In a sleek showroom in China’s eastern city of Suzhou, Cambridge-trained scientist and entrepreneur Yu Kai switches fluidly between British-accented English and his native Chinese as he rattles off the achievements of his unicorn AI firm.
Dr. Yu and a co-founder launched AISpeech in the United Kingdom in 2007, but relocated to China the following year, betting on China’s favorable policies and fledgling artificial intelligence industry.
That bet paid off. AISpeech is now a pioneer in conversational AI, with its breakthrough technology winning clients from Mercedes-Benz to Chinese electric-car giant BYD.
Why We Wrote This
In the United States, government regulation is often viewed as the enemy of innovation. But China’s AI rise tells a different regulatory story – one that could shape the global race for AI dominance.
“China’s industrial policy, planning, and continuity is obviously stronger than that abroad,” says Dr. Yu.
Beijing’s prioritization of AI as a strategic sector for the past two decades has helped create a vibrant AI ecosphere, and narrow the technological gap with the United States, long the global leader in AI innovation.
China is producing increasingly cutting-edge large language models – programs trained on huge amounts of data to recognize text and perform tasks – and spreading free, open-source models around the world. It is on track to achieve its goal of an AI industry worth $100 billion by 2030.
Still, experts stress that China’s race to surpass the U.S. in AI leadership involves much more than the short-term sprint to develop the best models. The true contest will be spreading AI technology throughout various sectors of the economy. That will be more like a long-distance run, says Jeffrey Ding, an assistant professor of political science at George Washington University. It could take decades, similar to the adoption of past technologies such as electricity and the computer.
Each country’s ability to run AI systems, known as compute power, will be a significant factor, as will its decisions on how to regulate emerging technologies.
“We shouldn’t overhype China’s progress,” Dr. Ding says. “The U.S. is still very well positioned to diffuse AI at scale.”
However, China’s gains suggest “prudent regulation could actually make that diffusion more sustainable … [and] cultivate more public trust,” he adds.
China’s strategy and the race to innovate
Beijing regulates primarily to control content – so models do not produce politically sensitive information. Yet this censorship requirement has not slowed China’s AI innovation, as some experts predicted.
“We were wrong about the barrier that censorship would pose toward China’s development of large language models,” says Dr. Ding. “Chinese companies and the government have developed a workable model for censoring … sensitive content in a way that doesn’t slow down innovation.”
Ann Scott Tyson/The Christian Science Monitor
In 2023, China’s AI models were estimated to be three years behind those in the U.S. Today, it’s “more like a six- to 12-month gap,” says Helen Toner, an expert in U.S. and Chinese AI policy at Georgetown’s Center for Security and Emerging Technology.
More relevant for the U.S., China has set guardrails around generative AI – think ChatGPT and other tools that create seemingly original video or audio. Beijing requires public-facing services to register algorithms, and to disclose security assessments and training data used to develop the model.
For its part, the U.S. under the Trump administration has opposed such regulation, arguing it would hobble America in the AI race. Yet, to the contrary, rules requiring transparency about what companies are building can prevent AI mishaps that could alarm the public and slow adoption of AI.
“A sensible set of guardrails can actually support the industry and accelerate development,” says Ms. Toner.
In addition to these regulations, Beijing’s top-down AI strategy includes robust government investment into AI models, research, talent, and start-ups. The government is also amassing vast quantities of data that can help train AI models, subsidizing computing power, striving for self-sufficiency in chip production, and pushing for widespread AI adoption.
Overall, Beijing is prioritizing real-world applications in manufacturing, farming, and other traditional sectors, as well as in new industries like drones and self-driving vehicles.
Dr. Yu’s focus at AISpeech fits into this pattern. He says his firm has scores of original technologies and more than 600 invention patents which leverage AI to connect people to the hardware around them.
For example, AISpeech’s systems can perceive distinct voices in a crowd – what he describes as “the holy grail of speech processing” – and reliably allows them to direct hardware, such as smart cars or home appliances, to execute tasks amid the commotion.
“At a … party, everybody is speaking, but a human can listen to a particular speaker very clearly,” explains Dr. Yu. That’s much harder for machines, he adds.
Global governance and cooperation
China seeks to compete for AI markets and soft-power impact internationally, while maintaining self-sufficiency at home.
At the World Artificial Intelligence Conference in Shanghai last month, Chinese Premier Li Qiang announced an ambitious proposal to create a world AI cooperation organization that would expand China’s role in the global governance of the technology.
Mr. Li said AI governance is fragmented, with regulations varying widely between countries. “We should strengthen coordination to form a global AI governance framework as soon as possible,” he said.
Meanwhile, China is spreading its open-source models abroad, with many countries adopting them. According to Ms. Toner, this is allowing Chinese programs such as DeepSeek’s R1 model, Alibaba’s Qwen series, and Moonshot’s Kimi to take over a lead once held by Meta. “[This] does matter from a soft-power perspective, and U.S. companies have ceded a lot of ground by not prioritizing open models,” she says.
China’s moves overseas also offer a workaround to U.S. restrictions on advanced semiconductor chips, by encouraging outside computing power to run the Chinese models.
Amid the U.S.-China technological decoupling, the mantra of self-sufficiency is common in Chinese AI firms, including AISpeech in Suzhou. “The current technology dispute doesn’t affect us very much, because we primarily rely on ourselves,” says Dr. Yu, whose company has its own voice-control chip manufactured inside China.
Yet Dr. Yu urges U.S.-China cooperation on AI as “a must for the good of the world.”
Beijing and Washington have already held talks aimed at preventing some of the biggest anticipated risks of AI, including the use of AI weapons by terrorist groups and other nonstate actors, or the unleashing of a rogue superintelligence.
Dr. Yu believes humans can manage these threats and “co-evolve” with AI. “I’m pretty much optimistic,” he says.
