Commentary: How China Could Keep Up With the Global AI Race

13 Aug 2025

On Aug. 7, OpenAI launched GPT-5, its most advanced large language model to date. The model delivers comprehensive performance improvements in reasoning, code generation, writing, and multitasking, with a significantly reduced hallucination rate and optimized response speed through an intelligent routing mechanism. A basic version is available to free users, while premium users receive higher API call limits. CEO Sam Altman described it as being “like having a Ph.D. expert at your side,” calling it a significant step toward artificial general intelligence. The release has sparked a global frenzy and once again put the question of who will lead the next generation of AI development into the spotlight.

In fact, as early as May 8, the U.S. Senate held a hearing titled “Winning the AI Race: Bolstering American Capabilities in Computing and Innovation.” In his testimony, Altman emphasized that AI is strategic national infrastructure, on par with electricity and the internet. He proposed a “twin revolution” of “abundant intelligence” and “abundant energy,” calling for massive investment in computing power, chips, data centers, and energy supply. He also argued against cumbersome pre-approval processes, advocating for methods like “regulatory sandboxes” to protect the speed of innovation. At the hearing, figures from American politics and business unequivocally identified AI as the commanding height of future national competitiveness, stressing the necessity of winning the race against China.

Against this backdrop, China must face the gaps in its infrastructure, regulatory policy, and innovative capacity, and formulate a more forward-looking strategy in response.

Infrastructure: computing investment and hardware layout

The U.S. maintains a lead in computing power and chips, thanks to its tech giants and capital advantages. OpenAI is building the world’s largest AI training center in Texas, and CoreWeave operates data center clusters with more than 250,000 GPUs. The U.S. dominates high-end chip design and manufacturing through companies like Nvidia and AMD. In 2024, its total computing power reached 291 exaflops, accounting for 32% of the global total.

China is rapidly expanding its computing power through the “Eastern Data, Western Computing” project. As of June 2024, its total capacity reached 246 exaflops, with a forecast of 300 exaflops by 2025. The growth rate of its intelligent computing power has reached 65%. Major AI computing platforms have been built in Beijing, Shanghai, and Shenzhen. However, high-end chips remain constrained by U.S. export controls, and domestic alternatives like Ascend and Cambricon have slightly inferior performance for the short term. The overall situation is one where the U.S. leads in existing capacity while China is catching up in incremental growth.

Regulatory systems: policy orientation and governance models

The U.S. favors “light-touch regulation,” opposing the European model of preemptive controls. It plans to establish a national AI regulatory sandbox and is issuing voluntary standards through the National Institute of Standards and Technology. China, by contrast, emphasizes preemptive control and filing requirements, such as the Interim Measures for the Management of Generative Artificial Intelligence Services, which mandate content security reviews and real-name registration. By the end of 2024, 302 generative AI services had completed these filing requirements.

The American model is conducive to rapid iteration, while the Chinese model is more robust in terms of security and social stability. In the future, China could introduce flexible mechanisms like tiered and categorized regulation and sandbox pilot programs, provided security is ensured, to avoid “one-size-fits-all” approaches that stifle innovation.

Innovative capacity: research strength and industrial ecosystem

The U.S. remains the primary source of top-tier models and original algorithms. In 2024, it produced 40 representative large models, far exceeding China’s 15. The launch of GPT-5 further solidifies its technological advantage. The U.S. possesses the world’s most mature capital and talent ecosystem, with AI investment reaching $109.1 billion in 2024, nearly 12 times that of China.

Although a latecomer, China is making rapid progress. Models like Baidu’s ERNIE 4.0 and DeepSeek R1 are approaching American levels on several benchmarks. The value of China’s core AI industry reached 578.4 billion yuan ($80.3 billion) in 2023, with over 4,000 companies. The country also has a wealth of application scenarios and high user acceptance (83% of respondents believe AI’s benefits outweigh its harms). The challenges lie in the conversion rate of basic research, the influence of its open-source ecosystem, and the supply of high-end talent.

Overall, the U.S. leads in technology and ecosystems, while China has clear advantages in markets and applications. The competition for global influence between the two sides will only intensify.

China’s strategic response

1. Solidify AI infrastructure to build a highland of computing power and hardware.

Given the trend that “computing power is national power,” China should accelerate the inclusion of AI-related infrastructure in its national plan for new infrastructure, achieving forward-looking deployment and “zìzhǔ kěkòng” (autonomous and controllable) systems.

First, accelerate the layout of computing hubs. Building on the “Eastern Data, Western Computing” project, China should increase the construction of large AI computing centers to meet the 300 exaflops goal by 2025, while also planning ahead for E-class and Z-class computing facilities by 2030. It should optimize the computing network layout, enhance inter-regional scheduling efficiency, and build a high-speed backbone network to supply AI computing power as precisely as water and electricity. A national AI computing service platform could be established to integrate supercomputing centers and cloud resources to provide low-cost computing rentals for research institutions and small and medium-sized enterprises.

Second, strengthen guarantees for key hardware. China should leverage major national projects and industrial funds to promote the R&D and mass production of domestic GPUs, NPUs, and high-bandwidth memory to reduce reliance on imports. Companies making major breakthroughs should receive tax incentives and preferential treatment in government procurement, and industry-university-research collaboration should be encouraged to tackle “choke-point” technologies. Cooperation with friendly countries on semiconductor materials and equipment should be strengthened to diversify supply chain risks.

