Chinese AI progress and top companies

Chinese AI is not just DeepSeek. Nearly half of the world’s top AI researchers come from China, and only 18% come from the United States. China owns 65% of the world’s artificial intelligence patents.

Chinese AI

The emergence of DeepSeek has suddenly aroused the general public’s great interest in the development of Chinese AI. In fact, this is not news at all, nor is it an achievement that suddenly emerged.

DeepSeek is not the only one

Some sour grapes and ignorant frogs in the well who are not convinced will certainly mock China for having only come up with DeepSeek in the field of artificial intelligence after all its efforts. This is just luck and coincidence, a special situation, and cannot be regarded as any great achievement of China in the field of artificial intelligence.

For more information about DeepSeek, please refer to my previous post “DeepSeek routed the global AI and stock

Academic achievements

half world’s top AI researchers from China

In March 2024, according to the latest research by the think tank MacroPolo, nearly half of the world’s top AI research talents are from China, and only 18% are from the United States. Three years ago, China accounted for only about one-third of the total talent, while the United States accounted for a similar proportion. Today, 38% of the top AI research talent in the United States are from China, while Americans account for only 37%; three years ago, 27% were from China, while Americans accounted for 31%.

A 2020 New York Times article pointed out that American AI institutions and companies account for 60% of the world’s top researchers. This is six times the 10.6% of China, which ranks second, and six times the 10.2% of Europe, which ranks third. But are these 60% produced in the United States? No. Native American researchers account for only 31%. Another major force is researchers from China, accounting for 27%.

China holds 65% of AI patents

Analyzing the sources of papers adopted by the three top AI societies, among the top 100 institutions in terms of number of authors in 2024, 31 were from mainland China, second only to 37 from the United States.

In September 2024, according to the latest “2024 Artificial Intelligence Index Report” released by Stanford University, more than 61.1% of the authorized artificial intelligence patents worldwide are from China, while 20.9% are from the United States. In terms of the total number of patent authorizations, China is nearly three times that of the United States.

From 2014 to 2023, China’s number of AI patent applications has far surpassed the United States, South Korea, Japan, India and other countries. China has more than 38,000 generative artificial intelligence inventions, which is 6 times that of the second-ranked United States.

On July 3, the World Intellectual Property Organization released the “Generative Artificial Intelligence Patent Landscape Report”. The report shows that from 2014 to 2023, Chinese inventors applied for the largest number of generative artificial intelligence patents, far exceeding the United States, South Korea, Japan, India and other countries. From 2014 to 2023, the number of invention applications related to generative artificial intelligence worldwide reached 54,000, of which more than 25% appeared in the past year alone. Generative AI allows users to create content such as text, images, music and computer code, powering a range of industrial and consumer products. Between 2014 and 2023, China’s generative artificial intelligence inventions exceeded 38,000, six times that of the second-ranked United States.

China prefers open source model

China’s three best-performing AI models currently are: DeepSeek, Alibaba, and Baidu, all of which use open source models. In contrast, among the United States’ major competitors, only Meta and X.ai’s Glok adopt an open source model.

China’s Tech Behemoths on AI

Alibaba

The latest version “2.5-Max” of the artificial intelligence model “QWen” released by Alibaba’s Alibaba Cloud one month ago has surpassed DeepSeek’s “V3”.

On February 11, 2025, according to a report by technology media The Information citing sources, Apple and Alibaba will work together to develop AI features for Chinese iPhone users. Apple began testing different AI models from well-known Chinese AI developers in 2023 and chose Baidu as its main partner last year, but the cooperation was later canceled because Baidu’s progress in developing models for Apple’s intelligence did not meet the standards of American companies.

Alibaba’s Tongyi Qianwen Qwen series of large models have performed well in the open source field and performance evaluation, and are in the first echelon of the industry. Apple eventually gave up on DeepSeek, which had been highly sought after recently, because the DeepSeek team lacked the manpower and experience required to support a large customer like Apple.

iOS18.4 may be officially released in April 2025, and the Chinese version of Apple Intelligence will also be officially unveiled in this version.

