If you have deeper insights or opinions on this article, please contact the author:Zhang Surong, General Manager of Investment Department 4, Nanjing Innovation Investment GroupEmail: zhangsr@njicg.com
01 Brief Analysis of Key Development Points in the Industry Track
(I) Industry Development Context
As the core driving force of a new round of technological revolution and industrial transformation, Artificial Intelligence (AI) traces its origins back to the 1950s. From early symbolic logic reasoning, to the rise of expert systems in the 1980s, and then to breakthroughs in machine learning—especially deep learning—at the beginning of the 21st century, AI has undergone a tortuous evolution of "three rises and two falls".
In recent years, with the sharp decline in computing costs, the accumulation of massive data, and continuous optimization of algorithm models, AI has entered its "third wave", accelerating from laboratories to industrial applications.
Currently, AI is transitioning from perceptual intelligence to cognitive intelligence, and gradually integrating cutting-edge technologies such as large models, generative AI (AIGC), and multimodal interaction, forming a technological evolution path with Artificial General Intelligence (AGI) as the long-term goal. This context has not only reshaped the technological paradigm but also profoundly influenced the global tech competition landscape and industrial ecosystem restructuring.
(II) Industry Development History
The development of China’s AI industry can be roughly divided into three phases:
Phase 1 (2010–2016): Technological Embryo and Preliminary Exploration
This stage was dominated by breakthroughs in single-point technologies such as speech recognition and image processing. Representative enterprises including iFlytek and SenseTime began to emerge. The state had not yet formed systematic supporting policies, but the capital market had started to focus on AI startups.
Phase 2 (2017–2020): Policy-driven and Large-scale Implementation
In 2017, the State Council issued the New Generation Artificial Intelligence Development Plan, clearly positioning AI as a national strategy and promoting its deep integration with the real economy. Local governments established AI industrial parks, capital flooded in, and application scenarios expanded from security and finance to medical care, manufacturing, education, and other fields.The number of AI enterprises surged, but homogeneous competition was severe, and some enterprises exited due to insufficient commercialization capabilities.
Phase 3 (2021–Present): High-quality Development and Ecosystem Construction
After a "bubble squeeze", the industry entered a rational development stage. Large models became the new focus, with major releases such as Baidu ERNIE, Alibaba Tongyi, Huawei Pangu, and ByteDance Doubao, marking China’s AI entry into an era of "platform + ecosystem" competition.Meanwhile, the state strengthened institutional frameworks for data security and algorithm supervision to guide the healthy and orderly development of AI.
(III) Market Size
According to IDC, iResearch, and other institutions:
The global AI market exceeded $500 billion in 2025 and is projected to surpass $1.5 trillion by 2027, with a CAGR of over 30%.
China’s market reached approximately ¥450 billion in 2025, accounting for about 10% of the global total, but growing significantly faster than the world average. It is expected to exceed ¥800 billion by 2027.
Among segments, AI chips, large model platforms, intelligent driving, AIGC content generation, and industrial AI quality inspection are growing particularly rapidly.Since ChatGPT sparked global attention in late 2022, the generative AI market has expanded by over 300% within a year, becoming the most explosive sub-track.
(IV) Industrial Chain
The AI industrial chain is divided into three layers:
1. Foundation Layer
Including AI chips (e.g., Cambricon, Horizon), computing infrastructure (cloud computing, data centers), data resources, and annotation services.This layer features high technical barriers and capital intensity, serving as the foundation of the entire ecosystem.
2. Technology Layer
Covering algorithm frameworks (e.g., TensorFlow, PyTorch), large model training platforms, computer vision, Natural Language Processing (NLP), speech recognition, and other core capabilities.Leading tech companies dominate this layer.
3. Application Layer
Embedding AI capabilities into specific industry scenarios, such as intelligent customer service, smart healthcare, autonomous driving, intelligent manufacturing, and fintech.This layer is close to end-users with clear commercialization paths, but requires in-depth industry know-how.
Notably, with the popularization of large models, the Model-as-a-Service (MaaS) model is blurring the boundary between the technology and application layers, driving the industrial chain to evolve toward an integrated "cloud + device + model" structure.
(V) Constraints
Despite broad prospects, the AI track faces multiple challenges:
1. Technical Bottlenecks
Essentially, current mainstream large language models (GPT series, Gemini, LLaMA, etc.) rely on probabilistic pattern matching and sequence prediction. They still suffer from hallucinations, weak reasoning, and high energy consumption. AGI is far from realization, and most applications still depend on scenario-specific fine-tuning.
2. Data and Privacy Risks
High-quality training data is scarce, involving user privacy and copyright disputes (e.g., legitimacy of AIGC training data sources).
