American businesses and governments have been shaken by China’s aim to make its open-source AI models a worldwide standard. They are preparing their answers to the threat because they fear that U.S. models may be overshadowed.
Beginning with the much-discussed DeepSeek and its R1 reasoning model in January, China has made a series of significant strides in AI this year. Alibaba’s Qwen and a slew of others with names like Moonshot, Z.ai, and MiniMax followed in July.
There are free copies of each model available for people to download and modify. This strategy, often known as open source or open weight, is propelling the worldwide uptake of Chinese artificial intelligence technologies.
American corporations that have kept their models secret are under pressure. In early August, OpenAI, the creator of ChatGPT, published its first open-source model, gpt-oss.
There are several instances in the history of technology when a crowded market at the beginning of an industry eventually gave way to a monopoly or oligopoly of a small number of participants. Examples include Google’s search engine, Microsoft’s Windows operating system for desktop computers, and the iOS and Android operating systems for smartphones.
History also shows that the most technologically sophisticated player does not always win the fight to become an industry standard. Many in Silicon Valley and Washington are concerned about China’s advancements in open-source AI because of its easy access and adaptability.
The Trump administration stated that open-source models “could become global standards in some areas of business and in academic research” in an AI action plan published in July. The United States was urged to create “leading open models founded on American values” in the report.
Since they invest hundreds of millions of dollars in building models and receive no direct compensation, the incentives for open-source AI champions are now minimal. However, companies that lock customers in could be able to sell other services that capitalize on the free portion, just how Google bundles its Android operating system with search, YouTube, and other revenue-generating goods.
Linux is still a popular open-source operating system in the business, and Android is based on it.
Researchers have long supported open source as a means of hastening the development of developing technologies, because it allows any user to view the code and make changes. Chinese policymakers have fostered open-source research and development in a variety of fields, including artificial intelligence, operating systems, semiconductor architecture, and engineering software.
“China is promoting open-source projects as a strategic backup and emergency resource out of fear of being cut off from American technologies,” said Lian Jye Su, an analyst with the AI research firm Omdia.
Each side can use its economic advantages—such as rare-earth materials for China and Nvidia chips for the United States—to demand concessions from the other side, as demonstrated by this year’s trade battle between the United States and China. American officials are concerned that Beijing would find a way to use AI for geopolitical gain if Chinese models take the world by storm.
Beyond the political arena, companies are competing to use open-source AI models. Open-source AI appeals to many clients since they may freely modify and install it on their computers while maintaining the privacy of sensitive data.
One of the largest banks in Southeast Asia, Oversea-Chinese Banking, located in Singapore, has created over 30 internal products utilizing open-source models. These tools include DeepSeek for market trend analysis, Qwen for computer code assistance, and Google’s Gemma for document summarization.
According to the bank, it stayed away from being restricted to a single model. It keeps an eye on new models and has the option to swap if it likes one. In order to receive technical help, it also favors models that are well-known to many developers.
OCBC CEO Donald MacDonald stated, “We probably have a stable of about 10 open-source models that we’re using at any point in time.”
Since November, the world’s top open-weight model from China has outperformed the open-source champion from the United States, according to research company Artificial Analysis. The company that evaluates models’ proficiency in coding, algebra, and other domains discovered that a variant of Alibaba’s Qwen3 outperformed OpenAI’s gpt-oss.
Nonetheless, the Chinese model is about twice as large as OpenAI’s, indicating that Qwen may need more processing power to do easier tasks. According to OpenAI, their free and open-source model scored well on reasoning tasks and was reasonably priced.
Major cloud-service providers in the United States have begun to supply their consumers with gpt-oss. The OpenAI methodology, according to Amazon Web Services, was more affordable than DeepSeek’s R1 operating on its infrastructure.
Engineers, particularly those in Asia, said that they discovered Chinese models were frequently more adept at grasping local dialects and cultural quirks. More Chinese data is used to train Chinese models, which is similar to several other Asian languages.
Yokohama, Japan-based developer Shinichi Usami recently created a chatbot for customer care for a retail customer. He selected Qwen from Alibaba.
Usami stated that with a prominent U.S. model, “we’ve seen cases where the chatbot struggles to grasp the implicit intent from users’ words and the responses can occasionally be not polite enough.” “Qwen seems to manage these subtleties more effectively.”
Businesses first concentrated on undercutting one another’s costs for closed-source models in China’s fiercely competitive AI market. In recent months, this rivalry has spread to open-source models as all parties vie for acceptance and public acclaim.
Chinese businesses frequently put customer loyalty ahead of short-term profits, according to Charlie Chai, a tech researcher at 86Research located in Shanghai.
Larger tech companies are frequently best-positioned to profit from a huge user base by providing complementary services like specialized applications or cloud services, experts said, adding that startups have a limited time to draw consumers.
In a recent blog, Andrew Ng, the CEO of Silicon Valley firm DeepLearning.AI, stated that “many of the current players will fail in this Darwinian life-or-death struggle, but the fierce competition creates strong companies.”
