AI in the People's Republic China Impresses with Rapid AI Adoption and Open Models

Related Vendors

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

China relies on openness instead of isolation in the global AI race, disrupting an apparently entrenched balance of power. But will China's open approach ultimately tip the scales and threaten the U.S.'s lead in AI development?

Is China winning the AI race against the USA with the help of "open-weight" models? Current indications suggest so. While American companies like OpenAI carefully guard their secrets (closed-weight), Chinese companies are making their weights publicly accessible. The result is an even faster, almost universal adoption of models like Doubao, Deepseek, or Yuanbao among the Chinese population and many businesses.

The Chinese now use generative AI in their daily lives almost every day. Participants include not only computer nerds and well-educated knowledge workers but also elementary school students and retirees. The "penetration" of generative AI in everyday Chinese life is already truly impressive.

The "Tencent Research Institute" recently published the results of a survey. 96.2 percent of the approximately 3,500 Chinese respondents already use the Chinese equivalents of ChatGPT. And not just superficially or occasionally. 67.7 percent use generative AI at least once a day—at work, for learning, or for everyday questions.

"Hey Doubao!"

While this survey might be dismissed as optimistic since Tencent itself has business interests in this field, the adoption of chatbots on mobile devices in everyday Chinese life is undeniable. Hardly any Chinese citizen, regardless of age, goes without saying, “Let’s just ask Doubao” or “Let’s just ask Deepseek,” as it’s called here. Companies, ranging from very large to very small, are also integrating Chinese LLMs into their workflows with astonishing speed.

Leading the applications is Doubao from Bytedance, the parent company of TikTok, followed by Deepseek. In third place is Yuanbao from Tencent. The decision of all these Chinese AI companies to offer their models as "open-weight" contributes to their growing popularity.

Unlike open-source software, where the source code is made available and can be modified by anyone, open-weight LLMs typically only release the numerical parameters or "weights" learned during training online. Anyone can download them and experiment with them. It is a much more transparent way of enabling the use of chatbots and models compared to proprietary, closed models.

Affordable And in Demand

The Chinese also take a different approach to pricing than the Americans. Unlike the expensive closed-weight models from the U.S., such as OpenAI's ChatGPT or Anthropic's Claude, using Chinese AI models is almost free. Rarely is it priced significantly higher than a subscription to a music app on a mobile phone.

Behind all this lies an entirely different business model than that in the USA. Chinese companies are not trying to improve proprietary models with billion-dollar investments to sell them at high prices but are focusing on the rapid adoption of their models. Currently, they are making much less money than their American competitors, but whichever company prevails in this race for market share will be able to earn as much as it wants later.

Open-weight models also encourage different types of adoption compared to closed, secret-weight models. “They can be more easily tailored by businesses, governments, and researchers to the specific requirements of individual use cases and allow users to operate their AI tools locally rather than in the cloud. Revenue can still be generated through accompanying services, such as support with customization,” the Economist recently quoted Percy Liang, an associate professor of computer science at Stanford University and director of the “Center for Research on Foundation Models (CRFM).”

Strategic Differences

Thus, different strategies can be observed. While American companies speculate on huge profits by attempting to develop the best models (and, as seen recently with OpenAI's "gpt-oss," only half-heartedly participate in releasing open-weight models), Chinese companies are primarily focused on spreading their models and "gathering users" for now. Making money comes later.

Since the Chinese start-up Deepseek, previously almost unknown, released its LLM "R1" in January 2025—which, despite its lower training costs, has greatly impressed many American AI experts—the debate about dominance in the global AI race has shifted. While most observers previously believed that the USA would always remain ahead, many of them are no longer so sure today.

This moment in January has since been referred to as the "Deepseek Shock." In China, Deepseek has now made the leap from an insider tip for AI enthusiasts to a widely used everyday tool for the Chinese population. Other providers, such as Bytedance and Tencent, quickly followed with their own models.

Secret Approach

The fact that affordable language models, which can be even better integrated into one's own processes thanks to open weights, are very popular can be seen not only in China but also in Silicon Valley, California. Even there, many start-ups secretly use Deepseek, although they rarely admit it openly due to the currently widespread China hysteria in politics and media.

"This might be the biggest secret in Silicon Valley that cannot be revealed," writes the Chinese tech portal Huxiu. "Since DeepSeek exists—who would still be willing to pay for OpenAI or Anthropic?" Huxiu quotes Paddy Cosgrave, CEO and founder of "Web Summit."

If adoption truly proves to be a decisive success factor for AI companies and open weights play a role in it, the next "Deepseek Shock" may soon be upon us. (sb)

Secure and Compliant Authentication in Laboratories

Secure & Contactless User Authentication at GMP/GLP