ByteDance

Tik Tok itself is a company that has swept the global social network in just a few years by relying on its excellent artificial intelligence algorithms. Its parent company, ByteDance, is also among the best in the world in terms of development, achievements, and investment in artificial intelligence.

An upgraded version of its flagship AI model, Doubao-1.5-pro, was released during the Chinese Lunar New Year and is said to be able to outperform OpenAI’s o1 in some tests. In addition to performance, products released by Chinese AI companies also challenge American competitors on price. For example, the most powerful version of its AI chat program Doubao is priced at US$9 per million units, which is only half of the DeepSeek DeepSeek-R1 version. In comparison, OpenAI’s o1 costs $438 for the same purpose.

In early February, ByteDance launched OmniHuman-1, an advanced artificial intelligence system that can generate ultra-realistic deepfake videos using only an image and audio. The system is trained on 19,000 hours of video data and can accurately give vivid dynamic effects to various objects, from cartoon characters to complex human movements, all of which can be presented vividly. This technology pushes the boundaries of AI-generated content, enabling smooth and detailed motion synthesis.

Baidu

Baidu is the Chinese version of Google, with three decades of accumulated Internet search and China-specific Internet data, which are moats that cannot be matched by its American or Chinese counterparts. This is also the reason why when Apple Intelligence was launched in China, the first partner considered was Baidu.

In June 2024, an IDC report stated: Baidu Ernie artificial intelligence large model is ahead of OpenAI GPT-4 and other Chinese and American competitors in 7 evaluation projects. By December 2024, half of China’s mobile phone manufacturers will use the Ernie model. As of November 2024, the user scale of Wenxin Yiyan is 430 million, and the average daily call volume of Wenxin big model exceeds 1.5 billion times, an increase of more than 30 times compared with 2023.

Apollo, an independent company under Baidu, is the startup that launched the Apollo Go autonomous taxi. Investors can regard it as the Chinese version of Waymo, a subsidiary of Google. The main core technology of automatic driverless taxis is artificial intelligence.

Huawei

Huawei’s greatest contribution to artificial intelligence so far is to help Chinese manufacturers break through the US embargo on China’s high-end artificial intelligence chips.

In June 2024, Huawei launched the cutting-edge Ascend 910B AI chip, which can achieve 80% of the efficiency of Nvidia A100 when training large-scale artificial intelligence language models, and has surpassed Nvidia A100 AI GPU by 20% in specific test performance. Leading mainland companies including Tencent Holdings and Baidu have also purchased the Ascend 910B chip.

Huawei itself has also made great achievements in the development of artificial intelligence. In 2020, it announced that it would officially open source its AI computing framework MindSpore. The Pangu model was announced to the public in 2023.

Tencent

The flagship AI model released by Tencent is the text-to-video generator Hunyuan. In January 2025, Tencent stated that the performance of its Hunyuan-Large model is comparable to the Llama 3.1 model launched by Meta in July 2024, and the computing power required for training is only 1/10 of that of Meta. In May 2024, some of the Chinese capabilities of Tencent’s Hunyuan-Large model have caught up with GPT-4 and support 16-second video generation.

Pure AI software startups

Moonshot AI

Moonshot AI was founded in 2023. As of January, its corporate valuation had reached $3.3 billion, according to data from US research firm CB Insights.

Just two hours after DeepSeek released its latest generative AI large language model (LLM) “R1” on January 20, Dark Side of the Moon also released its LLM “Kimi k1.5”, claiming that its reasoning capability exceeds that of Anthropic’s “Claude 3.5 Sonnet” and its performance is comparable to OpenAI’s “OpenAI o1”.

Currently, the company’s product Kimi k1.5 is an upgraded version of the Kimi launched in October 2023. It has attracted much attention because it is the first AI assistant that can process 200,000 Chinese characters at a time. It is the world’s first artificial intelligence assistant product that supports the input of 200,000 Chinese characters. In this regard, Dark Side of the Moon once stated that Kimi’s ability has been further upgraded and can process up to 2 million Chinese characters. As of October 2024, Kimi’s monthly active users exceeded 36 million.

Zhipu AI

Zhipu AI, founded in 2019, is valued at US$3 billion. It was added by the US government on January 15 to become one of more than two dozen Chinese entities included in the US trade restriction list. Zhipu was specifically included in the list because it assists China’s military advancement through AI development.

