In-depth Research Report on the "New Paradigm" of AI Application Entrepreneurship in China

China’s AI application entrepreneurship is undergoing a profound paradigm shift. The “AI Brain + Made in China + Global Vision” model is redefining the valuation logic and investment opportunity framework of the AI industry. According to the latest reports from 36Kr and industry data, the total financing amount in China’s AI application track reached RMB 107.07 billion in 2025, involving 930 companies [1][2]. This model’s core value lies in: by bypassing direct competition with Silicon Valley in terms of computing power, it instead leverages the supply chain advantages of China’s manufacturing industry to establish differentiated competitive barriers at the application layer, thereby creating an investment main line with unique valuation logic.

---

This model was officially proposed by Zhou Qi, Partner of GSR United Capital, at the “AI Spark · Open Mic” forum co-hosted by Alibaba Cloud and 36Kr [2]. Its core framework consists of three mutually supporting dimensions:

: Driven by large models and multimodal AI technology as core capabilities, it provides intelligent “thinking” and “decision-making” functions. Unlike basic model research and development that required massive computing power accumulation in the past, the core of this dimension lies in deeply integrating general AI capabilities with specific scenarios to realize the value release of “small models, large applications”.

: Relying on China’s globally most competitive manufacturing supply chain system, it provides physical carriers and large-scale production capabilities for AI technology. From consumer-grade hardware to industrial robots, from smart wearables to auxiliary medical devices, the cost advantages, response speed, and process accumulation of Made in China provide a solid foundation for the rapid implementation of AI applications.

: Taking the global market space as the ultimate goal of commercialization, it avoids involutionary competition in the domestic market and verifies the feasibility and replicability of the business model on a larger scale.

The rise of this model is not accidental, but the result of the combined effect of multiple structural factors:

First, the

makes direct competition with Silicon Valley in the field of large models increasingly difficult. The launch of NVIDIA’s Rubin chip once again proves that the iteration speed of computing power far exceeds the catching-up ability of application-layer entrepreneurs [2]. Against this background, bypassing computing power competition and instead establishing advantages at the application layer has become a more pragmatic strategic choice.

Second, the

has gained a new channel for value release in the AI era. Whether it is smartphones, drones, or new energy vehicles, Made in China has proven its supply chain integration capabilities and cost control capabilities in multiple fields. This capability can be quickly transferred to the AI hardware field, forming a superimposed effect of “Made in China + AI Intelligence”.

Third, the

provides broad space for Chinese entrepreneurs. Compared with the fierce competition in the domestic market, the global market, especially in fields such as healthcare, education, and auxiliary functions, has a large number of unmet needs, providing incremental space for Chinese AI enterprises with differentiated capabilities.

---

In the past, the valuation logic of AI companies was relatively simple and direct: computing resource reserves, model parameter scale, and technical team background constituted the main supporting factors for valuation. However, this paradigm is facing fundamental challenges.

Zhou Qi, Partner of GSR United Capital, pointed out that almost all NLP companies invested in 10 years ago have become obsolete under today’s Transformer architecture — “all previously trained models and accumulated data no longer exist” [2]. This uncertainty in technological iteration makes valuation models that solely rely on technical barriers fragile.

A deeper problem is that the inference cost of large models is decreasing at a rate of ten times per year, while API call volumes are rising at a rate of ten times per month [1]. This means that the window period for pure technological advantages is shrinking sharply, and the strategy of building a moat by “technological leadership” is difficult to sustain.

Based on observations of industrial changes, a new valuation framework is taking shape, whose core elements include:

: Investors are increasingly valuing enterprises’ ability to adapt to technological iteration, rather than their technological leadership at a certain point in time. Shi Mao, Founding Partner of Changlei Capital, said that the core points of investment are threefold: first, whether the ceiling is high enough; second, relative competitive advantages; third, having a verifiable positive feedback cycle. It is not required for the company to be profitable immediately, but the technology must be gradually verified [3].

: Under the “AI Brain + Made in China” model, manufacturing capabilities have become valuation elements as important as technological capabilities. Investors’ enthusiasm for the embodied intelligence track is a reflection of this logic — they are betting that large models will solve the embodied “brain”, and China’s supply chain will provide the embodied “body” [1].

