Everyone is Waiting for Liang Wenfeng

Giants burn cash, models fly

China’s internet titans are openly wagering on who will own the next generation of user attention. Tencent (腾讯) handed out RMB 1 billion in red packets, Baidu (百度) offered RMB 500 million to promote its Wenxin assistant, Alibaba (阿里巴巴) splashed out RMB 3 billion around its Qwen family, and ByteDance (字节跳动) pushed BeanBao (豆包) features and prizes to pull users into everyday AI habits. Product releases have been rapid too: Seedance 2.0, BeanBao 2.0 and Qwen‑Image 2.0—new models, new hooks into commerce and content—arrived within weeks. Who can capture real-world use cases and become the “super‑entry” to everyday life? That question is driving the scramble.

Where is DeepSeek?

In that noise, DeepSeek—founded by Liang Wenfeng (梁文锋)—has grown strangely quiet. It was DeepSeek’s surprise open release of R1 last year that shocked the global AI community: reportedly matching top-tier inference performance at a fraction of the cost, R1 challenged the assumption that only enormous hardware budgets could produce world-class models. Yet as rivals roll out aggressive consumer gambits, DeepSeek has largely stayed silent. It has been reported that The Information said DeepSeek planned a V4 flagship timed around the Lunar New Year but then held back; the company briefly increased its context window from 128K to 1M tokens on February 11 and has since kept product and timing details close to the chest.

Liang’s backstory and the stakes

Liang, born 1985 in Zhanjiang and educated at Zhejiang University, moved from quantitative trading powerhouse Huanfang (幻方量化) to build DeepSeek with the explicit aim of “deep searching” for new model architectures and techniques. He is reportedly worth more than RMB 10 billion and has kept a low public profile, yet his technical papers in January — on manifold‑constrained hyperconnections (mHC) and conditional memory (Engram) — have fueled market speculation that DeepSeek’s next step could be substantive. Against a backdrop of Western export controls on advanced AI chips and intense domestic competition to reduce hardware dependence, a software‑efficient breakthrough would be strategically potent.

Why everyone waits

This is not just a product race. It is a fight to define the next decade of China’s internet: who builds the interfaces people use daily, who controls attention, and who turns AI into habit. Giants buy scale with subsidies and distribution. Startups and research‑driven players like DeepSeek chase efficiency and architecture breakthroughs. Will Liang and his team reappear with a technical and product leap that redraws the map? For investors, competitors and millions of users, the answer will determine who sets the rules in China’s fast‑evolving AI ecosystem.