From 'Completely Following' to 'Partially Leading': China's Leapfrog AI Development and Industry Empowerment
Abstract
Artificial intelligence, as the core driver of the new round of technological revolution and industrial transformation, is profoundly reshaping global economy and society. In recent years, China's AI industry has shown strong momentum: domestic large language models like DeepSeek, Tongyi Qianwen, and Doubao have ranked among the global leaders in multiple key indicators; a batch of domestic chip ten-thousand-card clusters have been built; AI-related industry scale has exceeded one trillion yuan; and the overall AI penetration rate in key industries has surpassed 80%. These landmark achievements mark that China's AI technology has moved from "completely following" to "completely parallel running with partial leading," deeply empowering all industries and injecting strong new momentum into China's high-quality economic development. This article systematically reviews the leapfrog progress of China's AI development since the 14th Five-Year Plan from four dimensions: technological breakthroughs, infrastructure, industry scale, and application penetration, and looks forward to future development paths.
I. Introduction: China's Coordinates in the AI Era
Artificial Intelligence (AI) has experienced multiple booms and winters since its birth at the 1956 Dartmouth Conference. Entering the second decade of the 21st century, with breakthroughs in deep learning, declining computing costs, and accumulation of big data, AI ushered in its third wave of development. In this global technology competition, China, with its huge market size, abundant data resources, sustained policy support, and growing innovation capability, has gradually transformed from a follower to a leader.
The 14th Five-Year Plan clearly stated the need to "target frontier areas such as AI, quantum information, and integrated circuits, and implement a number of forward-looking and strategic national major science and technology projects." Over the past five years, China's AI field has achieved remarkable results: from basic theory to key technologies, from hardware support to software ecology, from model development to scenario implementation, forming a complete innovation chain. Domestic large language models like DeepSeek, Tongyi Qianwen, and Doubao have repeatedly achieved excellent results in international authoritative evaluations; a batch of domestic chip ten-thousand-card clusters have broken computing bottlenecks; the AI industry scale has exceeded one trillion yuan; and the penetration rate has exceeded 80%—these figures are vivid footnotes to China's AI march from "running alongside" to "leading."
Caption: China's AI industry ecosystem map shows the complete industrial chain layout from basic layer to technology layer and application layer
II. Technological Breakthroughs: Domestic Large Models Join World's First Tier
Large Language Models (LLMs) are the strategic high ground of current AI technology competition. Since 2023, international large models like ChatGPT have attracted global attention; Chinese tech companies quickly followed and achieved differentiated breakthroughs. DeepSeek, with its efficient MoE (Mixture of Experts) architecture and innovative training methods, has reached internationally leading levels in multiple natural language processing benchmarks; Alibaba's Tongyi Qianwen series models excel in understanding, reasoning, and generation, and are the first to open-source, promoting industry collaborative innovation; ByteDance's Doubao model forms unique advantages in multimodal interaction and personalized services.
The rise of these domestic large models is no accident. On one hand, China has the world's largest internet user base and rich Chinese corpus resources, providing natural advantages for model training; on the other hand, domestic research teams have continuously broken through in model architecture optimization, training efficiency improvement, and inference cost control, achieving "better results with less computing power." For example, DeepSeek has reduced training costs to one-tenth of comparable international models through algorithmic innovation while maintaining leading performance, a result accepted by the top international conference NeurIPS, demonstrating China's original capability in AI underlying technology.
More importantly, domestic large models are transforming from "usable" to "good to use." In areas such as Chinese understanding, cultural adaptability, and domain knowledge integration, domestic models show more precise localization capabilities than international models. For example, in vertical fields like law, healthcare, and finance, fine-tuned Tongyi Qianwen and Doubao models can provide professional-level Q&A services, laying the technical foundation for their large-scale deployment.
III. Infrastructure: Domestic Chip Ten-Thousand-Card Clusters Build Computing Foundation
AI development relies on strong computing power support. In the past, high-end AI training chips were heavily dependent on imports, a bottleneck constraining China's AI development. In recent years, driven by policy guidance and market forces, domestic chip manufacturers have accelerated their rise; the AI chip performances of companies like Cambricon, Huawei Ascend, and Haiguang Information have continuously improved. Even more encouraging, a batch of ten-thousand-card clusters based on domestic chips (i.e., computing clusters connecting tens of thousands of AI accelerators) have been built and put into operation, marking that China has achieved self-controllable large-scale computing infrastructure capabilities.
These ten-thousand-card clusters use self-developed interconnection architecture and distributed training frameworks, achieving efficient data parallelism and model parallelism. For example, a national-level intelligent computing center deployed a ten-thousand-card cluster using Ascend 910B chips, achieving internationally advanced linearity and acceleration ratios in training tasks with hundreds of billions of parameters, reducing single training costs by over 30% compared to traditional solutions. This breakthrough not only ensures self-control of core technology but also provides continuous computing support for the iterative upgrade of domestic large models.
