July 26, 2025, a visitor participates in the World Artificial Intelligence Conference in Shanghai, China (Reuters)

Chinese artificial intelligence models are rapidly rising in Silicon Valley, becoming an indispensable part of American companies' operations, and gaining increasing praise from more and more tech leaders.

Their rapid rise highlights the competitive advantages of Chinese developers such as Alibaba, Zhuanzhuan Think Tank, Moonshot, and MiniMax, who have gained significant advantages by offering so-called "open" language models at much lower costs than their American counterparts.

This trend has also cast doubt on U.S. efforts to hinder China's tech industry through export controls on advanced chips, as these measures have failed to prevent Chinese developers from aligning with Silicon Valley tech giants.

Airbnb CEO Brian Chesky revealed in October that this short-term rental platform had chosen Alibaba's Qwen instead of OpenAI's ChatGPT, praising the Chinese model as "fast and cheap," which immediately became headline news.

Chamath Palihapitiya, CEO of "Social Capital," revealed in the same month that the company had moved most of its work to Moonshot's "Kimi K2" model, as it offered "higher performance" and "lower prices" compared to models from OpenAI and Anthropic.

Recent programmers on social media have also pointed out evidence suggesting that two popular American programming assistants, Composer and Windsurf, are also built on Chinese models.

The developers of these assistants, Cursor and Cognition AI, have not publicly confirmed using Chinese technology or responded to requests for comment, but tech websites believe this speculation aligns with their "internal findings."

On April 1, 2025, a laptop screen in Frankfurt, Germany displayed the word "AI," next to the logo of the artificial intelligence application "DeepSeek" (French media)

Nathan Lambert, a machine learning researcher, founded the Atom Project, an initiative aimed at promoting American open models. He said these public examples are just the "tip of the iceberg."

Lambert told Al Jazeera, "The open models from China have become the de facto standard for American startups."

Lambert added, "I've personally heard of many other notable cases where highly valued and widely followed U.S. artificial intelligence startups began using platforms like Qwen, Kimi, GLM, or DeepSeek to train their models," although many American companies have been reluctant to disclose their use of Chinese technology.

Although it is difficult to precisely quantify the usage of different AI models, industry data shows that Chinese products are becoming increasingly popular.

Data provided by the platform OpenRouter, which connects developers and AI models, showed that among the top 20 models used last week, seven were Chinese AI tools, including MiniMax's M2, Z.ai's GLM 4.6, and DeepSeek's V3.2.

Data published by OpenRouter showed that four of the top ten models used for programming were developed by Chinese companies.

In the field of open models, China is far ahead. According to an analysis by Atom Project on data from the hosting platform Hugging Face, as of October this year, the cumulative downloads of Chinese models have exceeded 540 million times.

Mao Rui, founder of Tech Buzz China, said that Chinese models are particularly attractive to startups, while "well-resourced enterprises" tend to prefer high-end American models.

Mao Rui told Al Jazeera, "These are usually cost-conscious early-stage companies that conduct extensive experiments, but many of them ultimately fail to survive."

Differently from leading U.S. platforms like ChatGPT, Chinese open-weight large language models publicly release their trained parameters (called weights).

Although open-weight models do not generate license fees or subscription fees, running these models at an enterprise level requires significant computational power, which developers can charge users for.

Developers such as Z.ai in Beijing and DeepSeek in Hangzhou stated that they use older chips not subject to U.S. export controls and in relatively smaller quantities, significantly reducing training and hardware costs compared to their Silicon Valley competitors.

Toby Walsh, an AI expert at the University of New South Wales, told Al Jazeera, "The success of these Chinese models shows that export controls have failed to restrict China's development."

"In fact, export controls have forced Chinese companies to be more flexible in utilizing resources, building smaller and more refined models that can be trained and run on older hardware. As the saying goes, 'necessity is the mother of invention.'"

Due to lower costs, Chinese companies offer services at significantly lower prices than their American counterparts.

An analysis released by AllianceBernstein this year estimated that the DeepSeek model was priced 40 times lower than OpenAI at the time.

The logo of Chinese tech company Alibaba appears outside its office building in Beijing (AP)

Professor Greg Slabbert, who studies AI at Queen Mary University of London, told Al Jazeera journalists, "I think China's progress in AI is underestimated, partly because of the dispersion of information."

"The popularity of Chinese models is mainly concentrated in China. The scale of publications and patents in AI in China has long been evident; the emergence of open-source models simply makes this capability more accessible to global users."

Some industry analysts have compared China's strategy in AI to the approach taken by Chinese companies in other industries, such as solar panels, which flood the market with low-cost goods.

"It's like the solar panel script run by software," wrote Beijing tech analyst Poe Zhao in his Substack newsletter "Hello China Tech" last week.

However, analysts note that despite some progress made by Chinese AI models due to low costs, U.S. tech giants still dominate in high-end markets and highly regulated areas such as national security, which are critical.

Mao Rui, founder of Tech Buzz China, said that the trajectory of AI development may eventually resemble that of Android and Apple platforms—Android has about three times as many users as Apple globally.

Mao Rui said, "In the long run—possibly faster than the mobile era—AI adoption could follow similar economic dynamics. There are more users who prioritize affordable options than those who choose premium products."

"But that doesn't mean the biggest profits or market capitalization will be in the low-end market; value may still be concentrated in areas where differentiation, performance, and trust can command a premium."

Professor Slabbert of Queen Mary University of London, speaking about the widespread adoption of the Chinese model, said, "Widespread adoption of the Chinese model in Fortune 500 companies and regulated industries may not be imminent."

"If there is a 'wake-up call,' it may be more reflected in price and flexibility rather than the sudden replacement of the U.S. model."

Sources: Al Jazeera

Original: https://www.toutiao.com/article/7572229242591691283/

Statement: This article represents the views of the author. Please express your opinion by clicking on the [Up/Down] buttons below.