3. Computing Dependence and Cost Pressure
Training trillion-parameter models costs tens of millions of dollars, unaffordable for SMEs. Domestic AI chips still lag behind NVIDIA in performance and ecosystem.
4. Regulatory Uncertainty
New rules on AI ethics, algorithm transparency, and deepfakes (e.g., EU AI Act) may increase compliance costs.
5. Difficult Commercialization
Many AI projects are popular but unprofitable, with unclear ROI and low customer willingness to pay, especially slow penetration in traditional industries.
02 Fund Investment Strategy
(I) Track Investment Status
In recent years, AI has become a shared hotspot in both primary and secondary markets.According to Zero2IPO Data, there were over 1,200 financing events in China’s AI sector in 2023, with disclosed amounts exceeding ¥200 billion, mainly concentrated in four directions: large models, AI chips, autonomous driving, and AIGC.Top VC/PE firms such as Sequoia, Hillhouse, and Qiming Venture Partners have all launched dedicated AI funds.
In the secondary market, AI concept stocks have fluctuated sharply.Boosted by ChatGPT, related stocks rose collectively in 2023; however, in 2024–2025, as performance fell short of expectations, some targets corrected significantly.This reflects the market shifting from "concept speculation" to "fundamental verification".
Notably, state-owned funds (e.g., National Integrated Circuit Industry Fund, local guidance funds) are increasing investment in the AI foundation layer, especially supporting domestic substitution projects, reflecting a strategic orientation of "security and controllability".
(II) Investment Risk Factors
Although high-growth, the AI track carries notable risks:
Technological iteration risk: AI evolves extremely rapidly; leaders today may be eliminated tomorrow (e.g., Transformer replacing RNN). Betting on a single technical route can lead to heavy losses.
Valuation bubble risk: Some startups obtain excessive valuations under the "large model" label but lack sustainable revenue models, facing downward pressure.
Policy and compliance risk: Unclear rules on cross-border data, algorithm filing, and liability for generated content may disrupt operations.
Talent competition risk: Scarcity of top AI talent drives up labor costs via high salaries from giants, reducing stability for small teams.
Geopolitical risk: U.S. export controls on high-end chips directly impact domestic AI computing power; supply chain security has become a critical variable.
(III) Investment Layout Strategy
Facing a complex environment, the fund should adopt a strategy of "layered focus, combination of short and long term, and ecological synergy":
Layered Layout
Foundation Layer: Focus on hard-tech projects such as domestic AI chips, photonic computing, and in-memory computing, emphasizing independent and controllable technology.
Technology Layer: Watch vertical-domain large models (legal, medical, financial) and model compression & inference optimization.
Application Layer: Select scenarios with clear payers, high repurchase rates, and strong industry barriers, such as industrial inspection, intelligent O&M, and actuarial insurance.
Stage Balancing
Early stage (Angel/Pre-A): Bet on disruptive technologies, tolerate high failure rates.
Growth stage (B/C rounds): Focus on product validation and customer expansion.
Mature stage (Pre-IPO): Emphasize profit models and listing paths.
Ecological SynergyCooperate with industrial capital (e.g., automakers investing in autonomous driving, hospitals in AI diagnosis) to empower portfolio companies via "capital + scenarios" and accelerate commercial closed loops.
Global VisionUnder compliance, monitor overseas open-source AI communities and emerging market applications (e.g., digital governance in Southeast Asia) to diversify regional risks.
(IV) Investment Project Selection Criteria
In project screening, focus on:
Technical barriers: Original algorithms, patents, or unique datasets; ability to build a 6–12 month technological moat.
Team DNA: Core team with top academic background + engineering implementation capabilities + successful startup or leading AI project experience.
Business model: Clear revenue streams (SaaS subscription, API calls, project-based); customer LTV significantly higher than CAC.
Strategic alignment: Consistency with national "new quality productivity" and integration into industrial ecosystems or fund portfolio synergy networks.
Beware of "pseudo-AI" projects that use AI only as a marketing gimmick with low technical content and high substitutability.Truly valuable AI companies can significantly improve efficiency, reduce costs, or create new value through technology.
03 Conclusion
AI is evolving from "correlation" to "causation", and from "statistics" to "modeling".The track stands at the intersection of "technological explosion" and "commercial validation".
In the short term: large models and AIGC will continue to lead innovation.
In the medium term: deep integration of AI and industries will spawn a trillion-yuan market.
In the long term: breakthroughs in AGI may reshape human society.
For investors, it is essential to maintain insight into technological trends while adhering to value investing, avoiding "technology-only" or "trend-only" mindsets.Only by understanding the essence of technology, respecting industrial laws, and controlling risk bottom lines can we achieve steady progress and long-term returns in this century-defining transformation of AI.
Source: Zhang Surong, General Manager of Investment Department 4
Reviewed by: Xue Yao
Released by: You Yi
