On 4 September 2026, the General Office of the Ministry of Industry and Information Technology (MIIT) issued the “Artificial Intelligence Support Plan for SME Entrepreneurship (2026—2028)” (《人工智能中小企业创业支持计划（2026—2028年）》).
The day the news broke, most commentaries were passing around the same set of numbers: over 10,000 new science and innovation SMEs to be cultivated in three years, and the specialized-refined-differentiated-and-unique “little giant” companies to break through 2,000.
The numbers, of course, matter. But if you are an individual entrepreneur, or you run a small team of three to five people, the line in this document that is worth reading three times is a different one — a line that nearly every commentary missed.
1. What is different this time: the document writes in “one-person company”
In the official commentary to the plan, the item on “sticking to innovation-driven development” reads like this:
Following the accelerating evolution of artificial intelligence technologies such as large models and agents, encourage the development of new types of business entities such as one-person companies (《一人公司》), super individuals, and intelligence-native enterprises, and support innovation and entrepreneurship among new groups such as core contributors to open-source communities and agent developers.
Elsewhere, when talking about “keeping pace with technological evolution and changes in the form of entrepreneurship”, the commentary points it out even more directly:
Right now, frontier technologies such as large models are advancing rapidly, driving a significant leap in individual productivity. One-person companies (OPC) and super individuals and other new types of business entities rely on intelligent tools to start agile businesses and are growing fast… Targeted support measures urgently need to be improved in line with the realities of industry development.
A commentary by Sina Finance translates this into plainer language: the support plan explicitly calls for localities to give inclusive support to micro-entities such as “one-person companies” and super individuals that run agile startups on top of intelligent tools.
Where does the weight of this line lie?
For the past decade or more, the policy imagination of an “enterprise” in China has basically been fixed: registered capital, premises, staff, operating revenue. “SME” in the policy system means an organization by default. The one-person-company form — one person, using a stack of AI tools, getting done what used to take ten people — has long been absent from the policy context.
This is the first time a national-level document has singled out this type of entity and named it, using verbs like “encourage its development” and “give it inclusive support”.
For individual entrepreneurs, this means you are no longer a “gray zone” in the policy system.
2. Three things behind the four numbers
The goal the document sets: by 2028 —
- more than 10,000 new science and innovation SMEs cultivated
- specialized-refined-differentiated-and-unique “little giant” companies breaking 2,000
- 10 science and technology enterprise incubators and 10 national SME public service demonstration platforms (bases) built to high standards
- new cultivation of 10 national-level characteristic SME industrial clusters
If you only look at these four numbers, it is easy to conclude “this is prepared for large companies and large institutions”. But expand the document’s “4 areas, 15 key tasks” and you will find that the first three items target exactly the real-world difficulties of micro and small entities.
First: computing power cost, solved with the two characters “inclusive”
The document explicitly calls for “deeply implementing the special action of inclusive computing power empowering SME development”, and, relying on the China Computing Power Platform (《中国算力平台》), “advancing pooled integration and elastic allocation of computing power resources, so that entrepreneurial enterprises get low-cost inclusive computing power”.
Translated: in the past, when you trained a small model or ran an industry fine-tune, the most expensive thing was often not the people — it was the GPUs. The direction of the policy is to turn computing power into a public infrastructure that can be bought on demand, with the price pushed down. For a one-person company, that means the experiments that used to be “out of reach” become testable.
Second: you can’t get the data, so build datasets around industries and open them up
The original wording of the commentary: build high-quality industry datasets around fields such as smart manufacturing, materials R&D, and biomedicine, and open them in an orderly way to entrepreneurial enterprises.
This is, in my view, the single most valuable item in the whole document for vertical-industry entrepreneurs. Because over the past two years the true bottleneck of AI landing was never model capability — it was that you did not have data.
Public data everyone has; industry data nobody can get. And the value of vertical AI sits exactly in the industry data. Writing “build high-quality industry datasets and open them in an orderly way” as a task means that in the next three years a batch of industry datasets led by the government or by platforms will appear, and they will explicitly “open to entrepreneurial enterprises”. This is a door opened for vertical-industry entrepreneurs.
Third: no scenarios? Use “Hundred Events, Ten Thousand Enterprises” to bring the chain-leader enterprises in and open up scenarios
The document says that, relying on matching events such as “Hundred Events, Ten Thousand Enterprises” (《百场万企》) linking large, medium and small enterprises, it will guide chain-leader enterprises to open up scenarios — pilot verification, demonstration application, scaled commercial use — to entrepreneurial enterprises.
