On the last day of 2025, DeepSeek published a new technical paper, with founder and CEO Liang Wenfeng among the 19 co-authors, about “manifold-constrained hyper-connections” – a general framework for training artificial intelligence systems at scale, which suggested “promising directions for the evolution of foundational models”.
That release was a fitting reminder to the world, especially during the peak of the Christmas holiday season, about Chinese AI companies’ sharpened focus on innovation to stay ahead in this fast-developing industry.
It was around this time last year when DeepSeek started to get widely noticed days after releasing its namesake large language model (LLM), DeepSeek-V3. Weeks later on January 20, reasoning model DeepSeek-R1 was released.
Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.
The two models either surpassed or matched the performance of rival models across a range of industry benchmark tests. They were also built at a fraction of the cost and computing power that major US tech companies invest to build LLMs. The result: a massive sell-off on January 27 wiped out nearly US$1 trillion in tech stocks, including US$600 billion from Nvidia alone.
Analysts expected the momentum of Chinese AI companies to continue this year, thanks to Beijing’s policy support, improved funding prospects, greater adoption of AI systems across industries and a growing number of talent being recruited for innovative projects.
A domestic AI start-up’s co-founder, who declined to be identified, predicted that China would overtake the US to become “the world’s leading AI power in 2027”. China’s advantage was its deep talent pool, according to the co-founder.
In his New Year’s address, Chinese President Xi Jinping pointed out that the domestic market has “many large AI models competing in a race to the top”, while new breakthroughs were being achieved in domestic semiconductor development. All of that “has turned China into one of the economies with the fastest growing innovation capabilities”, Xi said.
DeepSeek founder and CEO Liang Wenfeng. Photo: Shutterstock alt=DeepSeek founder and CEO Liang Wenfeng. Photo: Shutterstock>
“China’s tech innovation is poised for policy-driven growth in 2026, with AI placed at the centre of the country’s economic agenda and industrial upgrading plans,” said Winston Ma, an adjunct professor at the New York University School of Law, with a focus on AI and the digital economy.
China currently has a strong AI bench of more than a dozen players developing powerful open models beyond DeepSeek, according to a report last month by Stanford University‘s DigiChina Project, which is under the school’s Centre for International Security and Cooperation, and its Institute for Human-Centered Artificial Intelligence, which compiles the annual AI Index report.
Those included Qwen model developer Alibaba Cloud and fast-growing start-ups such as Moonshot AI, MiniMax and Zhipu AI, known internationally as Z.ai. Alibaba Cloud is the AI and cloud computing services unit of Alibaba Group Holding, owner of the Post.
China’s open-source AI models may have already caught up or “even pulled ahead” of their US counterparts in capabilities and adoption, the report found.
Alibaba Cloud is the artificial intelligence and cloud computing services unit of Alibaba Group Holding. Photo: Shutterstock alt=Alibaba Cloud is the artificial intelligence and cloud computing services unit of Alibaba Group Holding. Photo: Shutterstock>
Facebook parent Meta Platforms, for example, was said to be using Alibaba Cloud’s open-source Qwen model as part of the training process for a new model code-named Avocado, which was expected to be released in the spring.
The report did not specify which Qwen model was being used by Meta.
Qwen is the world’s most popular series of open-source LLMs, with many US businesses using it to develop their own applications. Brian Chesky, co-founder and CEO of San Francisco-based Airbnb, said the company “relies heavily” on Alibaba’s Qwen models to power its AI-driven customer service agent.
In 2025, ChatGPT-style apps – including Alibaba Cloud’s Qwen AI assistant, ByteDance-owned Doubao chatbot and Tencent Holdings‘ Yuanbao – increasingly became part of everyday life in China, as these were typically free of charge for individual users.
An open secret within China’s provincial governments was that local cadres increasingly relied on AI tools to draft official documents and reports, according to the official Xinhua News Agency‘s report in September.
Zhang Yi, the founder and chief analyst at market consultancy iiMedia, said he was bullish about China’s AI prospects in 2026 as the technology’s development was “a national priority”.
China’s central economic work conference, which charts policies, and the country’s 15th five-year plan from 2026 to 2030, highlighted AI’s potential to transform and make more efficient China’s massive industrial and manufacturing industries, according to Zhang.
Meta Platforms is reportedly using Alibaba Cloud’s open-source Qwen model as part of the training process for a new model code-named Avocado. Photo: Shutterstock alt=Meta Platforms is reportedly using Alibaba Cloud’s open-source Qwen model as part of the training process for a new model code-named Avocado. Photo: Shutterstock>
Apart from chatbots, humanoid robots have become a major application for AI in China, where such implementation is called embodied AI.
