This summer, in less than a month, the World Artificial Intelligence Conference and the World Robot Conference were held in Shanghai and Beijing, respectively, highlighting China’s ambitions in the most strategic new technology fields.
Although the two conferences are called "world" conferences, they are actually "1+N" diplomatic platforms held by China as its home court. This model can be traced back to the World Internet Conference founded in Wuzhen, Zhejiang in 2014, and has now expanded from the Internet to artificial intelligence and robots.
In the United States, artificial intelligence has also been elevated to a strategic level. Just between the two conferences, US President Trump approved the inclusion of "advanced foreign-made robotic equipment" on the national security controlled list. The impact on Chinese companies is of particular concern. China's humanoid robot shipments account for more than 80% of the world's total, with Yushu Technology accounting for 32%.
From AI models, advanced chips to robots, the competition between China and the United States is extending from the product competition of technology companies to industry and national security. It has also been described as a new "AI arms race." So, what exactly are China and the United States competing against?
Is AI really becoming an “arms race”?
Former US Secretary of State Kissinger once warned that AI could become a "crazy race to disaster"; Musk also said that AI could lead to "World War III." The U.S.-China Economic and Security Review Commission of the U.S. Congress even recommended launching an “AI version of the Manhattan Project.”
The Manhattan Project was the United States' secret plan to develop nuclear weapons during World War II.
Since then, more and more voices have compared today’s AI competition to the nuclear race in the 20th century.
In this arena, there is a fierce scene of "you chasing me". In November 2022, OpenAI released ChatGPT, which was a blockbuster. At that time, the industry believed that the AI gap between China and the United States was two years; in January 2025, DeepSeek released R1, which shocked the U.S. capital market, and the gap was changed to one year; after the release of Kimi K3 in July this year, some analysts believed that the gap had shrunk to four to nine months.
The United States and the Soviet Union also engaged in similar pursuits in nuclear technology.
In July 1945, the United States took the lead in testing an atomic bomb; in August 1949, the Soviet Union successfully tested its first atomic bomb. In November 1952, the United States tested its first hydrogen bomb. Less than ten months later, in August 1953, the Soviet Union also successfully tested a hydrogen bomb. In the field of civilian nuclear energy, the Soviet Union even overtook it for a time: in June 1954, the world's first nuclear power station was built and connected to the grid in Obninsk, the Soviet Union.
"This analogy is valid to a certain extent, because AI and nuclear weapons have great strategic significance." Lu Xiaomeng, director of the geopolitical department of the Eurasia Group, an American political risk consulting company, told BBC Chinese.
But there are differences between the two.
Lu Xiaomeng said that AI has a wide range of commercial applications, but nuclear energy does not. It’s hard to imagine ordinary people using some kind of “civilian version of the core system” on a daily basis, but AI has appeared on everyone’s personal devices and is widely used.
Kyle Chan, a researcher at the John Thornton China Center at the Brookings Institution, believes that there are more differences between the two. Nuclear weapons have only one purpose, mass destruction. Artificial intelligence is a diversified technology with different forms, capabilities, and uses; it can indeed be used for bad purposes, such as enhancing cyber attack capabilities or creating biological weapons, but it is also a common tool used daily by millions of individuals and businesses.
"Therefore, equating artificial intelligence with nuclear weapons is a misjudgment and may lead to wrong policy responses." Chen Kaixin said.
But both China and the United States seem to be taking this competition more and more seriously: When both countries believe that AI will determine future national competitiveness, what exactly do they want to achieve with this technology?
The United States wants to break through the limits, and China wants to expand applications
"China and the United States are taking completely different paths in artificial intelligence." Chen Kaixin said.
Artificial intelligence in the United States is dominated by a few private companies that provide proprietary models. The government basically does not intervene and mainly relies on industry self-regulation.
This model has led to extremely powerful models, huge investments and revenue, but it has also led to growing concerns about how to deal with risks and what role government regulation should play.
Several large AI companies in the United States have shown their pursuit of general artificial intelligence (AGI).
OpenAI states in its charter that the company's mission is to ensure the realization of AGI and make it benefit all mankind, and defines AGI as "a highly autonomous system that surpasses humans in most economically valuable tasks."
Google DeepMind said that the final AGI is expected to bring historic changes and defines it as "AI that is at least as capable as humans on most cognitive tasks."
The core strategy of this route is the Scaling Law, which means to continuously expand model parameters and stack computing power, so that large models continue to evolve, and finally break through key nodes to achieve AGI.
Under this model, most of the models of AI companies are closed source, users pay by token, cash flow is stable, and the company has strong control over the model.
Simply put, the United States is betting on the next technological breakthrough.
China has taken a fundamentally different path.
Chen Kaixin said that China's artificial intelligence is strongly guided and supervised by the central government, and the industry focuses more on promotion, popularization and integration into the real economy. Policymakers hope that AI will improve productivity in all walks of life, rather than insisting on building super intelligence that may lead to mass unemployment.