Third, improve energy and communication support. Green data center clusters should be built in energy-rich regions, with costs lowered through direct supply of new energy and preferential electricity rates. Trunk networks should be upgraded by deploying high-speed fiber optics and low-latency networks to support cross-regional computing calls. Urban planning should reserve land and power quotas for data centers to avoid restrictions from “dual control” energy consumption policies.

2. Optimize the regulatory system to both ensure security and promote innovation.

While guaranteeing security and control, regulatory policies should be more flexible and tolerant to leave room for technological development and application.

First, pilot regulatory sandboxes. AI innovation regulatory sandboxes should be established in areas like the Guangdong-Hong Kong-Macao Greater Bay Area, Zhongguancun in Beijing, and the West Bund in Shanghai. These would provide regulatory exemptions or simplified approvals for qualified companies and technologies, allowing new tech to be tested in a controlled environment where risks can be identified and mitigated in a timely manner.

Second, implement tiered and categorized regulation. High-risk fields such as medical diagnostics and autonomous driving should continue to face strict access and security assessments. For low- and medium-risk applications like office assistants and content generation, a model of filing plus post-hoc supervision should be adopted to reduce innovation costs. A risk assessment standards system should be established to clarify mandatory review and exemption scenarios, replacing vague requirements with a list-based management approach.

Third, strengthen coordination and synergy. It is recommended that the State Council or the Cyberspace Administration of China lead the establishment of a cross-departmental governance coordination mechanism to unify regulatory standards and reduce uncertainty from multi-agency oversight. Laws and regulations that are ill-suited for AI development should be revised to facilitate compliant data flows and international cooperation.

Fourth, encourage industry self-discipline and standard-setting. Support should be given to forming AI ethics and security self-regulatory alliances to formulate industry guidelines. China should also actively participate in international standards organizations to gain a voice on issues like algorithmic transparency and privacy protection, and reduce overseas regulatory barriers through dialogue with major economies.

3. Cultivate an innovation ecosystem to enhance sustained technological capabilities.

Facing the formidable strength of the U.S. in talent, capital, and ecosystems, China has an even greater need to build fertile ground for AI innovation and strengthen its indigenous innovation capacity from the source.

First, implement talent programs. Establish interdisciplinary AI schools or research institutes in universities and expand enrollment. Encourage joint training bases between universities and enterprises. Open green channels and offer tax incentives to attract high-end talent in short supply, encouraging overseas talent to return. The evaluation mechanism for researchers should be improved, allowing part-time work and the commercialization of research results to count toward professional titles.

Second, increase investment in basic research. Funding and support from key R&D programs should be increased for foundational AI theories and “choke-point” technologies. Long-term special projects should be established in areas like brain-inspired intelligence, general AI, and quantum computing + AI, with tolerance for failure in exploratory research. Open and shared datasets and evaluation benchmarks should be built to alleviate data bottlenecks for research institutions and SMEs.

Third, optimize the industrial environment. The investment and financing ecosystem should be improved by setting up AI industry funds focused on foundational models, key software, and AI + industry startups. Fair opportunities should be provided to startups in government procurement and bidding, with “first-purchase, first-use” policies to open up markets. Large enterprises should be encouraged to open their platform resources and innovate collaboratively with smaller firms.

Fourth, build an open and cooperative ecosystem. Participation in international open-source communities should be encouraged, along with leading influential projects. Non-sensitive model APIs should be opened to grow the local developer community. Openness should be used to both “bring in” and “go global.”

Fifth, promote AI going abroad. Overseas R&D centers or labs should be established along the Belt and Road to provide localized solutions. The government should incorporate AI cooperation into its diplomatic agenda and advance the Digital Silk Road initiative, exporting Chinese standards through technical assistance and training to create global benchmark projects and enhance international influence.

Conclusion

The launch of GPT-5 is not merely a technological iteration but a signal that the global AI race is accelerating. China should plan for the development of artificial intelligence with a more long-term vision and pragmatic measures. On the one hand, it must directly confront the competitive pressure from leading countries like the U.S. and shore up its weaknesses in infrastructure and core technologies. On the other hand, it must leverage its own institutional and market advantages to accelerate policy support and application deployment. As the U.S. hearing demonstrated, artificial intelligence has become the new focal point of competition over national strength. We must have a sense of urgency, but more importantly, strategic resolve. Through sustained investment and an optimized policy environment, we must cultivate the soil of innovation. While ensuring that development is secure and controllable, we must boldly embrace the AI revolution and strive to achieve the strategic goal of making China a global leader in AI by 2030.

Tan Yinliang and Liu Geng are researchers at China Europe International Business School.

This is an AI-generated English rendering of original reporting or commentary published by Caixin Media. In the event of any discrepancies, the Chinese version shall prevail.

The views expressed in third-party articles are those of the authors and do not necessarily reflect the positions of Caixin.

caixinglobal.com is the English-language online news portal of Chinese financial and business news media group Caixin. Global Neighbours is authorized to reprint this article.

Image: Creative Library – stock.adobe.com