On the last day of 2025, DeepSeek released a technical paper co-authored by founder and CEO Liang Wenfeng and 18 others, detailing "manifold-constrained hyper-connections, " a general framework for scaling AI system training that proposed promising directions for foundational models’ evolution. This release underscored Chinese AI companies' intensified focus on innovation amid the Christmas holiday season to maintain their competitive edge in a rapidly evolving field. Around the same time the previous year, DeepSeek gained significant attention after launching its large language model, DeepSeek-V3, followed weeks later by the reasoning model DeepSeek-R1 on January 20. Both models matched or outperformed rival systems on industry benchmarks while being developed at a fraction of the cost and computing power used by major U. S. tech firms. This triggered a massive tech stock sell-off on January 27, wiping out nearly US$1 trillion, with Nvidia alone losing US$600 billion. Analysts predict continued momentum for Chinese AI companies in 2026, fueled by Beijing's policy support, better funding prospects, broader AI adoption across industries, and a growing talent pool. A co-founder of a domestic AI startup, speaking anonymously, forecasted China would surpass the U. S. as the world’s leading AI power by 2027, highlighting China’s deep talent reservoir. Chinese President Xi Jinping’s New Year's address emphasized a domestic market vigorous with big AI models competing at the top and breakthroughs in semiconductor development, collectively propelling China into one of the fastest-growing innovation economies. Winston Ma, an adjunct professor at NYU School of Law focusing on AI and digital economy, noted that China’s tech innovation is poised for policy-driven growth in 2026, with AI central to economic and industrial upgrading plans. According to a recent Stanford University DigiChina Project report, China boasts over a dozen prominent developers of powerful open AI models beyond DeepSeek, including Alibaba Cloud (Qwen model), Moonshot AI, MiniMax, and Zhipu AI (Z. ai internationally). The report suggested China’s open-source AI models may have caught up or even surpassed U. S. counterparts in capability and adoption. Meta Platforms, for example, reportedly uses Alibaba Cloud’s Qwen model in training its upcoming model Avocado, expected in spring. Qwen is the world’s most popular open-source LLM series, powering various U. S. applications, with Airbnb CEO Brian Chesky stating the company heavily relies on these models for its AI customer service agent. In 2025, ChatGPT-style apps like Alibaba’s Qwen AI assistant, ByteDance’s Doubao, and Tencent’s Yuanbao increasingly became part of daily life in China, generally free to individual users. Xinhua reported that local Chinese government officials routinely used AI tools for drafting official documents, reflecting widespread AI integration. Market analyst Zhang Yi of iiMedia expressed strong confidence in China’s AI prospects for 2026, noting that AI development is a national priority, backed by China’s five-year plan and economic policy framework highlighting AI’s transformative potential in industrial and manufacturing efficiency.

Beyond chatbots, humanoid robots—referred to as embodied AI—are gaining traction; Unitree Robotics founder Wang Xingxing predicted major advancements within three to five years but cautioned household adoption may lag due to ethical and safety issues. Analysts at China Merchants Securities forecast humanoid robots as the ultimate AI application, though the "ChatGPT moment" for robotics has yet to arrive. Consumer AI devices, like AI glasses and smartphones, are also advancing. In November, Alibaba launched Quark AI Glasses, named after its AI assistant, aiming to capture the smart wearables market. Alibaba’s vice-president Wu Jia described AI glasses as revolutionary for human-computer interaction, comparable in importance to smartphones. ByteDance’s AI smartphone, co-developed with ZTE and powered by the Doubao AI, faced restrictions on voice capabilities by top Chinese apps, reflecting regulatory challenges. Chinese AI startups face the challenge of transitioning from innovation to profitability in 2026. Zhang noted that profitability will become the key metric for AI companies. MiniMax and Zhipu AI are preparing Hong Kong IPOs targeting HK$4. 19 billion and HK$4. 35 billion, respectively. Moonshot AI recently raised US$500 million in Series C funding, boosting its valuation to US$4. 3 billion, with no immediate IPO plans, according to founder Yang Zhilin. Meta Platforms’ acquisition of China-founded AI agent specialist Manus signals a new "cash-out" avenue for AI entrepreneurs beyond traditional IPOs, fostering innovation under major corporate umbrellas. Niu Zhangming, CEO of AI biotech firm MindRank, viewed this deal positively for the ecosystem, citing boosted confidence among entrepreneurs, investors, and scientists. Meta stated Manus serves millions globally and aims to bring leading AI agents to billions. Meanwhile, AI’s rapid growth has raised concerns among professionals like Adam Tang, a software engineer in Beijing. His employer encourages extensive use of Cursor, an AI coding assistant, ranking employees monthly based on usage to boost productivity and reduce costs. Tang fears AI tools like Cursor could eventually make software programmers obsolete. This coverage originally appeared in the South China Morning Post (SCMP), a leading authority on China and Asia, offering more insights via the SCMP app and social media channels. Copyright © 2026 South China Morning Post Publishers Ltd. All rights reserved.