Vivian Toh, a semiconductor commentator and editor-in-chief of TechTechChina, a Singapore-based technology consulting network, believes that China's unique path for AI is not to break the Western ecosystem or subvert the existing system, but to open up a new track of "low cost and high collaboration".
Lu Xiaomeng also pointed out the differences between the two countries: Chinese companies adopt a "fast follower" strategy and launch "good enough" models at low cost to promote rapid global adoption; while their American counterparts pay more attention to cutting-edge technologies and strive to lead the industry.
In other words, China’s strategy is “diffusion”, or whether AI can quickly enter more industries.
After proposing "artificial intelligence +" last year, this year's "15th Five-Year Plan" further clarified this path. The logic is consistent with the "Internet +" 11 years ago: embedding technological capabilities into all walks of life and improving total factor productivity.
This proliferation is not limited to domestic applications.
This year, it has been further integrated with the "Belt and Road Initiative": at the World Artificial Intelligence Conference, 29 countries signed an agreement to establish the World Artificial Intelligence Cooperation Organization, with most members being developing countries, including Russia, Brazil, Venezuela, etc. The organization has a secretariat and provides 5,000 AI training places in China.
"Of course, this is also a pragmatic industrial strategy for China under the reality of limited access to advanced chips." Zhuo Wei'an said.
Which route is more advantageous in the end is still a question mark.
Lu Xiaomeng believes that in the global AI competition, the role of government intervention is secondary to corporate strategy. Ultimately, AI laboratories are the forefront of technological competition.
If American companies can come up with services for the "next big thing" (such as AGI), they can establish a substantial lead over their Chinese counterparts; otherwise, the Chinese model will surpass the American leader in application ratio and occupy a larger global market share.
Whether the two routes can continue depends on another question: Who has enough resources to really make AI bigger?
China is short of chips and the United States is short of electricity
The underlying infrastructure of AI is chips and electricity. Chips combine with electricity to generate computing power, which is the oil of the AI era.
China and the United States' respective shortcomings happen to be the other's strong points: China is subject to dual constraints of chips and capital, while the United States faces dual constraints of power supply and local community public opinion.
As far as China is concerned, Chen Kaixin further explained that the United States’ export controls on Nvidia’s top chips have made it more difficult for China’s AI laboratories to obtain the computing power needed to train and deploy large models. The scale of Chinese technology companies is also much smaller than that of their American counterparts, and their investment in computing infrastructure cannot be compared.
Recently, popular AI products have experienced congestion and current limitations, which reflects the tight computing power on the supply side.
DeepSeek recently announced a possible price increase to relieve server pressure; Kimi suspended registration and introduced a candidate list because too many users exceeded the server's capacity.
In addition to high-end chips, Chen Kaixin believes that another constraint in China is capital.
There is a gap between China and the United States in investment in the AI field of more than 500 billion U.S. dollars and more than 90 billion U.S. dollars, which affects investment in various aspects such as computing power stacking, talent salaries, and model training.
But China also has its own advantages.
Chen Kaixin believes that China will be able to gradually solve the computing power problem over time: although it may not be able to catch up with the United States, domestic AI chips and hardware systems are providing "sufficient" alternatives, enough to allow China's AI industry to continue to move forward.
The problem facing the United States is different.
Lu Xiaomeng pointed out that the United States is facing energy constraints, partly due to the relatively old power systems of various states. China's power generation is about 2.5 times that of the United States.
Chen Kaixin said that although the United States continues to expand its power supply, growing community opposition to new data centers may also hinder its future expansion of computing power. For example, the governor of Texas recently suspended the approval of new data center grid connection requests.
"In comparison, the problem facing the United States is more difficult, because in the final analysis it is a political and social issue, not a technical issue." Chen Kaixin said.
“Many Americans do not recognize the value of artificial intelligence and even view it as a net negative force, so their support for data center construction is mixed.”
Three possible futures
Will the competition between China and the United States in AI be like the competition between the United States and the Soviet Union in nuclear technology, until one side collapses and disintegrates?
Many experts have proposed two versions of the future, light and dark.
Another possibility is that the global AI ecosystem will further fragment.
Chen Kaixin believes that if the Chinese model really narrows the gap or even surpasses the American model, the United States may take tougher measures to slow down China's AI development, which may include more comprehensive technical controls.
The worst-case scenario may be that AI competition directly intersects with national security crises.
Lu Xiaomeng believes that the worst-case scenario is some kind of unexpected war between the two countries, which can be imagined as the "AI version of the Cuban Missile Crisis" getting out of control. This is why it is necessary to establish strategic communication channels to maintain the flow of information between the two countries and avoid global crises caused or promoted by AI.
In addition, what really changes the rules of the competition may not be either China or the United States, but the breakthrough in AI's own capabilities.
Oxford University philosopher William MacAskill said in an interview in May this year that if AI can independently develop stronger AI, it will develop stronger AI at a larger scale and faster efficiency, entering an exponential "intelligence explosion" that will compress a century of human progress in a few years.
What worries him most is that power is highly concentrated in a certain country, company or even individual: AI has no votes, property or organizational loyalty, and its behavior depends entirely on training goals, and traditional political checks and balances may fail.
