From manufacturing to consumer electronics to high-tech transportation, China has earned a dominant global position. Hence, it isn’t surprising that the country is now on a mission to dethrone the United States in the AI race.
While OpenAI, Microsoft, Google, and even Anthropic, for that matter, are prominent names that frequently come up in most AI discourses, the Chinese ecosystem remains a less-explored aspect of the conversation. While DeepSeek has captured global attention, it is only the tip of the iceberg.
To begin with, it is crucial to recognise some of the other top tech giants in China – ByteDance, Tencent, Alibaba, Baidu and Huawei. What are they all up to?
Tencent
Tencent, based in Shenzhen, is the company behind the popular instant messaging platform WeChat. The company has developed the Hunyuan family of models, which includes large and small text generation models, a video generation model, and a 3D environment generation model.
Tencent’s latest offering is the Hunyuan3D 2.0, which turns images or text into three-dimensional models.
In November last year, Tencent launched its flagship language model, Hunyuan Large, which has 389 billion parameters. Its performance was on par with Meta’s Llama 3.1 405B, the company’s flagship AI model.
As reported by Omdia, it is estimated that Tencent, along with ByteDance, ordered about 2.3 lakh NVIDIA chips in 2024.
Tencent’s AI models are free and can be used open source in HuggingFace.
Alibaba
Chinese technology conglomerate Alibaba Group is one of the world’s biggest companies and, along with Tencent, only the second company to break the $500 billion valuation. The company’s co-founder, Jack Ma, is also widely revered as one of the most influential minds in the business world.
The company introduced its Qwen series of models in April 2023. Most recently, it unveiled the Qwen2.5 Max, which outperforms DeepSeek’s V3 model. The model can also control mobile and computer screens, similar to Anthropic’s Computer Use and OpenAI’s Operator.
On the other hand, the company has also built a vision language model called Qwen2.5-VL, which can understand charts, images, graphics, etc. Alibaba also offers a coding-centric model called Qwen 2.5 Coder. This model offers performance parity with OpenAI’s GPT-4o and Claude 3.5 Sonnet.
All of Qwen’s models can be accessed for free on the chatbot application on the web. With the exception of their latest Qwen2.5 Max, most of Qwen’s models are available for open-source use via Hugging Face.
The company also built a video generation model called Tongyi Wanxiang, which, according to Alibaba, scored impressive results in benchmarks.
As per recent reports, Alibaba has also planned to introduce AI features to iPhones in China.
Baidu
Baidu is one of the leading internet business companies in China and was the first Chinese company to be included in the NASDAQ-100 index. Beyond AI models, the company offers a comprehensive AI stack that includes chips, applications, and cloud infrastructure.
The company recently announced that it will be making its AI model ERNIE available for open-source use from June 30. ERNIE was released in 2023, and as of last year, Baidu revealed that it reached 300 million users. The ERNIE 4.0 is the company’s latest iteration, and it plans to release ERNIE 5 later this year.
Besides, the company has also built Apollo VTA, a vision foundational model designed for autonomous driving.
DeepSeek
DeepSeek, an AI research lab from High-Flyer Capital Management, has been in the news for building the DeepSeek-R1 model with a surprisingly low amount of capital and resources and yet, managing to beat most of the competition. In December last year, the company launched the DeepSeek-V3 model, which only required $5.5 million and 2048 NVIDIA GPUs for training.
Last year, a technical paper revealed that the most expensive publicly announced training runs to date are OpenAI’s GPT-4, which cost $40 million, and Google’s Gemini Ultra, which cost $30 million.
DeepSeek’s models are free to use, are available as open source, and can be accessed on chat.deepseek.com. More recently, the company announced Janus Pro, an AI image generation model that is claimed to offer better results than OpenAI’s DALL·E 3.
Recently, it was reported that both Tencent and Baidu are planning to integrate DeepSeek’s AI models into their consumer applications.
ByteDance
ByteDance is the parent company behind TikTok, one of the most popular social media apps. In recent years, it has been subjected to bans and restrictions. Amidst the hype DeepSeek has created, ByteDance has managed to stand out with its OmniHuman model.
This model can generate realistic videos from a single image while handling lifelike movements, natural gestures, and “stunning attention to detail”, as described by an official blog post. Recently, the company also unveiled a family of image and video generation models called Goku. Moreover, as mentioned earlier, ByteDance, along with Tencent, purchased 2.3 lakh NVIDIA chips in 2024.
Recently, ByteDance also launched a coding environment called Trae AI, which is available for free and offers similar features like Cursor, Windsurf and so on.
Besides, ByteDance also develops the Doubao family of AI models, which consists of an exhaustive list of AI models, including language, character generation, speech recognition and sound replica models.
It’s Not Big Players Alone
While the Chinese AI model ecosystem has been dominated by big players, several other startups are making waves in addition to DeepSeek. For instance, Kling AI is a capable video generation model, and its latest iteration was launched in December last year. Kling is also popularly regarded as an alternative to OpenAI’s Sora.
MiniMax, another text-to-video generator launched by a Chinese startup, has recently been recognised by AIM for some of its best AI-generated videos.
Moreover, Chinese startup Moonshot unveiled its latest iteration of the Kimi 1.5 model, which offered o1-level performance.
Recently, a startup called StepFun announced an open-source video generation model known as Step-Video-T2V and an open-source voice interaction model called Step-Audio.
Step-Video-T2V is now the world’s largest open-source video generation model. It comprises 300 billion parameters and can generate high-quality videos with 204 frames at 540P resolution.
Besides, Chinese Prime Minister Xi Jinping hosted a landmark event involving leaders of AI in the country yesterday.
Besides, Chinese President Xi Jinping hosted a landmark event involving leaders of AI in the country yesterday.
China is Going Big on AI
Jinping recently met with leading private sector business figures to discuss boosting the country’s economy, which has struggled with issues like the pandemic, regulatory crackdowns, and a real estate crisis. The agenda, according to sources, was also to encourage innovation and confidence in the country’s economic model.
Attendees included Alibaba Group co-founder Jack Ma, Contemporary Amperex Technology (CATL) chairman Zeng Yuqun, BYD chairman Wang Chuanfu, Tencent CEO Pony Ma and Huawei founder Ren Zhengfei.
Notably, Chuanfu and Zhengfei were photographed seated directly in front of Jinping, a position of honour for leaders in electric vehicles and chip development. Sources told Reuters that Liang Wenfeng, founder of DeepSeek, also attended.
Moreover, this was also the first time Ma was seen in public with Jinping. In 2020, his speech criticising Chinese regulators led to the suspension of Ant Group’s stock market debut and significant restructuring of the company. Alibaba was also fined $2.8 billion for antitrust violations. Since then, Ma was barely visible to the public eye.