: The cycle from the birth of an AI company to large-scale profitability has been extremely compressed, and valuation places more emphasis on the company’s short-term cash flow performance and its ability to seize positions in vertical tracks [2]. Ren Penghao pointed out that investment trends have changed, and investors who originally invested in later rounds now hope to invest in early-stage startups [2].

: The success rate of isolated projects has decreased, and projects that can integrate into the industrial chain ecosystem and have cross-border cooperation potential are more valuable for investment. Cloud service providers such as Alibaba Cloud are building AI application ecosystems to provide infrastructure and commercialization channels for startups [2].

The embodied intelligence track ranked first among AI application sub-tracks in 2025 with a total financing amount of RMB 33.77 billion, accounting for 31.5% [1]. The changes in valuation logic of this track are typical:

The high valuations given by investors to embodied intelligence companies are essentially paying for the grand narrative of “AI + Physical World”. Compared with intangible software SaaS, a movable robot seems easier to tell a high-valuation story [1]. However, bubbles have followed — among the 1,000 mass production orders announced by a certain embodied intelligence company, the actual delivery volume was less than 20% [1].

A hard technology investor’s confession is quite representative: “The moat of hard technology is precisely built on these ‘dirty and tiring jobs’. Once the mass production threshold is crossed, latecomers will find it difficult to catch up. Therefore, to some extent, capital is betting not only on robots, but also on the industrial upgrading window of Made in China in the AI era” [1].

---

AI entrepreneurship opportunities are moving towards two extremes:

: There is no longer any opportunity to establish a basic large model startup, as resources are concentrating on leading players [2]. The entrepreneurial window in this field has closed, and giants such as ByteDance and Alibaba have established insurmountable competitive barriers.

: Startups with one person, two people, or five people may instead become new investment hotspots [2]. Projects focusing on vertical niche scenarios and balancing market capacity and verticality are more favored. 2025 financing data shows that 36% of AI application companies are in the early stage (seed, angel round), and 46% are in the growth stage (A-B round), accounting for a total of 82% [1].

Based on 2025 financing data and industrial trends, the following sub-tracks show strong investment value:

: 194 financing cases, accounting for 20.9%, with a total financing amount of RMB 33.77 billion [1]. This track perfectly fits the model characteristics of “AI Brain + Made in China”, and China has obvious leading advantages in the field of embodied intelligence [3].

: Ranked second and third in terms of financing cases, accounting for more than 18% in total [1]. These two scenarios share the characteristics of real demand, strong willingness to pay, and a large number of traditional digital service providers that have transformed to obtain AI concepts, accounting for more than 50% [1]. It should be noted that traditional software companies that have switched to the “intelligent” label by accessing large model APIs have increased financing activity, but the original AI innovation content needs to be carefully screened.

: Including vertical categories such as children’s AI hardware and visually impaired assistive devices. The business logic of this track is clear — the hardware itself can generate profits through scale, and at the same time, the hardware serves as an entry point to provide customized content based on AI’s in-depth understanding of users, realizing paid monetization on the content side [2].

: Falling out of and re-entering the top 10 in total financing amount reflects that investment institutions are willing to place heavy bets on this field for a future [1]. This field has a high concentration of core technologies and talents, and has the potential for “small teams, big breakthroughs”.

: Changes in investment trends have led investors who originally invested in later rounds to turn to early-stage projects, because the cycle from the emergence of an AI company to its establishment, development, and final profitability is too short [2]. No matter how many BPs investors review in a year, they may not be able to capture a good timing.

: In terms of capital, Beijing, Shanghai, and Shenzhen each have a national innovation guidance mother fund of RMB 50 billion, Hangzhou has a RMB 50 billion fund from the National Social Security Fund and AIC, and Suzhou and Wuhan also have RMB 50 billion each. These funds still need to be allocated and expanded, possibly multiplied by 3 times, which means that a large amount of capital will flow into the AI entrepreneurship field in the future [2].

: The success rate of isolated projects has decreased, and investors pay more attention to the collaborative ability of projects in the ecosystem. Cloud service providers such as Alibaba Cloud are building AI application ecosystems to provide infrastructure and commercialization channels for startups, and projects integrated into this ecosystem are more likely to be favored by investors [2].

---

: Build a short drama creation tool with AI as the core, using AI technology to break the limitations of traditional film and television content creation, and realize rapid, efficient, and low-cost content production [2].

