By betting on AI, China hopes to develop cost-effective and practical business tools for global markets, especially low-income countries.When American technology entrepreneur Peter Thiel's book "Zero to One" was published in China in 2015, it hit upon an insecurity in the hearts of many Chinese people. Thiel believes thatAlthough China has done a great job in expanding and commercializing emerging technologies, its real innovation is not as good as that of the United States.——Real innovation here means creating something completely new from scratch. Take the iPhone as an example: Engineers in Cupertino, California are responsible for design, and workers in Shenzhen, China are responsible for manufacturing.

For more than a decade, China's decision-makers have been trying to shake off this impression, integrating the pursuit of innovation into national industrial policies such as Made in China 2025. Some early results are already showing. For example, in 2023, Shenzhen technology company Huawei launched the Mate60, a smartphone made using locally produced chips. This was seen as a symbolic breakthrough,Proving that China can still manufacture advanced semiconductors even in the face of severe U.S. sanctions on core equipment and high-end design software. The recent release of DeepSeek-R1—a large Chinese language model built at a fraction of the cost of its Western counterparts—has caused an uproar in the U.S. tech world.

Famous venture capitalist Marc Andreessen calls it“AI’s ‘Sputnik’ moment”——The allusion is that the US-Soviet space race began after the Soviet Union launched the world's first satellite "Sputnik" in the mid-20th century. Using DeepSeek through an application programming interface (API, a protocol that connects different software applications), the price is only 1/13 of a similar model developed by California-based OpenAI.

The emergence of DeepSeek shines the spotlight on China's artificial intelligence (AI) ecosystem——And the operating model here is very different from that in Silicon Valley.. Although user-oriented applications have received a lot of attention, Chinese AI companies, unlike American AI companies, are actually more focused on how to solve large-scale industrial and manufacturing problems.This difference mirrors the differences in the drivers of innovation in the two countries: venture capital in the United States, and large manufacturing companies and state institutions in China.

Technology diversion

The root of this difference is China's comparative advantage in the global economy - manufacturing - and the fact that the government is the largest customer of new technologies. The Chinese government hopes to develop low-cost, scalable AI applications to help modernize this rapidly developing country.

At a regular press conference in September last year, Foreign Ministry spokesperson Lin Jian said that China has always placed scientific and technological innovation at the core of national development. In response to the government's call, various AI research laboratories are working hard to create practical applications - making trains on time, monitoring fish stocks, and providing automated telemedicine services.

Beijing-based company ZhipuAI works with governments and state-owned enterprises to deploy its agent models to automate tasks such as form filling and financial report analysis. In January, Alibaba, the technology giant headquartered in Hangzhou, and 01.AI, a Beijing-based start-up company, established the "Industrialization Large Model Joint Laboratory" to focus on accelerating the commercial and industrial implementation of AI.

The role of AI in China’s energy transformation cannot be ignored, from large-scale pilots of integrated smart homes to huge investments in the national smart grid (approximately US$800 billion).

Therefore, the Chinese government’s goal is not necessarily to lead the world in the field of AI conversational robots, but to use the underlying technology to create affordable and commercially viable solutions. Its applications can be exported to other regions, especially low-income countries. in other words,China’s goal is not necessarily “cutting-edge AI” but “mass-market AI.”China’s new AI blueprint echoes its approach to other technologies, such as electric vehicles and clean energy:Don’t be the first to innovate, but be the first to make technology affordable and widely available.

Forced

A technology stack refers to a set of interrelated resources used to develop advanced AI models, including hardware such as semiconductors, cutting-edge learning algorithms optimized for that hardware, and a backend that includes energy-intensive data centers and predictable financial flows.

In order to secure its global dominance in AI technology, the United States has imposed export sanctions on key components from time to time.On October 7, 2022, the administration of former U.S. President Biden issued a series of export control measures on advanced computing and semiconductor manufacturing components, aiming to prevent China from purchasing high-performance chips from companies such as California-based NVIDIA. Then-national security adviser Jake Sullivan called it"High wall of small courtyard"(smallyard, highfence) strategy: The United States will build a "wall" around core AI technology and even encourage companies in allied countries such as the Netherlands and South Korea to restrict exports to China.