But Zhipu opposed this and said the accusation lacked factual basis. And putting aside the accusations of providing China with military advances, Zhipu is making rapid progress in the field of AI. Its latest product is AutoGLM, an AI assistant app released in October 2024 that helps users operate their smartphones through complex voice commands.

Baichuan Intelligence

Baichuan AI was founded by Sogou founder Wang Xiaochuan in 2023 and is currently valued at US$2.8 billion.

Unlike companies such as DeepSeek and 01.AI that emphasize global competition, Baichuan focuses on the domestic Chinese market and concentrates on developing fields such as medical assistance and health care. Developed the Baichuan model and a chat robot named “Bai Xiaoying”. It excels at processing very long texts, and its latest version of the Baichuan4 large model is said to surpass OpenAI’s GPT-4 in Chinese capabilities.

MiniMax

MiniMax was founded in 2021 by Junjie Yu, a former senior R&D director at Chinese facial recognition company SenseTime. The company focuses on developing the ABAB large model, and its main products are AI companion products. The current valuation is $2.5 billion.

“Xingye”, the most downloaded AI companion app in China, and “Talkie”, the most downloaded AI companion app in the United States, are both products of MiniMax, which also has an AI chatbot called “Hailouwen”.

01.AI

01.AI was founded by Kai-Fu Lee in 2023. Kai-Fu Lee, who has served as a senior executive at Microsoft and Google for a long time, was once the chairman of venture capital firm Sinovation Ventures, focusing on promoting the birth of AI startups. Born in Taiwan, he was once a naturalized American citizen. In 2011, he renounced his American citizenship. The company is currently valued at over $1 billion.

Develop a series of large models of Yi and take the international open source route. Kai-Fu Lee emphasized at last year’s World Economic Forum that open source is the key for any country to fully develop the potential of AI.

The Yi-34B large model published in November 2023 surpassed the Llama2 large model of the American Meta company on the open source large model ranking Hugging Face. The Yi-VL-34B multimodal large model released in January 2024 supports image understanding and text interaction, similar to OpenAI’s GPT-4V.

Stepfun

Stepfun was founded in 2023 by Jiang Daxin, who was formerly the chief scientist of the Software Technology Center at Microsoft Research Asia. The company is currently valued at $2 billion.

The company focuses on the research and development of the Step series of open source large models and multimodal AI. It has quickly become an industry contender with its portfolio of foundation models. “Unlike many Chinese AI startups that have given up on their vision of artificial general intelligence (AGI), StepFun remains committed to building AGI,” MIT Technology Review wrote.

The latest Step-2 large model has more than 1 trillion parameters (OpenAI’s GPT-4 has about 1.8 trillion parameters).

Pure AI hardware startups

Not as good as China’s AI software achievements

In recent years, China has also seen the emergence of some pure AI hardware startups, mainly in the areas of AI chips, self-driving car chips, and GPUs. Their main rival is Nvidia. Their strength is not as good as China’s achievements in artificial intelligence software, and is even less than Nvidia, the world’s only leader in this field. This is true, but if you don’t force it, it’s enough for your own use.

AI chips by technology behemoths

The main players in China’s domestic AI chip industry include Huawei HiSilicon, Baidu, Alibaba Pingtou Ge, ByteDance and other manufacturers. The hardware and software configurations of these domestic AI chips are catching up, and some domestic chips can basically match Nvidia’s A800/H800 series products.

Cambrian Technologies

The most famous one is Cambricon Technologies, a listed company in China. Its business is mainly to build core processor chips for various smart cloud servers, smart terminals and smart robots for enterprises, provide users with IP licensing and chip services, and smart platform services.

When Cambrian was first established, it received an angel round of financing of tens of millions of RMB from the Chinese Academy of Sciences. In August 2016, it received a Pre-A round of financing from Yuangan Yuandian, iFlytek, and Yonghua Capital. In August 2017, Cambricon announced the completion of a US$100 million Series A financing round, with participation from Alibaba, Lenovo, and Turing Technology. This round of financing made Cambricon the first unicorn company in the global AI chip field.

