Wu Peiwen: ¥1.5M a Year, One Laptop, and an AI Team That Does 90% of the Work

Introduction

On April 1, 2026, a phrase climbed the trending charts of Chinese social media: "a man who opened a one-person company with AI earns ¥1.5 million a year." The man was Wu Peiwen, an entrepreneur in Hangzhou. His entire company is himself, one laptop, and a squad of AI agents that scrape data, analyze markets, write copy, and run ad placements. His reported monthly cost is about ¥3,000 (under US$450) — mostly AI tool subscriptions and server fees — with a reported net margin above 65%.

Take the headline numbers as media reports, not audited accounts. They come from the original reporting that made him famous, and nobody has independently verified them. But the mechanism underneath is the interesting part, and it checks out: one person, AI as the execution team, the founder as the only decision-maker. In Hangzhou, that model now has a name, a policy framework, and an entire neighborhood behind it — the OPC, or One Person Company.

Wu is not an isolated case. A report from the Zhongguancun Talent Association put the number of single-shareholder companies in China at over 16 million, with 2.86 million new ones registered in the first half of 2025 alone — 23.8% of all new business registrations, up 47% year on year. Most of those are conventional small businesses, not AI operations; but the statistic shows the direction of travel. Hangzhou sits at the center of the most organized version of this wave: district-level policies, dedicated founder communities, and a local nickname for the new breed of founder.

The Story

Wu is not a newcomer who stumbled into AI. He has years of experience building cross-border marketing platforms and running startups. When he landed in Honghu Hui — an OPC community near Hangzhou's East Railway Station — he brought something most AI hype merchants lack: a domain where AI actually fits. His company provides full-process AI marketing solutions to small and mid-size e-commerce sellers overseas: market analysis, ad copywriting, visual design, video editing, and even customer-service replies. Every step is executed by AI agents he commands; he handles strategy, breaks down the brief, and signs off on the result.

The numbers reported around his operation are deliberately extreme. Roughly 90% of the execution work is done by AI; the human does the deciding. Monthly operating cost hovers near ¥3,000, which means almost the entire ¥1.5 million in yearly revenue falls through to the bottom line. That cost structure is the real headline — not the revenue, which a dozen marketing agencies could match with forty people, but the margin, which they couldn't.

Wu's own explanation for why this works is disarmingly practical. He argues that every person's knowledge and skill reserves are uneven — even a dozen-person startup team hits the same problem. The difference now is that AI tools can fill the skill gaps and level the execution standard, so a single person can keep the whole process and its delivery quality consistent. He is not replacing workers with AI. He is replacing the need for a team to be well-rounded with a system that makes one person well-rounded.

The timing was not accidental either. On March 3, 2026, Hangzhou's Shangcheng district released the first district-level OPC policy in the province and declared it would become the "number one city for OPC entrepreneurship." Neighboring districts followed with their own packages — free desks, computing-power subsidies, streamlined registration. Honghu Hui itself hands founders over 1,100 square meters of rent-free space and a one-time startup grant worth US$50,000, taking no equity and locking no valuation, with the local government adding roughly ¥1 million in special funds every quarter. Wu's one-person company runs inside a policy experiment designed to make one-person companies cheap to start and hard to kill.

Inside these communities, the culture is already developing its own language. In Hangzhou's OPC scene, solo founders are playfully called lobsters — 「龍蝦」 — and one district's policy package literally speaks of "raising lobsters." The joke captures the thesis: small creatures, hard shells, thriving in new water. Honghu Hui, the community Wu belongs to, reportedly received over 1,300 project applications and admitted just 32 — a reminder that the barrier to starting may be low, but the bar for the work itself hasn't moved.

That same day his story trended, one of China's best-known tech founders weighed in. Zhou Hongyi, founder of 360, argued that the root reason most people fail with AI is excessive control: people treat AI as a tool or a subordinate, dissecting every step of the process, which throttles what it can do. His proposed method is blunt — write only two sentences. Sentence one: you are a top-ten expert in this field. Sentence two: here is my goal. Then let it generate multiple times. Delegation with boundaries, he added: money and core data stay under human lock, with real permission isolation. It is, in essence, Wu's operating model restated as philosophy: define outcomes, don't choreograph the process.

The story has its skeptics, and the reporting names them. Experts warn that the model is not copy-paste friendly — it still demands industry experience and technical ability: AI workflow design, data compliance, model tuning. The market has already spawned high-priced "AI solo entrepreneurship" courses, some of which edge into loan-inducement scams. And founders on the ground report the unglamorous side: the body is free, but the mind is more exhausted; work stretches past midnight, and all the decision pressure and loneliness land on one person. The super-individual looks like freedom from the outside and feels like weight from the inside.

Lessons

Three mechanics worth stealing from Wu's solo run: 1. Define outcomes, don't choreograph the process: The Zhou Hongyi rule that mirrors Wu's own setup. For every workflow, draw two columns — delegate versus never-delegate. Money movement and core data stay human-gated; everything else gets a two-sentence brief (expert identity plus goal) and multiple generation rounds. The founder's job is review, not authorship. Copy it this week: pick one repeating task you currently do by hand, write the two-sentence brief, and run it three times before touching it yourself. 2. Use AI as a skill leveler, not a headcount reducer: Wu's insight is that a team fails because skills are unevenly distributed, not because there are too few people. The playbook: list every capability a ten-person team would need to deliver your offer — research, copy, design, video, support, ops — then assign each one to an agent or a human checkpoint. What you are buying is not labor substitution; it is a delivery standard that doesn't wobble. If a step's output quality can't be judged by you in under five minutes, it isn't agent-ready yet. 3. Sell the finished process, not the tool: Wu doesn't sell "AI services" — he sells marketing outcomes to overseas e-commerce sellers: analysis through ad placement to replies, one accountable pipeline. Clients buy the result, never the machinery. Package your agents into a single priced outcome with a single point of responsibility, and you escape the hourly-rate race that AI is commoditizing. The moat is the pipeline design and the sign-off judgment, not the model behind it.

Closing

¥1.5 million a year, one person, ¥3,000 a month in costs — not because the AI is magic, but because the human stopped doing what machines do and kept only what humans must: judge the goal, sign the result, own the consequence. Wu Peiwen's real company isn't a marketing agency. It's a demonstration that in the AI era, the smallest viable team is now a team of one — plus the discipline to let the other workers actually work.