Additionally, progress has been made in the centralized scheduling of computing resources. Through unified computing scheduling platforms, various intelligent computing centers have achieved cross-regional and cross-industry sharing of computing resources, effectively improving resource utilization efficiency. The advancement of national strategies such as the National Integrated Big Data Center System and the East-West Computing Transfer Project has further optimized computing layout, laying a solid foundation for large-scale AI applications.
IV. Industry Scale: Rise of a Trillion-Yuan Market and Structural Optimization
According to industry statistics, China's AI core industry scale has exceeded 1.2 trillion yuan in 2025, driving related industries to exceed 3.5 trillion yuan, with an average annual growth rate of over 25%. Behind this figure is the full acceleration of AI from lab to market. By subsector, computer vision, speech recognition, and natural language processing still dominate, but the rise of new tracks like generative AI, AI for Science, and embodied intelligence is reshaping the industry landscape.
Generative AI is the fastest-growing field. With open APIs and commercialization of models like DeepSeek and Tongyi Qianwen, applications such as AI writing, AI painting, AI coding, and AI office assistants have rapidly spread. In the first half of 2025 alone, domestic AIaaS market revenue exceeded 80 billion yuan, up 45% year-on-year. Meanwhile, deep integration of AI with scientific research gave rise to a new paradigm of "AI for Science," achieving breakthrough results in drug R&D, materials design, and weather prediction, providing new tools for China's scientific and technological self-reliance.
Structural optimization is also reflected in the diversification of enterprise entities. Besides traditional tech giants, a number of AI startups focusing on vertical scenarios have grown rapidly, forming a healthy ecosystem of large enterprises leading and SMEs innovating. Capital market enthusiasm for AI investment continues to be high, with total AI financing exceeding 300 billion yuan in 2025, of which over 60% goes to underlying technology and application scenarios.
V. Application Penetration: From Pilots to Full Embrace
The release of the data "overall AI penetration rate in key industries exceeds 80%" marks that AI has transformed from a "nice-to-have" pilot application to a "must-have" rigid demand. In manufacturing, AI-driven intelligent quality inspection systems improve defect detection accuracy to over 99.9%, while reducing labor costs by 80%; in healthcare, AI-assisted diagnosis systems cover over 300 tertiary hospitals, with recognition accuracy for lung nodules, fundus diseases, etc., reaching expert level; in finance, intelligent customer service, risk control models, and investment research assistants have become standard for financial institutions, improving service efficiency by over five times.
Notably, AI also plays an important role in public services and government services. Smart city platforms use AI algorithms for traffic flow prediction, public safety early warning, and environmental monitoring scheduling, making urban governance more refined; government large models provide instant policy consultation and service guidance for citizens, significantly improving administrative efficiency. Additionally, innovative AI applications are emerging in agriculture, education, energy, cultural tourism, and other industries, forming a thriving "all industries + AI" landscape.
High penetration depends on open platform ecosystems. Major domestic cloud service providers have launched AI development platforms, providing one-stop services from data annotation, model training to deployment, greatly lowering technical thresholds for SMEs and industry users. According to estimates, AI technology applications will contribute about 2.3 percentage points of growth to China's GDP in 2025, becoming a key force in driving domestic demand and promoting industrial upgrading.
VI. Challenges and Future Outlook
Despite remarkable achievements, China's AI development still faces many challenges. First, the "bottleneck" problem of core technologies is not fundamentally solved; gaps remain with international top levels in high-end chip manufacturing, basic software frameworks, etc. Second, the data governance system is not yet perfect; data silos, privacy security, ethical norms, etc., need urgent resolution. Third, there is a structural shortage of AI talent, especially top theoretical innovators and interdisciplinary compound talents. Fourth, international competition is intensifying; global AI governance rule games are complex, and China needs to find a balance between open cooperation and self-reliance.
Looking forward, China's AI development should focus on the following directions: first, continue to increase investment in basic research, achieving original breakthroughs in frontier areas like explainable AI, common sense reasoning, and multimodal fusion; second, accelerate the construction of a self-controllable computing ecosystem, closing the full-loop of chip, framework, model, and application; third, improve the data factor market mechanism, promoting data circulation and sharing under the premise of ensuring security; fourth, build an AI ethical governance system with Chinese characteristics, ensuring technology for good and safe controllability; fifth, deepen international cooperation, actively participate in global AI governance rule-making, contributing Chinese wisdom and Chinese solutions.
VII. Conclusion
From "completely following" to "completely parallel running with partial leading," China's AI has achieved a historic leap in just a few years. Behind the rise of domestic large models like DeepSeek, Tongyi Qianwen, and Doubao, the construction of ten-thousand-card clusters, the achievement of a trillion-yuan industry scale, and a penetration rate of over 80% lie the wisdom and hard work of countless researchers, entrepreneurs, and policymakers. Standing at a new historical starting point, we have reason to believe that as long as we maintain strategic focus, adhere to innovation-driven development, and strengthen ecological synergy, China's AI will surely move from "partial leading" to "full leading," making greater contributions to global technological progress and the development of human civilization.
