In the global arena of artificial intelligence, the United States and China are engaged in an unprecedented technological showdown. On the surface, the US holds the most advanced AI hardware technology and attempts to limit China’s development through export controls. However, a deeper analysis reveals that China is demonstrating surprising resilience and innovation through multiple pathways including talent advantages, patent accumulation, technological innovation, and open-source strategies.
America’s Hardware Hegemony and Control Strategy
Escalating Chip Export Controls
The United States indeed possesses significant advantages in AI hardware, particularly in high-end AI chips. In April 2025, the US government further tightened AI chip export restrictions to China, including the Nvidia H20 chip specifically designed for the Chinese market in the control list. This decision is expected to cost Nvidia $5.5 billion in revenue, demonstrating America’s determination to maintain its technological advantage.
The new guidance issued by the US Department of Commerce’s Bureau of Industry and Security (BIS) clearly states that using Huawei’s Ascend chips anywhere globally could violate US export control regulations. This “long-arm jurisdiction” approach reflects America’s strategic intent to maintain its leadership position in AI through technological blockade.
Allied Cooperation and Technology Barriers
The Trump administration abolished the Biden-era “AI Diffusion Framework,” instead adopting country-by-country negotiations to determine chip export levels. This strategy aims to improve diplomatic relations with countries like Saudi Arabia and the UAE while ensuring advanced AI technology doesn’t flow to adversary nations.
China’s Multi-dimensional Competitive Advantages
Talent Advantage: Half of Global AI Research Force
China possesses remarkable advantages in AI talent. According to Nvidia CEO Jensen Huang’s speech in Washington D.C., “50% of AI researchers globally are Chinese.” This data reveals China’s deep foundation in AI talent reserves.
More notably, Chinese AI companies like DeepSeek primarily rely on domestically trained talent. Although about 25% of DeepSeek researchers gained experience in the United States, they ultimately chose to return to China, creating a one-way knowledge transfer that further strengthens China’s AI ecosystem.
Patent Accumulation: Leading in Quantity but Quality Needs Improvement
China demonstrates overwhelming advantage in AI patent applications. In 2024, China filed 300,510 AI-related patents, accounting for 70% of the global total, while the US filed only 67,773, accounting for 15.8%. By the end of 2023, China’s effective AI invention patents reached 378,000, with an annual growth rate exceeding 40%, 1.4 times higher than the global average.
However, there remains a gap in patent influence. US AI patents have a citation rate about 7 times that of China (13.18 vs 1.90), indicating the US still maintains an advantage in patent quality and influence.
Technological Innovation: The “Quantity Over Quality” Distributed Computing Strategy
Breakthroughs in Distributed AI Training
Facing the challenge of high-end chip shortages, China has innovatively developed distributed AI training technology. China Unicom successfully implemented distributed AI model training across 300 kilometers, achieving training efficiency above 97%. In December 2024, the company stored 30TB of data in Jinhua and performed actual computation in Hangzhou 200 kilometers away, demonstrating the feasibility of distributed training.
This strategy of “if one can’t win, combine ten together” is redefining the possibilities of AI computing. China’s approach fully utilizes existing network infrastructure, distributing computing power across multiple locations to break through single-chip performance limitations.
Strategic Layout of Supercomputing Centers
China currently has 14 national-level supercomputing centers with total intelligent computing power reaching 5,000 petaflops. The fully domestically-produced supercomputing center built by China Telecom in Wuhan can train trillion-parameter large language models, with performance comparable to America’s Frontier supercomputer.
Zheng Weimin, academician of the Chinese Academy of Sciences and Engineering, stated that through coordinated hardware and software design, these resources can be used for large language model training and inference at only one-sixth the cost of renting Nvidia chips.
Cost Revolution: Knowledge Distillation and Open-Source Strategy
Disruptive Reduction in Training Costs
Chinese AI companies have achieved breakthrough progress in reducing model training costs. DeepSeek-V3’s training cost was only $5.6 million, while GPT-4’s training cost is estimated between $50-100 million. This cost difference reaches an order of magnitude.
In 2024, many Chinese AI companies began seeking to cut AI model training costs, with some developers reducing related costs by up to 80%. For example, Jijia Technology’s visual model training cost dropped from $5 million in 2023 to under $1 million in 2024.
The Controversy of Knowledge Distillation Technology
Chinese companies widely adopt knowledge distillation technology, “learning” from America’s leading AI models and creating lower-cost alternatives. US AI and cryptocurrency chief David Sacks pointed out, “There’s substantial evidence that what DeepSeek did was distill knowledge from OpenAI’s models.”
OpenAI stated in a declaration that companies in China and other countries “constantly attempt to distill models from leading US AI companies.” This technical approach allows China to achieve performance comparable to top US models at relatively low cost.
Strategic Significance of Open-Source Strategy
Chinese AI companies are adopting open-source strategies to expand their influence. DeepSeek’s R1 model was released under the MIT license, completely open for use. Baidu announced it would open-source its ERNIE 4.5 and ERNIE X1 models, with Alibaba and Tencent joining this trend.
This open-source strategy contrasts sharply with the proprietary approach of US companies like OpenAI and may redefine the global AI development landscape.
America’s Response Challenges and Strategic Dilemmas
Multiple Measures at the Policy Level
Facing the rapid rise of China’s AI capabilities, the US government has adopted multiple response measures. Congress has launched its first investigation into Nvidia’s business, assessing whether the company intentionally provided key technology to DeepSeek. Federal agencies including NASA, the Pentagon, and the Navy have banned the use of DeepSeek on government equipment.
The Trump administration is also considering punitive measures against DeepSeek, blocking its purchase of US technology, and discussing banning Americans from accessing its services.
Limitations of Technology Blockade
However, the technology blockade strategy faces fundamental challenges. China is effectively bypassing hardware limitations through innovative pathways such as distributed computing, knowledge distillation, and open-source strategies. As one analyst noted, “Necessity is the mother of invention” – US sanctions have actually stimulated Chinese technological innovation.
Uncertainty in Long-term Competition
US intelligence reports indicate that China aims to surpass the US in AI by 2030. The Chinese People’s Liberation Army may use large language models to create disinformation, impersonate identities, and facilitate cyberattacks. This strategic competition has transcended the purely technical level, involving national security and global influence contests.
Conclusion: Technology Democratization and Competitive Landscape Reshaping
The US-China AI competition is redefining the rules of global technological development. While the US still maintains advantages in hardware technology, China is building a more democratized and accessible AI ecosystem through talent aggregation, cost innovation, diversified technical routes, and open-source strategies.
The ultimate outcome of this competition will not only determine the two countries’ positions in AI but also influence the direction of global technological development. The game between open-source and proprietary, cost efficiency and technological leadership, hardware limitations and software innovation is shaping a more complex and diverse AI future. The US needs to rethink its technology hegemony strategy, while China’s innovation model also provides new pathways for technological development in other countries.
In this war without gunpowder, technological innovation, talent cultivation, and open cooperation may be more decisive than simple technology blockade in determining the final outcome.