In China, the domestic AI race intensifies as Chinese go gaga over DeepSeek

BEIJING/SHENZHEN – Chinese artificial intelligence (AI) sensation DeepSeek is having its moment in the sun, and users in China cannot get enough of its chatbot.

Over the eight-day Chinese New Year holiday that ended on Feb 4, ordinary people queried the start-up’s high-performance, free-to-use chatbot with their birth data – known as “bazi” or eight characters – and it became a fortune teller, advising them on love, life and wealth.

Others shared their discoveries on social media about how the DeepSeek-R1 reasoning model could carry out human-like conversations, recommend gym workouts and write poetry. And there are those who used DeepSeek’s various AI models to prepare for China’s highly competitive civil service examinations.

“DeepSeek has the effect of ‘breaking the circle’ – many ordinary people can now use such an advanced AI for free, raising their confidence in using such technology,” Mr Li Ruochong, 23, a master’s student in AI at the Hong Kong University of Science and Technology (Guangzhou), told The Straits Times.

The term “po quan”, or breaking the circle, is used online to describe something that has gained mainstream popularity, beyond a smaller circle of fans or enthusiasts.

Yet, DeepSeek’s chatbot is not the first Chinese generative AI model on the scene. The start-up was founded in 2023, around the same time that Chinese tech giants such as e-commerce platform Alibaba, search engine Baidu and TikTok creator ByteDance were launching their AI models – Alibaba

with its open-source Qwen,

Baidu with Ernie Bot and ByteDance with Doubao.

What

shot DeepSeek to fame

internationally and at home were its V3 large language model (LLM) and R1 reasoning model, released in the last two months, which have comparable results with the world’s best such as the US’ ChatGPT o1 but developed at a fraction of the cost, and without the most advanced chips.

Now, DeepSeek’s achievement could catalyse the race in China to develop ever more powerful and useful AI models, by attracting more investment and talent into the industry.

Already, some of its competitors like Alibaba, ByteDance and Moonshot AI have released updates of their models that are giving DeepSeek – and the US’ OpenAI – a run for its money.

All eyes are on whether China can keep up the pace of such breakthroughs to give the country a leg-up in its

ambition to be the global AI leader by 2030.

The race in China

Since OpenAI’s ChatGPT brought generative AI technology into the public consciousness in late 2022, China has been playing catch-up. Companies have raced to develop their own models, while the government has also supported development in the field.

China has invested in AI companies through government funds, including a 60 billion yuan (S$11 billion) national AI industry investment fund announced in January 2025. Between 2000 and 2023, one US study found, Chinese government venture capital funds invested US$184 billion (S$249 billion) in more than 9,600 companies in the AI space.

The government has also built AI infrastructure such as data centres and computing clusters.

Today, nine of the top 20 spots in Chatbot Arena, a benchmarking and ranking platform for LLMs popular with developers around the world, are occupied by Chinese companies (and 11 by the US).

Tech analyst Rui Ma, who runs the Tech Buzz China newsletter, said that if the latest frontier model is taken as the benchmark, then Chinese models have narrowed the gap with the best internationally, including those from the US.

“What is clear... is how far China has come since 2022, when ChatGPT launched and China had no real equivalent – mainly because very few researchers in the country had been working on LLMs. Today, it has a deep and formidable talent pool actively competing at the highest levels,” she said.

It would seem that the Chinese government’s policies of support and funding for the sector have gone some way in closing the gap between China and the West in generative AI models.

Leaders in China’s LLM start-up scene are known as the “six little tigers” – Zhipu AI, MiniMax, Baichuan AI, Moonshot AI, StepFun and 01.ai – whose founders have ties to academia or top tech companies such as Google and Microsoft. Valued at more than US$1 billion each, they raise funds from a mixture of private venture capital, larger companies and in some instances state-backed entities.

Breaking the mould

Yet, the breakthrough AI model came from DeepSeek which, according to some, succeeded not because of the innovation system nurtured by the government but in spite of it.

DeepSeek is seen as an outlier in the established ecosystem. It operates more like a passion project by a young and talented team, with little consideration given to commercialisation of their technology, and without profit-making pressures faced by bigger companies.

Mr J.S. Tan, a PhD student at the Massachusetts Institute of Technology who studies innovation policies in China, noted on media platform Substack that the company did not rely on state-backed initiatives or investments from tech incumbents.

Instead, its founder Liang Wenfeng personally funded its operations, Mr Tan wrote. Mr Liang became a billionaire after establishing the hedge fund High-Flyer in 2015.

“This allowed it to operate independently of the constraints often associated with state or corporate funding,” Mr Tan wrote in the Jan 29 post.

Asked why DeepSeek does not simultaneously focus on developing models and potential applications, Mr Liang, in a July 2024 interview with the Chinese media, said he believes China must shift from being a beneficiary of technology to a contributor, as its economy grows. “Our goal is not to turn a profit quickly, but to drive the technological frontier and fundamentally promote the growth of the entire ecosystem,” he had said.

Can DeepSeek’s success be replicated?

DeepSeek’s success was encouraging for Chinese AI companies because it was built partially on previous LLM work from China, including Alibaba’s open-source Qwen, said AI researcher Neil Zhu.

But “it may be very hard” for other AI companies in China to replicate DeepSeek’s successful organisational structure, which helped it achieve breakthroughs, said Mr Zhu, who is also the founder of the Centre for Safe AGI, a Shanghai-based non-profit that works with partners in China to devise ways in which artificial general intelligence can be safely deployed.

He was referring to the company’s flat and nimble set-up, and Mr Liang’s hands-on approach to research.

The start-up could also afford to take a longer-term approach in its research given its founder’s resources, and it had more computing power than many other companies in the field, he added.

Still, DeepSeek’s success pressures state-funded players to adapt and innovate, while opening new avenues for collaboration and investment, said Professor James Pang, who teaches AI and digital transformation at the NUS Business School.

“This dynamic landscape is poised to accelerate AI development in China, strengthening both its domestic industry and global competitiveness,” he told ST.