Some of China's AI models are already surpassing their U.S. counterparts, as the country aims to become a leader in the field.

• Some U.S. rivals are being surpassed by Chinese AI models, which are already very popular and keeping up with the pace, according to industry experts.
• Numerous Chinese companies are utilizing open-source, large language models to enhance innovation and expand their application.
• Hugging Face is a popular repository for AI models, and Chinese LLMs are among its most widely used models.

Chinese AI models are gaining popularity and keeping up with or surpassing US models in performance, according to industry insiders and technology analysts.

The U.S. and China are now competing over AI technology, with Washington restricting China's access to advanced chips used in AI systems, due to concerns about national security.

China has chosen to adopt its unique strategy to enhance the allure and effectiveness of its AI systems, which involves leveraging open-source technology and creating cutting-edge software and hardware.

China is creating popular LLMs

Some of the top U.S. companies are also working on large language models (LLMs), which are trained on vast amounts of data and used in chatbots, just like Chinese AI firms.

While ChatGPT is powered by OpenAI's models, many Chinese companies are developing open-source, or open-weight, LLMs that allow developers to download and build on top of them for free and without strict licensing requirements.

According to Tiezhen Wang, a machine learning engineer at Hugging Face, Chinese LLMs are the most downloaded. Meanwhile, Qwen, a family of AI models created by Chinese e-commerce giant, is the most popular on Hugging Face, he said.

Qwen is gaining popularity quickly because of its impressive performance on competitive benchmarks, as Wang stated in an email to CNBC.

Qwen has a "highly favorable licensing model" that allows companies to use it without requiring "extensive legal reviews."

The size of Qwen, also known as parameters, can vary, with larger models being more powerful but having higher computational costs, while smaller ones are more affordable to run.

Wang stated that Qwen is likely to be one of the best-performing models regardless of the size chosen.

Last month, DeepSeek announced that its R1 model is in competition with OpenAI's o1, a reasoning-based model designed for more complex tasks.

These companies assert that their models can match the performance of other open-source models like 's Llama, as well as closed LLMs from OpenAI, in a range of functions.

Grace Isford, a partner at Lux Capital, stated in an email to CNBC that in the past year, there has been a significant increase in open source Chinese contributions to AI, which have demonstrated exceptional performance, low cost to serve, and high throughput.

China pushes open source to go global

The open-sourcing of technology fosters innovation by allowing more developers to access it and creates a community around the product.

Besides Chinese companies, Meta, the parent of Facebook, and European start-up Mistral also offer open-source versions of their AI models.

With the technology industry at the center of the geopolitical conflict between Washington and Beijing, open-source LLMs provide Chinese companies with an edge by allowing their models to be utilized worldwide.

According to Paul Triolo, a partner at DGA Group, Chinese companies' desire to use their AI models globally indicates their ambition to become major players in the international AI industry.

There is ongoing discussion about the applications that will be developed using AI models and who will control the global internet market in the future.

Isford of Lux Capital stated that assuming frontier base AI models are basic requirements is about their application in advancing frontier science and engineering technology.

AI models today are being compared to operating systems like Windows, Android, and iOS, with the potential to dominate markets, similar to how these companies have done on mobile and PCs.

If true, this makes the stakes for building a dominant LLM higher.

According to Xin Sun, senior lecturer in Chinese and East Asian business at King's College London, Chinese companies view LLMs as the focal point of future tech ecosystems.

Sun stated that their future business models will depend on developers joining their ecosystems, creating new applications based on LLMs, and attracting users and data to generate profits through various means, including but not limited to directing users to use their cloud services.

Chip restrictions cast doubt over China's AI future

Currently, is the leading designer of the chips needed for training AI models, which are known as graphics processing units (GPUs).

While many leading AI companies are utilizing Nvidia's top-performing chips for system training, this practice is not prevalent in China.

The U.S. has intensified export restrictions on advanced semiconductor and chipmaking equipment to China, preventing the country from obtaining leading-edge chips and forcing companies to produce sanction-compliant semiconductors for export.

Despite the obstacles, Chinese companies have still successfully launched advanced AI models.

According to DGA Group's Triolo, Chinese technology platforms have enough computing power to enhance their models due to their accumulation of Nvidia GPUs and the utilization of domestic GPUs from Huawei and other companies.

In China, Chinese companies have intensified their efforts to develop alternatives to Nvidia. Huawei has been at the forefront of this pursuit, while companies like and Alibaba have also invested in semiconductor design.

Triolo stated that the difference in advanced hardware computing power will widen in the future, especially next year with Nvidia's Blackwell-based systems being restricted for export to China.

Baidu, a Chinese tech giant, has been systematically investing and growing their whole domestic AI infrastructure stack outside of Nvidia with high-performance AI chips.

Despite the potential ban on Nvidia chips in China, the country will still invest and develop their own infrastructure for building and training AI models.