The state has written "one-person company" into a ministry document for the first time. But another set of numbers surfaced the same day: of those sixteen million people, more than half earn under 7,000 yuan a month.
>
Policy hands out resources; the market wants orders. Between those two things lies a gulf — and today I want to lay that gulf out in front of you.
1. The real novelty of the policy hides in three sentences
On 4 September 2026, the Ministry of Industry and Information Technology issued the "Support Plan for AI-Based Entrepreneurship of Small and Medium-Sized Enterprises (2026–2028)". Xinhua, China News and Cailianshe carried it the same day.
The three-year goals are stated plainly: cultivate more than 10,000 new technology-based and innovative SMEs, push specialised and sophisticated "little giant" firms past 2,000, and build ten technology enterprise incubators.
The numbers are large. But the numbers are not the point.
The real novelty lies in three turns of phrase.
The first: "one-person company" and "super individual" were named.
The text encourages local authorities to give inclusive support to micro-entities such as "one-person companies" and super individuals that use intelligent tools to start up nimbly.
Where does the weight of that sentence lie?
In the old policy vocabulary the entrepreneurial subject was called an "SME" — you first had to be an enterprise to reach the policy at all. Terms such as "one-person company" and "super individual" previously circulated only in media headlines and informal discussion, and had never entered a formal ministry document.
From this document onward, the status has changed.
The second: the incubator service list was rewritten.
The policy requires incubation vehicles to break away from the model of "leasing office space first", to bring intelligent compute supply, industry data sharing and application-scenario matching into their core services, and it allows them to take equity in exchange for service fees or rent.
Translated: the startup parks of the past sold desks for a few hundred yuan a month, which was essentially a sub-landlord business.
What must be sold now is compute, data and scenarios — and those three are exactly what is most expensive and hardest for an AI startup to assemble on its own.
The policy is forcing incubators to turn from landlord into shareholder.
The third: compliance was written in as a "service" for the first time.
In the service support section, the text explicitly states that algorithm filing, technology ethics review and security testing and evaluation must keep pace, so as to give SMEs backing.
The subtext is clear: regulation arrives as part of the package, not afterwards.
While encouraging you to enter, the policy has already paved the compliance lane. It shows the top clearly understands that compliance cost is where small teams most easily fall down dead.
In the same batch of signals there are three more: three national teams — the National SME Development Fund, the National AI Industry Investment Fund and the National Integrated Circuit Industry Investment Fund — act as guiding funds; it explicitly names compute-for-equity, data-for-equity and investment-incubation linkage; and open-source projects are written into the cultivation indicators (AtomGit is named).
Local governments have already moved: Shenzhen issued an OPC startup ecosystem plan, Fujian set up a provincial one-person company alliance, and Beijing's Chaoyang district launched a platform where an individual can take commercial payment without registering a company.
Everything points in the same direction.
2. And yet another set of numbers was circulating the same day
While the policy filled feeds, another dataset was fermenting on content platforms in parallel.
As of mid-2025, the number of one-person companies nationwide had passed sixteen million, or 27.4% of all enterprises.
One in every three new companies is run by a single person.
Now the second half:
90% do not survive three years.
30.8% never generated any revenue at all.
Median monthly income is under 7,000 yuan, and 52.7% of OPC founders earn under 7,000 a month.
About 20% build a stable commercial loop; 3.6% earn over a million US dollars a year.
(Sources: a Huzhuo feature from August 2026, Tencent News reporting from the same period, and several industry surveys — three sources cross-checking each other.)
Lay the two sets of numbers on top of each other and you get today's real signal:
The policy is handing out "resources"; the market is asking for "orders".
Between those two words lies a gulf.
And standing in that gulf are sixteen million people.
3. AI lowers the organisational bar, not the commercial bar
This is the judgement I most want to make.
There is a fashionable line going around: one person plus one computer plus AI can match what used to take a whole company.
The first half is true.
AI really has brought down the organisational bar. One person can do what used to take ten — copywriting, design, spreadsheets, organising material, writing code, all within reach. That dividend is real and tangible.
But the second half is wrong.
Because AI has not lowered the commercial bar.
One person cannot sign the contracts that ten used to sign. One person cannot meet a customer's demand for delivery certainty. One person's credibility cannot survive the line in the other side's procurement process: "How many people are in your company?"
So those 30.8% with zero revenue are not lazy, and they are not incapable of using AI.
They have mistaken "capacity" for "business".
In an age of overcapacity, the scarce thing was never "people who can do the work".
What is scarce is the customer relationship, the certainty of delivery, the credential that makes someone willing to hand you money.
The distance between those two things is where this round's real work lies.
4. The opportunity is not in "teaching people to start up", it is in "supplying ammunition"
The policy lists compute, data and scenarios as the core services of incubators.
Read that sentence backwards and it is a list of demand.
One: do not crowd into the lane of "teaching people to run a one-person company".
Those AI startup courses selling for a few thousand yuan sell anxiety, not capability. And the moment the policy wind shifts, that kind of business is the first to collapse.
The genuinely stable business is doing the work for these people.
Nobody builds their knowledge base, nobody handles their customer service, nobody files their compliance, nobody updates their content — and all of that is hard demand, repeatable, and productisable into standard items.
Two: turn "compliance capability" into a product in advance.
Algorithm filing and technology ethics review being written into the service support section amounts to an official admission that this is a genuine barrier for SMEs.
That means a hard demand window opens over the next 12 months. Whoever gets the process working and turns delivery into a standard item first holds the pricing power.
Three: watch the "scenario partner" entry point.
The policy encourages local authorities to develop "AI entrepreneurship scenario partners" and requires incubation vehicles to match scenarios.
The implied message is: parks and incubators will become the first procurement gate for AI services aimed at SMEs.
One partnership conversation with a carrier is an order of magnitude more efficient than visiting customers one by one.
5. What I do myself
I spent years in operations management — building systems, writing SOPs, training newcomers, running internal audit reviews — and the thing that gave me the most headaches was never "the work does not get done".
It was the work gets done and the customer does not accept it.
The last two years have made that more concrete. Building knowledge bases and AI customer service for small and micro businesses, the most time-consuming part was never the technical integration.
It was turning the scattered experience in a customer's head into a delivery standard that a machine can read and the customer can maintain themselves.
It sounds unglamorous, but it is the line between that 20% and the 80%.
With a delivery standard, your service can be replicated, quoted and accepted.
Without one, you are stuck forever at the stage of "doing a friend a favour".
So I keep doing three things: organising scattered experience into a knowledge base, turning repeated enquiries into AI customer service, and distilling every delivery into a reusable process.
None of those three solves "is there an order".
But they decide whether I can catch the order when it comes.
In closing
Policy gives direction, not orders.
"One-person company" entering a ministry document is a confirmation of status, not a proof of income.
Sixteen million people have already entered, and half of them earn under 7,000 yuan a month.
That number should not excite you, and it should not drive you out.
It tells you one thing: in this era, people who can use the tools are not scarce; people who can turn the tools into something others will pay for are.
#one-person company #AI entrepreneurship #SME #knowledge base #AI customer service
Do you know anyone running a "one-person company"? Where do they get stuck — finding customers, or holding delivery together?
>
Share your observations in the comments. If this was useful, hit like or share so more of the people who need it see it.