China is focusing on large language models (LLMs) in the field of artificial intelligence.
Blackdovfx | Istock | Getty Images
China’s attempts to dominate the world of artificial intelligence could be paying off. Industry insiders and technology analysts told CNBC that Chinese AI models are already extremely popular, rivaling and even surpassing those from the United States in terms of performance.
AI has become the latest battleground between the US and China and is viewed as a strategic technology by both sides. Washington continues to restrict China’s access on cutting-edge chips designed to support artificial intelligence amid fears the technology could threaten U.S. national security.
This has led China to adopt its own approach to increasing the appeal and performance of its AI models, including through the use of open source technology and the development of its own superfast software and chips.
China creates popular LLMs
Like some of the leading US firms in the field, Chinese AI firms are developing so-called Large Language Models (LLMs), which are trained on huge amounts of data and support applications such as chatbots.
However, unlike the OpenAI models that support the hugely popular ChatGPT, this is the case with many of these Chinese companies Development of open source or open weight LLMs which developers can download and build upon for free and without strict licensing requirements from the inventor.
According to Tiezhen Wang, a machine learning engineer at the company, Chinese LLMs are the most downloaded on Hugging Face, a repository for LLMs. Qwen, a family of AI models developed by the Chinese e-commerce giant Alibabawas most popular at Hugging Face, he said.
“Qwen is rapidly growing in popularity due to its outstanding performance in competitive benchmarks,” Wang told CNBC via email.
He added that Qwen has an “extremely favorable licensing model,” meaning it can be used by companies without the need for “extensive legal reviews.”
Qwen comes in different sizes or parameters as they are called in the world of LLMs. Models with large parameters are more powerful but incur higher computational costs, while smaller models are cheaper to run.
“Regardless of the size you choose, Qwen is probably one of the best-performing models on the market today,” added Wang.
DeepSeek, a start-up, also recently caused a stir with a model called DeepSeek-R1. DeepSeek said last month that its R1 model competes with OpenAI’s o1 – a model designed for thinking or solving more complex tasks.
These companies claim that their models can compete with other open source offerings like Meta‘s Llama as well as closed LLMs like OpenAI’s across different functions.
“Over the last year we have seen a surge in Chinese open source contributions to AI with really strong performance, low deployment costs and high throughput,” Grace Isford, partner at Lux Capital, told CNBC by email.
China is pushing to spread open source globally
Open-sourcing a technology serves a number of purposes, including encouraging innovation as more developers have access to it and building a community around a product.
It’s not just Chinese companies that have introduced open source LLMs. Facebook parent company Meta and European start-up Mistral also have open source versions of AI models.
But with the tech industry in the crosshairs of the geopolitical battle between Washington and Beijing, open source LLMs give Chinese companies another advantage: They enable their models to be used globally.
“Chinese companies want their models to be used outside of China. So this is definitely an opportunity for companies to become global players in the AI space,” Paul Triolo, partner at global consulting firm DGA Group, told CNBC via email.
While the current focus is on AI models, there are also debates about what applications will be built on top of them – and who will dominate this global internet landscape in the future.
“If you take the critical question of these breakthrough AI models, it’s about what these models will be used for, such as accelerating breakthrough science and engineering technology,” Lux Capital’s Isford said.
Today’s AI models have been compared to operating systems, e.g Microsoft’s Window, Googleis Android and Appleis iOS with the potential to dominate a market like these companies do on mobile and PC.
If this is true, it increases the chances of building a dominant LLM.
“They (Chinese companies) see LLMs as the center of future technology ecosystems,” Xin Sun, a lecturer in Chinese and East Asian economics at King’s College London, told CNBC by email.
“Their future business models will depend on developers joining their ecosystems, building new applications on top of the LLMs, and attracting users and data from which profits can subsequently be generated, in a variety of ways, including, but well beyond, through directing users to use their cloud services,” Sun added.
Chip restrictions raise doubts about China’s AI future
AI models are trained on huge amounts of data, which requires enormous computing power. At the moment, Nvidia is the leading developer of the necessary chips, so-called graphics processors (GPUs).
Most leading AI companies train their systems on Nvidia’s most powerful chips – but not in China.
Last year, the United States tightened export restrictions on advanced semiconductor and chip manufacturing equipment to China. It means Nvidia’s cutting-edge chips cannot be exported to the country and the company had to produce sanctions-compliant semiconductors for export.
Despite these limitations, Chinese companies have still managed to bring advanced AI models to market.
“Major Chinese technology platforms currently have sufficient access to computing power to further improve their models. This is because they stock large quantities of Nvidia GPUs and also use domestic GPUs from Huawei and other companies,” said DGA Group’s Triolo.
In fact, they were Chinese companies Increase efforts to create viable alternatives to Nvidia. Huawei has been one of the leading players in pursuing this goal in China, while companies like it Baidu and Alibaba have also invested in semiconductor design.
“However, the gap in advanced hardware computing power will widen over time, particularly next year when Nvidia launches its Blackwell-based systems, exports of which are restricted to China,” Triolo said.
Lux Capital’s Isford noted that China is “systematically investing and expanding its entire domestic AI infrastructure stack outside of Nvidia with powerful AI chips from companies like Baidu.”
“Whether or not Nvidia chips are banned in China will not stop China from investing and building its own infrastructure to build and train AI models,” she added.
