Research Department

Insights from the Differences in AI Investment between China and the U.S.: U.S. Private Sector Dominance Leads to Return Constraints, Focusing on Infrastructure; China's Higher Government Investment Focuses on Chips and Models

The differences in the AI industry landscape, funding sources, and investment directions between China and the U.S. directly determine the direction of industry development and investment differences, leading to several insights:

1) U.S. AI investment is dominated by the private sector, with the core driving force being the pursuit of commercial returns. In the short term, its mobilization ability in coordinating public infrastructure is not as strong as China's. This is one of the reasons why China has rapidly narrowed the gap in AI infrastructure in recent years under policy support. On the other hand, the characteristic of private capital dominance means that once returns are below expectations or slow, the market can easily fall into concerns about bubbles 2) China's AI investment is government-led, guided by "patient capital" for long-term strategic direction. This model possesses strong resource allocation capabilities, allowing it to tackle and make advanced layouts in "bottleneck" areas such as computing power and chips without regard for short-term profits and losses, demonstrating strong investment resilience. However, its challenge lies in a lower sensitivity to financial returns.

3) U.S. AI investment focuses on data centers and energy-related infrastructure. According to Aerio's statistics, there are currently 628 data centers under construction in the United States. Investors, primarily large-scale cloud service providers, must ensure that their substantial upfront investments do not stagnate due to power supply bottlenecks, as this would directly lead to capital expenditures "going down the drain," resulting in deteriorating financial statements.

4) China's AI investment emphasizes the foundational layer, particularly in chip research and development. From the current state of the AI industry mentioned above, there is still a gap between China and the U.S. in computing power infrastructure, especially in advanced chip manufacturing processes. Based on the current funding directions of enterprises and the government, data centers and "bottleneck" areas are also the main investment directions.

How do the AI industries of China and the U.S. interact? The foundational layer has industrial chain interactions, while chips have a disconnect in domestic demand, and the application layer reflects mutual learning and reference.

The high connectivity and portability of the AI industry, combined with the differences in resource endowments, funding sources, and directions between China and the U.S., lead to industrial chain interactions at the foundational layer (such as the value chain needed for chip manufacturing and data center construction) and a domestic substitution demand for chips due to geopolitical disconnection. However, the application layer is more reflected in the mapping and mutual learning in commercial scenarios.

► The core of the foundational layer interaction is the joint pull of investment from both sides on the demand for related industrial chains. As analyzed above, the U.S. lacks high-density data centers and power infrastructure in the AI chain, as investment in this area will also stimulate demand from competitive Chinese enterprises in the value chain, such as liquid cooling and power equipment. Continuous investment in chips (although not as significant as China's share) will also drive demand for core hardware such as optical modules and PCBs. The same logic applies to China, which also needs to continue investing in data centers, with a larger proportion of investment in chips.

► The connection at the application layer mainly reflects mutual learning of business models. Both sides can provide references based on their practical experiences and business models in similar vertical fields, such as the health AI sector (e.g., Ant Group's Aifu, ChatGPT Health, etc.); the recent discussions in the A-share market about the GEO concept are related to Elon Musk's announcement to open-source the recommendation algorithm of the X platform soon. In the exploration of AI assistant business, the U.S. has personal assistant apps like ChatGPT, while Apple announced deep integration of Google's Gemini model into its ecosystem, and China has Qianwen deeply integrated into the Alibaba ecosystem, evolving towards an AI assistant (Figure 26).

Figure 26: There are industrial chain interactions at the foundational layer between China and the U.S., with a disconnect in domestic chip demand and mapping at the application layer

Note: This image is only a rough schematic and does not represent a complete industry chain.

