Twenty Chinese provinces and municipalities have issued over 100 measures for one-person AI companies

The 426 is a count of communities, and the coverage of the Fudan survey does not say how many founders work inside them or who owns the communities. ^[1] Across 65 cities the average is about 6.6 communities a city ^[15], and with roughly four-fifths of the total in the east ^[2] about 340 of the 426 sit there ^[14]. Jiangsu has the most, followed by Shandong, Guangdong, Zhejiang, Beijing, Hubei and Shanghai ^[3], and the report says new sites keep clustering where industrial bases, talent pools and technology ecosystems are strongest ^[9].

Whether that makes the communities "state-backed" depends on the definition. The sources describe each community as a hybrid of workspace, operators, enabling partners and founders ^[11]. The state appears as the payer for workspace subsidies, computing vouchers, training, scenario matching and risk mitigation ^[4]. Twenty provinces or municipalities issued more than 100 such measures by May 2026, so more than five each on average ^[16], and Suzhou moved first, with action plans in late 2025 ^[5]. Set that against the 28 regions with communities ^[1]: if the twenty sit inside the 28, eight regions host communities with no dedicated measure of their own ^[18].

The money goes to the cost side, where computing dominates. Pudong offers newly registered one-person companies up to 300,000 yuan in free computing power, according to news.bitcoin.com ^[6], while the Fudan report says computing can take more than four-fifths of a budget ^[7]. At an 80% compute share, a founder whose compute bill equals the full grant runs a 375,000-yuan operation ^[17]. Below that size the voucher covers all compute; above it the founder pays the difference.

The report recommends that governments move beyond space and cash toward order matching and ecosystem orchestration ^[8], and it warns that policies stressing rent-free desks and direct subsidies risk leaving physical space underused if operating quality lags ^[12]. According to Crowdfund Insider's account, the study treats the model as an emerging experiment that has yet to become a durable feature of China's employment and innovation system ^[10].

We think the editor's framing is half right. The sources document institutional support, with twenty places and 100-plus measures behind it, but the 426 counts sites and says nothing about whether the companies in them survive. The counter-thesis is that subsidised compute lets founders ship products and build track records that later attract paying customers, and that would show up as communities keeping their founders after the vouchers run out. We would be wrong if occupancy and retention hold once the subsidy ends. If the desks empty, the 426 is a count of subsidy sites.