04-02-2025, 06:01 PM
Article was written 2 April, 2025.
Translated 2 April, 2025.
Why China is distributing the most advanced neural networks for free.
China is actively building influence in the field of artificial intelligence, betting on an open development model. Following the release of the free DeepSeek model earlier this year, Yann LeCun, Meta's chief AI scientist, clarified that it's not that China has overtaken the US, but that open models are beginning to surpass closed models.
The largest Chinese tech companies - Baidu, Alibaba, Tencent - are eagerly moving to open source. This means that anyone can freely use, study, modify and distribute the program code of AI models. Moreover, the Chinese authorities officially support this approach: DeepSeek CEO Liang Wenfeng was presented to Premier Li Keqiang as a representative of the country's entire AI sector.
Unlike OpenAI, which closely guards its models, Chinese companies, including DeepSeek, publish source code with minimal restrictions. Even Meta, which promotes its Llama series of models, maintains restrictions on commercial use. Nevertheless, more and more industry participants are finding that openness is the key to long-term success.
The reason why China relies on open source is not just ideological. US restrictions prevent Chinese companies from purchasing the most powerful chips from Nvidia needed for AI training. Therefore, using open source models created with access to such resources has become an alternative solution. Prior to DeepSeek, almost all Chinese models were reworked versions of Llama.
Such an approach is helping to close the gap with the U.S. more quickly. In recent weeks, Baidu, Alibaba, Tencent and DeepSeek have unveiled new versions of their open source models. According to Bloomberg , Ant Group - an entity founded by Jack Ma - has learned how to train AI on Chinese Huawei chips, getting results comparable to using Nvidia processors. If such techniques enter widespread use, it will bring China closer to its strategic goal of technological independence.
China is not the only one seeking to develop open technologies. In Europe, for example, European Commission head Ursula von der Leyen recently unveiled a plan to invest 200 billion euros in collaborative AI innovation. French company Mistral AI is one of the flagships of this movement.
China's generosity in proliferating powerful AI models reinforces its soft power. According to DeepSeek's Liang, participation in open source brings respect, and this is not just for the company, but for the country as a whole. U.S. economist Tyler Cowen even noted that China is winning not only in technology, but also in “general sentiment” - how the rest of the world perceives it.
At the same time, the Chinese authorities are promoting open source in other areas, for example, in the development of chips based on RISC-V architecture, a free alternative to solutions from Arm, Intel and AMD. This is strategically important: China's access to Western technologies can be cut off at any time by sanctions from Washington.
However, the open approach has costs. Companies make less money, because they can't charge for access to models. For example, OpenAI builds its business on selling subscriptions and licenses. DeepSeek, on the other hand, can only rely on revenue from integrating its models into other people's products. This may not be a problem for Liang - he says he prioritizes innovation over profit. But for publicly traded companies like Alibaba, which has invested tens of billions of dollars in AI and cloud technology, low returns could hit the stock price.
There are political risks, too. While China has been supportive of open source so far, that could change if these technologies become key to military or cyber operations. Some DeepSeek employees have already faced travel restrictions, as The Information reported. And analysts at the U.S. Center for Strategic and International Studies say DeepSeek's unique model training techniques may be more useful to U.S. companies with more powerful computing resources.
Importantly, China's openness does not extend to areas where it is already a leader. For example, technologies for processing rare earth elements are banned for export, and plans to build BYD car plants in Mexico have been put on hold due to fears of technology leakage to the US. So free AI models from China are out there today, but how long that will last is an open question.
Translated 2 April, 2025.
Why China is distributing the most advanced neural networks for free. China is actively building influence in the field of artificial intelligence, betting on an open development model. Following the release of the free DeepSeek model earlier this year, Yann LeCun, Meta's chief AI scientist, clarified that it's not that China has overtaken the US, but that open models are beginning to surpass closed models.
The largest Chinese tech companies - Baidu, Alibaba, Tencent - are eagerly moving to open source. This means that anyone can freely use, study, modify and distribute the program code of AI models. Moreover, the Chinese authorities officially support this approach: DeepSeek CEO Liang Wenfeng was presented to Premier Li Keqiang as a representative of the country's entire AI sector.
Unlike OpenAI, which closely guards its models, Chinese companies, including DeepSeek, publish source code with minimal restrictions. Even Meta, which promotes its Llama series of models, maintains restrictions on commercial use. Nevertheless, more and more industry participants are finding that openness is the key to long-term success.
The reason why China relies on open source is not just ideological. US restrictions prevent Chinese companies from purchasing the most powerful chips from Nvidia needed for AI training. Therefore, using open source models created with access to such resources has become an alternative solution. Prior to DeepSeek, almost all Chinese models were reworked versions of Llama.
Such an approach is helping to close the gap with the U.S. more quickly. In recent weeks, Baidu, Alibaba, Tencent and DeepSeek have unveiled new versions of their open source models. According to Bloomberg , Ant Group - an entity founded by Jack Ma - has learned how to train AI on Chinese Huawei chips, getting results comparable to using Nvidia processors. If such techniques enter widespread use, it will bring China closer to its strategic goal of technological independence.
China is not the only one seeking to develop open technologies. In Europe, for example, European Commission head Ursula von der Leyen recently unveiled a plan to invest 200 billion euros in collaborative AI innovation. French company Mistral AI is one of the flagships of this movement.
China's generosity in proliferating powerful AI models reinforces its soft power. According to DeepSeek's Liang, participation in open source brings respect, and this is not just for the company, but for the country as a whole. U.S. economist Tyler Cowen even noted that China is winning not only in technology, but also in “general sentiment” - how the rest of the world perceives it.
At the same time, the Chinese authorities are promoting open source in other areas, for example, in the development of chips based on RISC-V architecture, a free alternative to solutions from Arm, Intel and AMD. This is strategically important: China's access to Western technologies can be cut off at any time by sanctions from Washington.
However, the open approach has costs. Companies make less money, because they can't charge for access to models. For example, OpenAI builds its business on selling subscriptions and licenses. DeepSeek, on the other hand, can only rely on revenue from integrating its models into other people's products. This may not be a problem for Liang - he says he prioritizes innovation over profit. But for publicly traded companies like Alibaba, which has invested tens of billions of dollars in AI and cloud technology, low returns could hit the stock price.
There are political risks, too. While China has been supportive of open source so far, that could change if these technologies become key to military or cyber operations. Some DeepSeek employees have already faced travel restrictions, as The Information reported. And analysts at the U.S. Center for Strategic and International Studies say DeepSeek's unique model training techniques may be more useful to U.S. companies with more powerful computing resources.
Importantly, China's openness does not extend to areas where it is already a leader. For example, technologies for processing rare earth elements are banned for export, and plans to build BYD car plants in Mexico have been put on hold due to fears of technology leakage to the US. So free AI models from China are out there today, but how long that will last is an open question.