At the same time, China sees the importance of upgrading its industrial chain for economic development, especially in areas such as aircraft and semiconductors. From this perspective, this kind of competition requires China’s efforts to be more urgent and vigorous. If the United States masters future technologies and insists on using export controls, China faces the risk of economic stagnation and subsequent political instability. The fight for AI hegemony is actually part of the entire geopolitical matrix.

To deal with this situation, China has already made preparations. In 2019, the United States included Huawei on the Entity List, a list of trade restrictions issued by the U.S. Department of Commerce. This has led China to rethink how it can be less at the mercy of U.S. export controls.The Ministry of Education of China has proposed multiple integrated research platforms (IRPs), through major institutional reforms to help the country develop in key areas vulnerable to U.S. sanctions or export controls, including robotics, driverless cars, and AI. There are currently 30 large integrated research platforms.

The new-generation integrated circuit technology integration research platform of Fudan University in Shanghai is a national-level scientific research team that is expected to promote breakthroughs. The 2022 export controls target chips with a "node" (the smallest unit of a semiconductor) of 14 nanometers or below. The smaller the chip node, the more transistors integrated in the same area can improve performance and efficiency. As early as 2021, Fudan's new generation integrated circuit technology integration research platform has launched positions to engage in 3-4 nanometer node research. Today, such high-performance chips are the driving force of the AI ​​technology stack. To resolve manufacturing bottlenecks,Last year, China launched the third phase of its “Big Fund” and plans to invest US$47 billion in the semiconductor ecosystem.The National Big Fund is a state-backed investment scheme that pools resources from public enterprises and local governments.

This shows China’s determination to localize its AI capabilities by investing a large amount of institutional, academic and scientific research resources. The U.S. policy of restricting China's access to chips has forced China to create conditions and find unconventional methods. Many of these attempts don't necessarily come to fruition, but it doesn't take a lot of success to have a global impact.

The integrated research platform has become an ideal platform for training engineers, filling the gap in the talent pool that existed even ten years ago.DeepSeek's instant fame is, to a certain extent, the result of years of investment.

the road ahead

In which direction will China's AI ecosystem develop in the future?Although the United States may still have a "stuck" advantage in the chip field in the short term, China is more prepared in other aspects: For example, there is growing demand for data centers that consume large amounts of electricity. The National Integrated Computing Power Network, as part of a major infrastructure project launched in 2021, plans to build large-scale data centers in western China, where land and electricity are relatively cheap.

China’s unique economic and political ecology will lead AI to develop along a specific path. China still faces development problems, so leveling medical resources, accurately predicting weather patterns, and managing industrial logistics are more pressing challenges than developing AI conversational robots and digital companions, especially given various compliance requirements on sensitive topics.

China’s huge consumer market and manufacturing base may promote the development of “embodied AI”, such as AI-driven robots, self-driving cars and industrial equipment. For example, when Hangzhou company Unitree Robotics' dog-shaped quadruped robots were used to repair infrastructure—using high-definition cameras to reach specific checkpoints and robotic arms to carry out repairs—they received coverage and praise from Xinhua, the state news agency.

"software(in China)Basically no money is made; people believe that all software should be open source. The combination of software and hardware can be profitable. "A machine learning researcher at Peking University said, "So, AI-driven robots may become the trend of the future in China."

In the field of language models, China's next focus may be to identify a series of products that can be exported to other emerging economies.Benjamin Bratton, a philosophy of technology researcher at the University of California, San Diego, said, "Essentially, China can provide an AI stack in a box. This is your power plant; this is your base model; this is the stack architecture for the application system - take it to India, take it to Brazil."

To prepare in advance, the Chinese government has implemented major policy reforms, revised economic theory, andTreat data as a resource equal to capital, people, and land. According to a white paper released last year by the Academy of Information and Communications Technology, China’s national scientific research agency,China currently accounts for 36% of all major language models in the world, second only to the United States.

However, even if the United States and China lead other countries in the world in AI development,As both sides develop in their own ways, the question of who comes first becomes less important.

The original article was published in the Essay section of "Nature" on February 18, 2025 under the title ChinamadewaveswithDeepseek,butitsrealambitionisAI-drivenindustrialinnovation