---
title: "The Voucher Went to the Incubator First"
date: 2026-11-17
category: AI in Action
site: NeuroAI
canonical: https://neuroai.site/a/na-app-opc-communities-1117
language: en
---

# The Voucher Went to the Incubator First

> Row one: Zhongguancun AI Beiwei Community, run by Beijing Kexing Chuangfu Information Technology Co., Haidian district. Sixteen more rows follow, same three columns: community, operator, district.

Row one: Zhongguancun AI Beiwei Community, run by Beijing Kexing Chuangfu Information Technology Co., Haidian district. Sixteen more rows follow, same three columns: community, operator, district.

The Beijing Municipal Bureau of Economy and Information Technology published the list on November 16 as the 2026 roster of OPC growth communities. 17 in total, 6 in Haidian.

OPC means one-person company. Not a legal form here, but a local incubation category for very small, AI-native firms. The label is new. The buildings are not.

## New names, old operators

Four of the 17 operators carry "incubator" in their names. The rest are park operators, technology service companies, asset managers and cultural businesses. None is called a one-person company service center.

Eleven of the 17 community names contain OPC. The name is new; the operator is not. Money follows the sign: a municipal OPC growth community can claim quarterly, up to 10 million yuan a year, plus 2 million for outstanding service results. The first layer of this table is a re-certification of space that already exists.

## Nine districts, seven without

Haidian 6, Chaoyang 2, Tongzhou 2, Shijingshan 2, and one each in E-Town, Fengtai, Huairou, Yanqing and Changping: 17 across 9 districts. Beijing has 16, so seven have none.

Not for lack of one-person companies, but because no institution can supply space, an operating entity and finance for quarterly claims at once. Daxing proves it: in August, Xihongmen town published draft measures with three hard criteria, more detailed than the municipal text. Daxing has zero.

Text can be written overnight. A community needs a building, an operator, and someone who can file a quarterly claim.

## The city never defines it

The Action Plan for Supporting AI OPC Innovation (Trial), issued June 17, 2026, has eight articles. None defines OPC.

The opening line says only: help more "super economic individuals (OPC)" grow fast.

So at city level OPC equals "super economic individual," a description, not a test for certifying a company.

The criteria appear at district level. Xihongmen's draft: an OPC enterprise is registered in the town, centered on a single founder, uses AI as its core production tool, and runs small and light.

Three tests follow. Registered locally, under 3 years old, clean record. No more than 10 staff in principle; the founder holds over 50% after look-through and leads technology, decisions and execution. AI as the core tool, with tokens, compute and AI tools averaging at least 800 yuan a month over the past 12 months.

The 800 is the important one.

At Tencent Hunyuan Hy4 flash's published 0.5 yuan per million input tokens, 800 yuan buys 1.6 billion tokens.

The threshold screens use, not capability. 800 a month is 26 a day—too small to be a cost, enough to separate "registered a company" from "running a business." Setting it there admits the fixed cost of starting an AI company is now counted in yuan.

OPC itself is not new: China's one-person limited liability company entered the Company Law in 2006. What changed is the statistical category. The city says "super economic individual," which need not be a company; the district says "AI one-person company," which must be one.

The scope was left open. The money was settled first.

## The timeline runs backwards

December 8, 2025: AI Beiwei launches an OPC service plan, the first in Beijing. June 17, 2026: the municipal plan. November 16, 2026: the list.

The community ran 6 months ahead of the policy.

By publication, AI Beiwei alone had more than 70 OPCs. Haidian's April measures went further: 50,000 yuan a year for two years in startup capital, and model vouchers up to 2 million a year. Communities ran first; policy ratified.

## 170 million, and 1,700 companies

10 million yuan over 100,000 is 100: one community can let at most 100 companies take the full 100,000 yuan resource package.

100 times 17 is 1,700.

170 million yuan is the annual pool and the ceiling on OPCs reached. AI Beiwei alone is already above 70, close to 100; all 17 at that density is 1,190. The design capacity is only slightly wider than what exists.

So the instrument is a sieve, not a runway. It gives 3 months of models, compute and data procurement, capped at 100,000 yuan per company, then it is gone. Three months is the length of a validation, not a business.

## The smallest flywheel is one person

At Lenovo Capital's CEO meeting in Hangzhou, Yang Yuanqing described a "private-domain data flywheel": devices and enterprise systems generate data, data trains agents, better agents earn more use, and use generates more data. "Public data is available to everyone, and models will converge," he said. "What really separates companies is private data that others cannot get and only you own."

That is a business prospectus for a one-person company: one computer, one agent, one dataset nobody else has. The smallest flywheel is not a company. It is a person.

But the first turn is the most expensive, and the 100,000 yuan buys it. After that, Yang says, the leader pulls further ahead, which means those who cannot get it turning are still stuck after 3 months. A voucher buys time, not demand.

## Policy artifact, or cost artifact

Costs fell first, people gathered, communities ran, and policy arrived to do paperwork, hand out vouchers and hang signs.

Look only at the November 16 list and policy seems to have created OPC. Pull the timeline back 11 months and the 70-plus companies at AI Beiwei moved in with no municipal document in existence. The policy's real job is cutting three costs: finding space, trying AI, and being seen.

All three are transaction costs, not production costs.

So it cannot change who succeeds. It can only change who starts testing earlier.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-opc-communities-1117
Free to quote with attribution and a link to the original.