The company has been making losses for many years since its listing in 2020. Its revenue in 2024 will be between 1.07 billion RMB and 1.2 billion RMB, a year-on-year increase of 50.83% to 69.16%; its net profit is expected to be a loss of 396 million yuan to 484 million RMB.

According to the “2024 Hurun China Top 50 Artificial Intelligence Enterprises”, Cambrian ranked first with a value of 238 billion yuan. After Cambrian’s stock price rose nearly fourfold in 2024, it has continued to soar since 2025, with its market value reaching the 300 billion RMB mark for the first time; on the 10th, the stock price closed at 729.97 RMB , a record high.

BiRen Technology

Biren Technology was founded in 2019. Its products mainly focus on graphics processing units (GPUs) and dedicated accelerators for driving safety alert systems (DSA). In terms of development path, BiRen Technology first focused on general intelligent computing in the cloud, and gradually expanded into multiple fields such as AI artificial intelligence training and reasoning.

BiRen Technology has attracted much attention since its establishment. It has not only established a luxurious technical team, but also obtained RMB 5 billion in funding in a short period of time.

Jingjia Microelectronics

Although Jingjia Micro is the leading GPU company in mainland China, its technology and output value are still far behind those of the world’s top company Nvidia.

Jingjiawei was listed on the Shenzhen Stock Exchange’s Growth Enterprise Market in March 2016. The company launched its first generation of low-power GPU products in 2014.

The second generation was launched in 2018, and the latest third-generation product JH920 was released at the end of 2021, emphasizing that it can be used in PCs, servers, graphics workstations, etc.

Muxi Tech

MetaX Tech was founded in Shanghai in September 2020. It focuses on AI chips and has launched the Xisi N series GPUs for intelligent computing reasoning, Xiyun C series GPUs for general computing, and Xicai G series GPUs for graphics rendering.

Muxi products use self-developed GPU IP, have completely independent instruction sets and architectures, and are equipped with a complete software stack compatible with the mainstream GPU ecosystem.

Horizon Robotics

Horizon Robotics was founded in 2015 and went public in 2024. It is a supplier of advanced driver assistance systems (ADAS) and advanced autonomous driving (AD) solutions for vehicles, with proprietary software and hardware technologies. It is currently the largest autonomous driving chip company in China, with revenue of 1.6 billion yuan in 2023.

Horizon has indeed made a number of breakthroughs in the past few years. Data shows that in 2023, Horizon ranked second in both China’s high-end and low-end intelligent driving markets, with markets shares of 35.5% and 21.3% respectively. The number one in the high-end intelligent driving market is Nvidia, with a market share of 49%. The number one in the low-end intelligent driving market is Mobileye, with a market share of 26.6%.

Black Sesame Technologies

Horizon Robotics and Black Sesame Intelligence, although the two companies were founded only one year apart and planed to go public in the same year, their situations are completely different. Horizon’s revenue soared 71.3% in 2023 to RMB 1.552 billion. However, Heizhima Intelligence only has 312 million yuan, and its cash flow is even tight.

Biometrics AI

Leading the world

China is far ahead of other countries in the world in biological AI, especially facial recognition; in particular, the ecosystem of the entire industrial chain is very complete. The quality, price, and performance are leading in various countries. Not only has it been widely used in China for many years, it has also been exported to countries around the world, defeating competitors around the world.

The four main companies

  • SenseTime: A leader in China’s computer vision field, providing technologies such as facial recognition and smart security.
  • Megvii: Expertise in image processing and edge computing, applied to smart cities.
  • Yitu Technology: Focuses on medical imaging AI and provides diagnosis assistance systems for hospitals.
  • CloudWalk: Focuses on smart finance and smart security, and cooperates with Chinese state-owned banks.
Chinese AI
credit: Gemini

Related articles

Disclaimer

  • The content of this site is the author’s personal opinions and is for reference only. I am not responsible for the correctness, opinions, and immediacy of the content and information of the article. Readers must make their own judgments.
  • I shall not be liable for any damages or other legal liabilities for the direct or indirect losses caused by the readers’ direct or indirect reliance on and reference to the information on this site, or all the responsibilities arising therefrom, as a result of any investment behavior.
error: Content is protected !!