: Meet users’ personalized content needs, such as creating content like Archery Junior for niche sports groups, which could not be commercialized due to insufficient audience scale in the past. This is a typical embodiment of the “creating supply” logic — AI applications no longer only “reduce costs and increase efficiency”, but can open up new markets for needs that originally did not exist or could not be met.

: Adopt a revenue model of user recharge and consumption, returning to the essence of business, with organizational capabilities and efficiency as core competitiveness, benchmarking the refined operation ideas of the catering industry [2].

: The case of Juruilu shows that in the field of content generation, the value of AI does not lie in replacing human creators, but in lowering the threshold and cost of content creation, so that long-tail content needs that were originally not commercially viable can be met. The valuation of such projects should focus more on the verification of users’ willingness to pay and the improvement of operational efficiency, rather than purely technical indicators.

: Taking AI hardware as an entry point, build an intelligent agent for the life, study, and entertainment scenarios of post-2010 AI natives, and use AI as a companion and growth tool for children [2].

: Help children shift from “passive input” to “active exploration”. Compared with the fierce competition in the adult AI hardware market, the commercial opportunities in the children’s AI hardware track may be greater — children have higher acceptance of AI, parents have a clear willingness to pay, and the supply of truly high-quality products in the market is insufficient [2].

: The hardware itself can generate profits through scale, and at the same time, the hardware serves as an entry point to provide customized content based on AI’s in-depth understanding of children, realizing paid monetization on the content side [2].

: The case of Tinglixiong reflects the value of in-depth integration of “AI + Scenarios”. Investors should focus on the team’s in-depth understanding of children’s user needs, and the ability to translate this understanding into product functions and commercial value. The valuation of such projects should consider indicators such as user retention rate, payment conversion rate, and user lifetime value.

: Use AI technology to create assistive tools for the visually impaired. Previously, products for the visually impaired could only achieve “image description”, but now with AI, they can achieve precise assistance at the “prescription” level [2].

: Open up a new window of life for the visually impaired, and at the same time, analogously to the path of domestic hearing aid companies breaking the monopoly of foreign brands, it has a clear domestic substitution logic [2].

: Relying on the advantages of China’s manufacturing industry, it can provide high-quality products with more competitive costs, which is particularly important for the disabled assistive device market with high price sensitivity.

: The case of Tongxing Technology reflects the potential of AI technology in creating social value. Investors should focus on the real improvement of users’ lives brought by the product, as well as the resulting market recognition and willingness to pay. The valuation of such projects should consider the brand premium brought by social value and policy support.

---

The technological iteration speed in the AI field far exceeds that of traditional industries, and the disruptive impact of the Transformer architecture on NLP companies is still a lesson to be learned [2]. Investors should pay attention to whether the project has the ability to continuously evolve, rather than its technological leadership at a certain point in time.

: Prioritize startups with rapid learning capabilities, flexible technical architectures, and teams with cross-domain experience; pay attention to the project’s ability to track and integrate the latest technological trends; reserve sufficient safety margins in valuation to deal with uncertainties brought by technological iteration.

The total financing amount of RMB 107 billion in the AI application track in 2025 shows obvious signs of overheating. The case where the actual delivery volume was less than 20% of the 1,000 mass production orders announced by a company in the embodied intelligence field indicates that valuation may be disconnected from actual commercial progress [1].

: Be cautious about projects whose valuations significantly deviate from industry levels or commercial progress; conduct in-depth verification of the project’s commercial data, including core indicators such as actual delivery volume, user retention, and payment conversion; pay attention to whether the project has real competitive advantages, rather than just relying on the AI concept.

From 2023 to 2025, China’s AI industry is basically dominated by giants, which is not friendly to entrepreneurs [3]. There are still some opportunities for startups in the multimodal field, but most tracks are occupied by companies such as ByteDance and Alibaba.

: When investing in startups, carefully evaluate their differentiated capabilities to compete with giants; prioritize vertical scenarios outside the advantageous fields of giants; pay attention to whether the project has unique resource endowments or scenario understanding advantages.

Currently, AI application companies are still in the stage of proving themselves. Most companies are either looking for product-market fit or verifying the sustainability of their business models, and there is still a long way to go before real large-scale profitability [1].