For people doing enterprise services, this one matters especially. Anyone who builds B-side products knows that the hardest part is not building the product, it is getting the first customer who will let you run through their real business. Writing “guide chain-leader enterprises to open scenarios” as a task is the document creating “the first real scenario” for entrepreneurial enterprises.
3. Open source, named for the first time as “infrastructure”
The document also has one unusual item: it devotes an entire section to open source — “deepen the empowerment of the open-source ecosystem”:
- support the national-level AI open-source community (AtomGit) to grow and get stronger
- encourage entrepreneurial enterprises to take an active part in building the open-source ecosystem; support secondary development and industry adaptation based on open-source models
- encourage localities to explore incentive mechanisms for the commercial conversion of open-source achievements
In the “precision cultivation” stage, the document lists multi-dimensional indicators for identifying high-growth-potential enterprises: leading enterprise talent, invention patents, highly cited papers — and the star count of open-source community projects.
The fact that the star count is written into the enterprise-identification indicators of a national-level policy document is itself a signal: what you are doing in the open-source community is starting to be seen by the policy system.
For developers, this means “I wrote a pile of code nobody understands” has now become something that can be quantified; for entrepreneurs, the “open source” path is no longer just technical idealism but a commercial path with policy catchment — first accumulate stars and contribution records on open-source projects, then build vertical products and industry adaptation on top of open source, and finally move to commercialization. The document writes out all three segments of that road.
4. Stay cool-headed: the document does not say it will hand you money
A lot of good news above; next is the most valuable part of this article — what this document is not.
First, it does not promise direct subsidies. The words “subsidy” and “reward funds” never appear anywhere in the text. What it provides is elements (computing power, data, scenarios), carriers (incubators, demonstration platforms, industrial clusters), services (financing matchmaking, talent, compliance) and channels (the tiered cultivation system). Where is the money? In this sentence: “give play to the guiding role of the National SME Development Fund, the National AI Industry Fund, and the National Integrated Circuit Industry Investment Fund, and drive social capital to increase early-stage investment” — what it leverages is capital, not direct fiscal subsidies.
Second, it has an explicit threshold. One phrase recurs throughout the document: “science and innovation SMEs”. This is not a loose term; it is an enterprise qualification with evaluation criteria. If you want to actually catch the policy bonus, the path is clear: enter the quality SME tiered cultivation system, first earn the “science-and-technology SME” or “innovation-type SME” evaluation, then move toward specialized-refined-differentiated-and-unique.
Third, implementation depends on localities. The document is a “support plan”, not an “implementation plan”. How it actually lands will show in the moves of each provincial and municipal MIIT department. We are already seeing local MIIT departments reposting the official commentary — this is the “local policy implementation window”: the next three months are the window genuinely worth watching.
5. Translate the document into three things you can do
If after reading all of the above you still think “this has little to do with me”, translate it into actions:
Action 1: solve the identity question first. Go to the Quality SME Tiered Cultivation Platform (《优质中小企业梯度培育平台》, run by MIIT) and look at the evaluation criteria and application entry for science-and-technology SMEs / innovation-type SMEs. The document already writes opening a dedicated column into the implementation mechanism, and concentrated information will show up there next. Without the identity, the computing power, data, scenarios and investment that follow — you cannot get in line for them.
Action 2: point your capabilities at the fields the document names. The key fields named are “industry applications, data services, intelligent computing power”; the data-building directions named are “smart manufacturing, materials R&D, biomedicine”. If your product and capabilities can lean toward these three directions, you are at the exact center of the policy narrative. If your story cannot be translated into these three words, then the policy bonus really does have little to do with you — that is the honest truth.
Action 3: turn open-source actions into traceable records. Since the “star count of open-source community projects” is written into the enterprise-identification indicators, every one of your moves in the open-source community — projects, contributions, being cited — shifts from “hobbyist behavior” to “an asset you can accumulate”. From today on, treat every open-source release seriously.
Last words
Looking back at this document, what is most worth remembering is not the three numbers, but one judgment in it:
“Frontier technologies such as large models are advancing rapidly, driving a significant leap in individual productivity.”
Policy documents rarely evaluate productivity directly. The context in which this sentence appears is explaining why “one-person companies” should be encouraged.
In other words, in the judgment of the people who set the policy: one person plus AI can already constitute a productive business entity. That this is acknowledged matters more than any single concrete clause.
How much you personally get to catch depends on one thing — whether you can turn yourself from “someone who can use AI” into “a business entity that delivers results with AI”.
The gap between those two is the opportunity of the next three years.
Do you see yourself as one of the “new types of business entities” the document talks about? Share your take in the comments. If this article was useful, give it a like so I know to keep translating policy documents into actions you can actually execute.