Unitree Robotics founder Wang Xingxing in August predicted significant progress for these machines within three to five years because of AI, although he cautioned that it may take longer for humanoids to enter households because of ethical and safety concerns. “In the long run, the surge of humanoid robots this year could be just a small spark, akin to the birth of the internet many years ago,” he said.
In a research note published this week, analysts at China Merchants Securities projected humanoid robots to become the ultimate application for AI models.
Still, Wang reiterated his earlier assertion that the “ChatGPT moment” for China’s robotics industry had yet to arrive.
Other AI-powered devices, meanwhile, are also being promoted to consumers as Chinese innovation in this field continued to progress. On-device AI products – such as AI glasses and AI smartphones – are expected to thrive in the next few years, iiMedia’s Zhang said.
Alibaba in November launched its first AI glasses – the Quark AI Glasses, named after the company’s Quark AI assistant – in China in a bid to carve out a solid niche in the growing smart wearables market.
“AI glasses are the intelligent devices that truly usher in a revolution in human-computer interaction in the AI era,” Alibaba vice-president Wu Jia said at the product launch. “In this respect, their importance is no less than that of mobile phones.”
Alibaba Group Holding’s Quark AI Glasses are named after the company’s Quark AI assistant. Photo: Handout alt=Alibaba Group Holding’s Quark AI Glasses are named after the company’s Quark AI assistant. Photo: Handout>
ByteDance, meanwhile, triggered controversy with a test version of its agentic AI smartphone made in partnership with ZTE. The company said it had scaled back the capabilities of Doubao, the agentic AI that runs on the handset, after several of China’s most widely used apps restricted its voice-operated functions.
One challenge for a number of Chinese AI start-ups in 2026 would be generating profits as they transform their technology into usable products.
“A tipping point will emerge in 2026, as profitability becomes the key metric to measure an AI company,” iiMedia’s Zhang said.
After they complete their Hong Kong listing efforts, MiniMax and Zhipu AI are expected to show they can steadily make money to their investors, according to Zhang.
MiniMax on Wednesday kicked off its Hong Kong initial public offering, seeking to raise up to HK$4.19 billion. Zhipu AI on Tuesday started its listing process, with an eye on raising HK$4.35 billion.
Moonshot AI, meanwhile, raised US$500 million in its recent Series C funding round, according to a report on Wednesday by the Chinese technology news outlet LatePost.
That fresh injection of capital pushed Moonshot AI’s valuation to US$4.3 billion, according to the report. With its financial war chest, Moonshot AI was “in no rush for an IPO in the short term”, founder and CEO Yang Zhilin said in an internal letter dated Wednesday.
China-founded Manus, an artificial intelligence agent specialist, now forms part of the operations of new owner Meta Platforms. Photo: Shutterstock alt=China-founded Manus, an artificial intelligence agent specialist, now forms part of the operations of new owner Meta Platforms. Photo: Shutterstock>
This week, Meta’s acquisition of China-founded AI agent specialist Manus for unspecified billions of dollars opened what some see as a “cash-out” route for AI entrepreneurs beyond the traditional IPO playbook, while continuing to push innovation under a large corporate parent.
“The overall sentiment has been positive and optimistic,” said Niu Zhangming, CEO of AI-driven biotech firm MindRank, adding that the deal could be a “significant boon for the ecosystem” by lifting confidence among entrepreneurs, investors and scientists.
Meta said in a statement that it was taking in Manus “to bring a leading agent to billions of people and unlock opportunities for businesses across our products”. It said Manus was already serving the daily needs of “millions of users and businesses worldwide”.
For Adam Tang, a 39‑year‑old software engineer in Beijing, the rapid expansion of AI capabilities has brought growing unease, as the pace of innovation has increased.
Tang’s employer, a multinational research facility in Beijing, has been encouraging the use of Cursor, an AI coding assistant popular among software developers, to improve productivity and save costs across the organisation.
He said the team he was on ranked staff monthly on the basis of who used Cursor at work more, an apparent effort to spur adoption of the AI coding tool that can turn natural language instructions into production-ready code in seconds.
Tang said his worry was that AI tools, such as Cursor, could render the role of software programmers obsolete in the not-so-remote future.
This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP’s Facebook and Twitter pages. Copyright © 2026 South China Morning Post Publishers Ltd. All rights reserved.
Copyright (c) 2026. South China Morning Post Publishers Ltd. All rights reserved.