: Have a clear understanding of the investment cycle and prepare for long-term holding; pay attention to the project’s cash flow status and self-hematopoietic ability; set reasonable exit protection mechanisms in investment terms.

---

According to data from IT Juzi and TMTBASE of Titanium Media, as of December 2025, the total number of companies with the label of AI application that received new financing was 930, with a total financing amount of up to RMB 107.07 billion [1]. This means that in 2025, 2.6 companies received financing every day, with an average of RMB 12 million in capital entering the market every hour.

This far exceeds the financing intensity of any single track in recent years, reflecting the strong optimism of the primary market for AI applications.

The number of financed companies in the top 10 scenarios accounts for 73% of the total, indicating a high degree of scene concentration [1]. This means that a consensus has been formed among capital, and funds mainly flow to these ten verified or promising directions.

Rank	Scenario	Number of Financed Companies	Total Financing Amount (100 Million RMB)
1	Embodied Intelligence	194	337.7
2	Industry	-	-
3	Medical	-	-
4	General	-	-
5	Consumer-Grade Hardware	-	-
6	Autonomous Driving	-	-
7	Content Generation	-	-
8	Marketing	-	-
9	Visual Intelligence	-	-
10	Data Governance	-	-

36% of the companies are in the early stage (seed, angel round) of financing, 46% are in the growth stage (A-B round), and the sum of the two accounts for 82% of the total number of companies [1]. The overall proportion of companies in the late stage and strategic investment stage is less than 18%.

This distribution indicates that the vast majority of AI application companies are still in the stage of proving themselves. They are either looking for product-market fit or verifying the sustainability of their business models.

From the perspective of the fund-absorbing capacity of individual projects, autonomous driving and general scenarios are the real winners, ranking first and second with an average single-project financing amount of RMB 450 million and RMB 170 million respectively [1]. Although embodied intelligence has the highest total financing amount, its average single-project financing amount is overtaken.

This reflects that the average financing rounds of autonomous driving and general scenarios are later, and large-scale financings are more concentrated. For example, the largest single financing this year — the US$600 million Series D financing of Neolix Unmanned Vehicles — occurred in the autonomous driving track, which directly raised the average financing amount of the entire track.

---

: This model represents a new direction of AI application entrepreneurship, with strategic value of bypassing computing power competition and establishing differentiated advantages. Investors should focus on startups with comprehensive advantages in manufacturing capabilities, scenario understanding, and continuous evolution capabilities.

: Under the polarization trend of tracks, projects focusing on vertical niche scenarios are more favored. Investors should pay attention to projects that can balance market capacity and verticality, and avoid chasing overly crowded tracks.

: The development cycle of AI companies is short, and early-stage positioning is more likely to capture investment opportunities. Investors should adjust their investment strategies and lay out high-quality projects at an earlier stage.

: The success rate of isolated projects has decreased, and investors should pay more attention to the collaborative ability of projects in the ecosystem. Prioritize projects with potential in ecosystems built by cloud service providers such as Alibaba Cloud.

Looking forward to the next five years, the “AI Brain + Made in China + Global Vision” model is expected to become an important paradigm leading the development of China’s AI industry. With the continuous decline of large model inference costs and the continuous expansion of application scenarios, Chinese AI application enterprises with differentiated competitive advantages are expected to occupy an important position in the global market.

From the perspective of investment returns, although there is a certain risk of valuation bubbles at present, this track still has the potential to generate excess returns in the long run. The key lies in whether investors can identify high-quality targets with real continuous evolution capabilities and commercialization potential among the complex projects.

As Shi Mao, Founding Partner of Changlei Capital, said: “The core of venture capital is not chasing trends, but finding opportunities in differences.” This investment wisdom still applies under the new paradigm of AI application entrepreneurship.

---

[1] Phoenix Net. “107 Billion RMB, 930 Companies: The Savage Consensus of China’s AI Applications in 2025” (https://i.ifeng.com/c/8pfKLis8lqL)

[2] 36Kr. “China’s AI Application Entrepreneurs are Changing Tracks to Lead | AI Spark Open Mic” (https://m.36kr.com/p/3641730074300038)

[3] Huxiu. “When Everyone is Talking About AI, What Opportunities Are Left for Entrepreneurship and Investment?” (https://www.huxiu.com/article/4817751.html)