Source: CICC Research Department

From the perspective of the capital market, 1) companies listed in the basic layer are mainly concentrated in the A-shares and US stocks (US stocks such as Nvidia, Broadcom, Qualcomm, etc.; A-shares such as Moore, Muxi, Cambricon, Haiguang Information, etc.; Hong Kong stocks such as Birun, as well as optical module companies like NewEase, Zhongji Xuchuang, and Tianfu Communication, and also including data centers, liquid cooling, energy storage, power equipment, etc.). The market's expectations for the performance growth of basic layer listed companies in China and the US are relatively higher than those for the technology layer and application layer (Chart 27, Chart 28), mainly due to more certain capital expenditures and demand formed by policy support. From the market performance perspective, the excess returns of the basic layer in China and the US show a high positive correlation, but it is not very stable (Chart 29), partly because the basic layer is more susceptible to trade frictions, and partly due to the high volatility brought about by high expectations and high valuations; 2) Leading listed companies in the technology layer are more in Hong Kong and US stocks, with Hong Kong stocks such as Alibaba, Zhizhu, and Minimax, Tencent, and US stocks such as Google, Meta, etc.; 3) The application layer has a relatively more balanced distribution in A-shares and Hong Kong stocks, not only due to internet platforms but also because the range of vertical application fields is quite broad. From the perspective of excess return correlation, after the launch of DeepSeek, the correlation between the two markets in China and the US has become more pronounced and stable (Chart 30), which is also a result of innovation linkage and mutual learning of business models.

Chart 27: In Chinese assets, the market's expectations for the profit growth of the basic layer are relatively higher.

Note: Industry classification refers to MSCI China secondary industry classification, data as of 2026/1/17.

Source: FactSet, CICC Research Department

Chart 28: The same is true in the US market, where the market's expectations for the profit growth of the basic layer are relatively higher.


Note: Industry classification refers to the S&P 500 secondary industry classification, data as of 2026/1/17

Source: FactSet, CICC Research Department

Chart 29: Although the industrial context shows a positive correlation between the performance of the Chinese and American foundational layers, it is not stable.

Note 1: The calculation of excess returns uses the Wind All A as the benchmark for A-shares, the S&P 500 for U.S. stocks, and the Hang Seng Index for Hong Kong stocks, same on the right;

Note 2: The A-share PCB is depicted by the CITIC PCB Industry Index, and optical modules are depicted by the Wind Optical Module Index;

Source: Wind, CICC Research Department

Chart 30: After the launch of DeepSeek, the assets in the technology and application layers in China and the U.S. show a positive and stable correlation of excess returns.

Note 1: The application layer of A-shares is depicted by the Guozheng AI Application Index, while the U.S. application layer includes Adobe, Salesforce, and Palantir, depicted by the equal-weighted average performance of major leading stocks;

Note 2: The U.S. technology layer includes Google, Microsoft, and Amazon, while the Hong Kong technology and application layers include Tencent, Baidu, Alibaba, SenseTime, Kingdee International, Kuaishou, Meitu, Fourth Paradigm, Alibaba Health, and Kingsoft, depicted by the equal-weighted average performance of major leading stocks.

Source: Wind, CICC Research Department

Looking ahead, the performance of the foundational layer has high certainty of realization, while the potential upward space for the technology and application layers is larger. From the perspective of the credit cycle, the technology chain represented by AI remains the main direction of prosperity at present. Specifically,

► The U.S. is still making large-scale investments in data centers and energy equipment, which will also drive demand in China’s computing power infrastructure (such as optical modules), data centers (such as liquid cooling), and energy infrastructure (such as related power and energy storage equipment);

► China still needs chips, whether from foundational capital expenditures or domestic substitution strategies, thus there remains certainty in demand and policy support in fields such as semiconductors, but the downside is that valuations and expectations are relatively high; ► The technical layer mainly focuses on the technological advancements of large models, and progress in the field of large models from China and the United States may catalyze each other;

► The catalysis on the application side comes from the progress in various industry vertical scenarios. If the C-end business model and demand continue to be realized, related sectors may have greater upward potential, and the overall progress on the application side may also drive the technical layer.

This article is sourced from: CICC Insight

Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk.