90% of One-Person AI Companies Fail for the Same Reason

Twenty years ago, the internet gave rise to platform-based companies, and Alibaba and Tencent became giants.

Hello everyone, I'm Xiaohu.

Twenty years later, AI is spawning another kind of company, where one person is an entire team.

And Hangzhou, my hometown, just held the nation's first OPC conference last month, writing this trend into government documents.

On June 29th, Hangzhou held the first 'AI + OPC Innovation Development Conference.' The mayor attended in person and announced 15 action plans and 12 supporting policies.

OPC, One Person Company.

If you think this is just another buzzword from Silicon Valley, a few data points might make you sit up.

The number of one-person companies in China has already exceeded 16 million, accounting for 27.4% of the total number of enterprises nationwide. New registrations in the first half of 2025 grew by 47% year-on-year.

Data from the Carta platform shows that the global proportion of solo founders rose from 23.7% in 2019 to 36.3% in 2025, a 53% increase in 6 years.

16 million. This is not a lab concept; these are real registered, operating companies.

What about Hangzhou itself? In 2025, the revenue of its core AI industry reached 460 billion yuan, accounting for 64.5% of the entire Zhejiang province.

The core digital economy industry accounts for over 29.5% of its GDP.

That's not all. Honghu Hui released a report with a staggering figure.

For every 1 yuan invested in AI cost, it equivalently replaces 72 yuan in human labor expenditure.

1 to 72.

So what? It means a business that previously required a 5-person team with an annual operating cost of 500,000 yuan can now be run by one person plus AI, with annual fixed costs squeezed to under 20,000 yuan.

This is no longer just an efficiency improvement; this is a complete restructuring of the cost base.

But 90% Will Die

If you're already tempted, hold on. Another set of data throws cold water on that.

At the Hangzhou conference, OPC expert Zhou Baoqiang gave a harsh judgment. Over 60% of OPC entrepreneurs nationwide have cognitive biases, and the overall survival rate for OPCs is less than 10%. 90% will fail.

He summarized the most common mistake OPC entrepreneurs make: mistaking tools for a business model and falling into the trap of the solo all-rounder.

What does that mean? Knowing how to use AI to write code doesn't mean you know how to do business.

AI allows everyone to quickly build products, but when everyone uses the same tools, the result is extremely rapid product homogenization.

Technical capability itself no longer constitutes a moat.

An in-depth analysis by 36kr put it even more bluntly.

79% of OPC entrepreneurs spend their time learning new tools, and 78% spend it on product development, but more than half are puzzled by the same question: how to acquire customers.

Knowing how to build but not how to sell. This is the true picture for most one-person companies.

83% of OPCs list customer acquisition difficulty as their biggest pain point. Products are made quickly and well, but no one knows about them, and no one pays.

What the Surviving 10% Did Right

After discussing the harsh reality, let's talk about something uplifting.

Mr. Chen, a merchant from Yiwu, makes artificial flowers. He started with a loan of 10,000 yuan and almost zero English.

He used AI agents for real-time translation plus 24/7 automated customer service and sold his flowers to 50 to 60 countries.

He doesn't understand AI programming, Cursor, or Claude Code.

But he understands the artificial flower market, knows what styles customers in different countries like, and knows how to price things to make them place an order.

AI just helped him eliminate the language and time barriers.

Another, more extreme example.

American Matthew Gallagher used $20,000 plus a dozen AI tools to build Medvi, a GLP-1 telemedicine platform.

In 2025, its revenue was $401 million, with a net profit margin of 16.2%. Full-time employees: 2 people.

Started with $20,000, $400 million in revenue, 2 employees.

For comparison, the publicly traded company in the same industry, Hims & Hers, has an annual revenue of $2.4 billion, 2,442 employees, and a net profit margin of 5.5%.

Medvi's revenue per employee is 200 times higher.

These two cases share a common point. Their core competitiveness is not the AI tools themselves, but deep industry knowledge.

Mr. Chen understands the artificial flower export trade, and Gallagher understands the compliance and operational logic of telemedicine. Tools are available to everyone; cognition cannot be copied.

You've probably also heard about Hu Yanbin.

A singer, using Cursor and Claude Code, built a fan community app called 'Yanhuo' in 45 days. It got 27,000 downloads and shot straight onto the App Store's social chart.

The programmer community collectively lost it. But if you think carefully, his greatest skill isn't knowing how to use Cursor; it's knowing exactly what functions fans need.

Tour maps, behind-the-scenes vlogs, limited private messages—these product definitions weren't dreamed up by AI; they are the industry intuition accumulated over his 20 years as a singer.

The Policy Dividend Window

Let's go back to that Hangzhou conference.

The level of real financial support this time is indeed significant.

Compute vouchers, with an annual cap of 10 million yuan per enterprise. Workstation registration plus one address for multiple licenses, eliminating the rigid expense of traditional office rental for startups.

AI product experience vouchers help entrepreneurs lower early customer acquisition thresholds. Shangcheng District even implemented a 1-to-3-year regulatory sandbox period, replacing severe penalties with risk warnings.

Not just Hangzhou. As of May 2026, 426 OPC-related entrepreneurial communities have been built nationwide, covering 65 cities and 26 provincial-level regions.

Shenzhen's Longgang District has planned 54 OPC communities, providing 170,000 square meters of rent-free office space and issuing 2.46 billion free tokens.

Ningbo's Haishu District is even more aggressive: a 5-square-meter workstation can be rent-free for up to 3 years, with an annual maximum of 300,000 yuan in computing power subsidies and a 3 million yuan credit loan at an interest rate of 2.31%.

Frankly, as a Hangzhou native, I feel a bit emotional seeing these policies.

Twenty years ago, Hangzhou took off on the platform economy, with Alibaba and NetEase pushing this city to the heights of the internet.

Now, Hangzhou is again the first to bet on AI plus one-person companies, shifting from platform thinking to individual empowerment. This city's sense of smell has always been quite sharp.

But policy is the soil, not the fruit. Having soil doesn't guarantee you can grow something.

Tools Will Change, Cognition Won't

Recently, while researching these cases, one realization has become increasingly clear.

The 36kr article quoted a line I particularly agree with.

Tools will iterate, models will change, but the accumulation curve of deep cognition is far steeper than the slope of the technology curve.

In plain English, whether you use Cursor today, Lovable tomorrow, or Claude Code the day after, switching these tools around won't affect your business.

But if you don't understand what your customers want, giving you more tools is useless.

AI has lowered the barrier to entry. From hundreds of thousands to a few thousand yuan, from assembling a team to just one person. But AI has simultaneously raised the bar for winning to a new height.

Because everyone can quickly build a product, the real competition becomes who understands the industry better, who is better at acquiring customers, and whose cognition is deeper.

This is actually good news.

For our generation, AI provides unprecedented leverage.

One person plus AI equals a company. This is no longer a slogan; it's a reality backed by 16 million registered enterprises, a $400 million revenue case study, and genuine government financial support.

But the premise is that you must figure out two things.

What industry do you truly understand, deep enough that your deep cognition cannot be copied by others?

And, after your product is built, how do you let the people who need it find you?

The answers to these two questions are not in any AI tool. They are in your own head.