--- title: "China Built a Policy Category for Companies of One. It Has Targets, Not a Denominator" date: 2026-11-03 category: China AI Decoded site: NeuroAI canonical: https://neuroai.site/a/na-decode-opc-policy-category language: en --- # China Built a Policy Category for Companies of One. It Has Targets, Not a Denominator > Seven ministries endorsed the AI one-person company in June. Provinces now compete on targets that count different things, and no register counts the category. The paperwork is the easy part. A small team registers at a shared address, applies for a desk in a government-backed community, and asks its landlord for compute on credit. What it gets is a label the province has just adopted, and a set of targets written for firms its size. The label is the one-person company — 一人公司, or OPC — and since March it has acquired provincial action plans, community standards, a credit product and a five-year horizon. It has not acquired a way of being counted. ## Key takeaways - Seven ministries endorsed "AI one-person companies" on 18 June 2026 — inside a platform-economy plan, not a dedicated AI document (MIIT). - Guangdong's 18-point action plan of 16 March 2026 targets 100 OPC communities and 1,000 benchmark firms by 2028 (Guangdong DRC). - Hunan's plan of 23 October 2026 targets 20,000 OPCs, 200 communities and 10 token factories by 2030, issued by a steering group rather than the provincial government. - OPC is defined by headcount, no more than 10 people, so no business register records it; the closest official count measures legal form instead. - The instruments are access rather than cash: desks, tokens, split orders and credit scored on usage. ## What the centre actually said On 18 June 2026 the Ministry of Industry and Information Technology and six other departments published a three-year action plan for coordination between large and small platform-economy firms. Inside a section on nurturing innovative entities sits one clause: "accelerate the cultivation of AI one-person companies (AI OPC)" — 加快培育人工智能一人公司 (Xinhua, via China Economic Net, 22 June 2026). That clause is the whole of the national position in the document. The category is named, not defined, and carries no target. The placement matters. OPC entered national policy as a sub-clause of a platform-economy plan about large and small firms working together, not as a standalone AI document. The reading is that the centre set a direction and left the instruments to provinces and cities — which is what happened next. ## Two provinces, two different counts Guangdong moved first at provincial level. On 16 March 2026 its development and reform commission, with provincial government approval, issued an 18-point action plan it describes as the first provincial policy dedicated to OPCs: 10 demonstration communities in 2026, then 100 communities, 1,000 benchmark OPC firms and 10,000 entrepreneurs by 2028 (Guangdong DRC, 16 March 2026). Hunan's version is dated 23 October 2026 — 湘制造强省〔2026〕3号, issued by the province's steering group for building a strong manufacturing province and published by its industry department, so it carries the group's document series rather than the provincial government's. Its 2030 targets: 10 token factories (词元工厂), more than 12,000P of intelligent compute, 200-plus communities, 20,000-plus OPCs and 30,000-plus people. Set the two side by side and the league table collapses. Guangdong counts 1,000 "benchmark firms"; Hunan counts 20,000 "agglomerated OPCs". Neither says whether a firm counts twice, or whether a two-person studio inside a community is one OPC or two. A target is a number the issuing body chose, not a measurement. ## The instruments are access, not cash What the provinces do agree on is the toolkit, and it is mostly not money. - **Space and registration.** "One address, multiple licences" and shared-office registration let a firm register where it works; communities supply desks, sometimes rent-free. - **Tokens and compute.** Communities buy compute and tokens in bulk and pass them on, sometimes against the province's compute-voucher (算力券) rules. Hunan also wants "token factories" — centres oriented to producing tokens rather than selling raw capacity. - **Orders.** Hunan's formulation is a chain: a large firm issues the order, a community host splits it, an OPC takes a piece — with "use first, pay later" and "pay by result" permitted as terms. - **Credit scored on usage.** Hunan asks lenders to accept token consumption, compute orders and settlement flows as evidence for a credit decision, and offers an interest subsidy calculated on half the loan amount for qualifying start-ups. The last item is the substantive shift. Collateral is what a ten-person studio lacks; a record of tokens bought and invoices settled is what it has. Whether banks accept it is open — the document invites rather than requires. ## The number with no denominator OPC is not a registration type. A Chinese founder registers a limited liability company or a sole proprietorship; the OPC label is applied afterwards, by headcount and business model. That is why Hunan's definition is a threshold rather than a form: no more than 10 people, working individually or in a very small team, creating value through large models and agents. So the closest official count measures something else. Guangzhou's Tianhe district reported 73,063 technology-sector sole proprietorships and sole-shareholder companies at the end of March 2026, with 4,379 opened in the first quarter, up 12.2% year on year and equal to 12.88% of all new firms in the district (Tianhe district government, 2026). That is a legal-form count, not an AI count. The policy target is a business-model count. Neither converts into the other, which is why the honest reading of "20,000 OPCs by 2030" is an aspiration without a baseline — and why the instrument design, not the target, is the part worth studying. ## Honest limitations Targets in both plans use 力争, "strive to", and neither published text carries a budget line. Neither plan states how many OPCs exist today, so no growth rate can be computed. The Tianhe figures are a legal-form proxy from one district, not a national series. Chinese coverage of OPC policy leans on profiles of individual founders, which I have not used as evidence. Whether lenders accept token consumption for credit is unknown. I have not verified how many cities have issued OPC policies, because no official tally exists. ## What readers can do now - Track instruments, not targets: a community with free desks and a token allocation is evidence; a 2030 number is not. - If you map China's AI start-up support, record the counting unit each province uses — firms, people or communities — before aggregating anything. - Before signing with a community, ask who holds the compute contract, who issues the order, and what the credit decision is actually based on. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-decode-opc-policy-category Free to quote with attribution and a link to the original.