A researcher at the Chinese Academy of Sciences kept coming back to one awkward question in 2011: had any of the clever algorithms his team built ever been used by a real person, for a real task? A few years later he walked away from a stable research career to find out. The company he started would one day ring the opening bell on one of China's most demanding stock exchanges — and he would barely tell anyone.
Key takeaways
- Zhou Xi (周曦) founded CloudWalk (云从科技) in March 2015, after building a facial-recognition group inside the Chinese Academy of Sciences (中国科学院).
- CloudWalk listed on Shanghai's STAR Market (科创板) on 27 May 2022 (stock code 688327), becoming the first of China's so-called "AI Four Little Dragons" (AI四小龙) to reach that board.
- His public pitch has stayed consistent for a decade: the test of artificial intelligence is not a benchmark score but whether it is actually put to use.
- The company pivoted from single-point vision tools to a "human-machine collaboration operating system" (人机协同操作系统) and, since 2023, to a large model (大模型) called Congrong (从容).
From the lab bench to a company
Zhou Xi was not a typical startup founder. He studied at the University of Science and Technology of China, then earned a doctorate in computer science at the University of Illinois at Urbana-Champaign. He returned to China in 2011 under the Academy's "Hundred Talents" (百人计划) program and set up a smart-multimedia research center in Chongqing. There, he realized the gap that would define his career: academic results were piling up, but deployment was almost nowhere.
"That question — can we truly put the technology to use? — was the starting point," Zhou has said in CloudWalk's own published profiles. It is a line he repeats more often than any product name, and it helps explain a career that looks, from the outside, like a slow march from science toward the messy work of selling software.
In 2015 he left the Academy and incorporated CloudWalk in Guangzhou, seed-funded partly by local government-linked funds. The name, a loose play on "cloud" and "walking," signaled an ambition to put perception everywhere.
A platform, not just a face
CloudWalk belongs to a cohort Chinese press dubbed the "AI Four Little Dragons" (AI四小龙): SenseTime (商汤), Megvii (旷视), Yitu (依图), and CloudWalk. All four cut their teeth on computer vision. CloudWalk's early wins were in banks and airports — remote account opening, identity checks, and the kind of back-office automation that is invisible to passengers but essential to operators.
What set CloudWalk apart, in Zhou's telling, was reluctance to stay a "face recognition company." Around 2018–2020 the firm began packaging its capabilities into a "human-machine collaboration operating system" (人机协同操作系统), meant to sit beneath finance, governance, mobility, and commerce applications rather than live inside one narrow product. The shift was also defensive: single-modality vision firms were being commoditized, and a platform story sells better to large enterprises than a point tool.
The company carried that logic into the large-model era. In 2023 CloudWalk released Congrong (从容), a large model (大模型) positioned for multimodal and agent-style tasks, and has since tied it to industry "intelligent agents" (智能体) in ports, customer service, and operations.
The "three waves" frame
Zhou is associated with a simple mental model he calls the "three waves of AI" (AI三浪理论), described in CloudWalk's interviews:
- Wave one — single-point breakthroughs: face recognition, speech, and other isolated capabilities find their first narrow uses.
- Wave two — a closed loop: technology combines into systems that deliver measurable industry value.
- Wave three — embedded collaboration: human and machine intelligence are woven into everyday business processes, raising both efficiency and experience.
It is less a forecast than a description of the path CloudWalk itself tried to walk — from a winning algorithm, to a product suite, to an operating layer. The honesty of the framework is that it admits most companies get stuck between wave one and wave two.
The listing, and the quiet founder
On 27 May 2022, CloudWalk debuted on the STAR Market (科创板) at code 688327, the first of the Four Little Dragons to list there. Shares opened more than 50% above the issue price on the first day, a moment of validation for a firm that had spent years explaining why perception software deserved a platform valuation.
Zhou, characteristically, stayed in the background. Colleagues and company profiles describe a founder who "rarely gives interviews" and distrusts the cult of the founder's persona — a view he has voiced publicly, arguing that when a founder's halo drowns out the company, strategy suffers. For a man whose face is attached to a facial-recognition empire, the instinct to look away from the camera is its own kind of statement.
Honest limitations
This profile draws mainly on CloudWalk's official corporate materials, Chinese financial-press coverage of the 2022 listing, and the company's published descriptions of Zhou's "three waves" framework. We did not have access to an independent, long-form English-language interview with Zhou, so attributions of his views rely on company-sourced wording rather than a verbatim transcript we could cross-check line by line. Financial figures such as market capitalization on listing day vary by source and by the hour of trading, so we describe the debut qualitatively. CloudWalk's more recent large-model commercial traction is reported by the company itself and is not independently audited here. The "Four Little Dragons" label is a press coinage, not a legal or technical category.
What readers can do now
- If you follow Chinese AI, track CloudWalk's Congrong (从容) large model and its industry-agent deployments as a test case for whether a vision-first company can reinvent itself as a platform player.
- Read Zhou's "three waves" idea as a due-diligence checklist: ask any AI vendor which wave they claim to be in, and what proof they have of the previous one.
- When a founder stays unusually quiet, look at the org chart and the product roadmap instead of the keynote — that is often where the real strategy lives.