Two lawyers argue about the same city. One cites a document the city legislature passed last month. The other cites an ordinance with a fixed start date. Both call it "Guangzhou's AI law" — and only one can tell a client what is enforceable today.
That gap is not a drafting quirk. It is the normal shape of Chinese local legislation, and in the current wave of city-level AI ordinances it is the only fact that matters.
Key takeaways
- On 22–23 October 2026 the Guangzhou Municipal People's Congress Standing Committee voted through the 《广州市人工智能发展促进条例》 (Guangzhou AI Development Promotion Regulation) at its 54th session. A city ordinance still needs Guangdong provincial standing committee approval before it can be promulgated and take effect.
- The published consultation draft runs to 21 articles, and its final article on the effective date is left blank — it reads 本条例自 年 月 日起施行 with no day filled in. The gap is real, not shorthand.
- Suzhou has completed the whole climb and is the only one with a dated start: passed on 1 September, approved by the Jiangsu provincial standing committee on 30 September, promulgated on 16 October, in force from 1 December 2026.
- Nanjing passed its own ordinance on 28 October 2026 — 7 chapters, 62 articles — and is also waiting on provincial approval. Jiangsu's provincial draft had a first reading in July, Xiamen's a second in October, and Hangzhou's is still out for comment.
- Shenzhen's 2022 ordinance and Shanghai's were the first AI-specific local laws of their kinds. Guangzhou is the latest in a queue now several cities deep.
- The stakes are not abstract. Hangzhou's AI core industry revenue reached RMB 460 billion (≈ US$68.5bn / HK$537bn) in 2025.
The four stages a Chinese city AI law has to clear
Each stage supports a completely different claim:
- 草案征求意见稿 — a consultation draft, published for comment. No legal effect.
- 表决通过 — passed by the city standing committee. Still no legal effect.
- 批准公布 — approved by the provincial standing committee and promulgated. Now it is a law, but may not be in force yet.
- 施行 — in force.
Guangzhou is at stage two. Suzhou has been at stage four since 16 October 2026. A supplier quoting "Guangzhou's AI ordinance" in a contract today is citing stage two.
What the Guangzhou draft actually obliges
The 21-article draft, published on 17 August 2026 with comments closing 18 September, is more operational than its name suggests:
- 算法模型辅导中心 — a city-run counselling centre for algorithm filing (备案), security testing and technical adaptation. Article 5.
- 全市算力协同调度平台 — a city-wide compute coordination and scheduling platform with a market mechanism for trading compute, alongside supercomputing centres, urban-edge inference nodes and industry clusters. Article 4.- Two annual lists — a public data resource catalogue and a high-quality dataset construction list, naming geographic spatial, meteorological, market supervision, transport, healthcare and education data as priorities. Article 6.
- Procurement outside open tender — government bodies, public institutions and state-owned enterprises must go first for products on an annual demonstration application list, and purchases above the open-tender threshold may use other methods. Article 10.
- A prohibited-behaviour list — five categories, including algorithmic price discrimination or consumer fraud, and failing to implement content review, labelling (标识) and algorithm review duties in a way that produces false information harming the public interest. Article 19.
- A regulatory grace period — for new technologies with no established market access rule, departments must explore convenient approval routes rather than banning them, and may grant a reasonable 执法观察期 (enforcement observation period). Article 18.
Article 16 cuts across provincial borders: talent already recognised in AI-leading cities can skip local recognition procedures.
The queue behind it
Suzhou is the informative case, because it shows the whole timetable. Its 《苏州市人工智能产业发展和应用促进条例》 was adopted on 1 September 2026, approved by the Jiangsu provincial standing committee on 30 September, promulgated on 16 October, and enters into force on 1 December 2026. Seven chapters, including a safety governance chapter codifying graded supervision alongside sandbox supervision (沙盒监管) and trigger-based supervision (触发式监管).
Nanjing followed on 28 October with a 7-chapter, 62-article ordinance awaiting the same provincial step. Hangzhou's draft is still collecting comments, Xiamen's special economic zone ordinance had a second reading in October, and the Jiangsu provincial draft had a first reading in July.
Before this wave there were two landmarks: Shenzhen's 《深圳经济特区人工智能产业促进条例》, the first AI-specific industrial legislation of its kind, in force since 1 November 2022; and Shanghai's 《上海市促进人工智能产业发展条例》, the first provincial-level AI statute, in force since 1 October 2022.
Honest limitations
I am reading Guangzhou's draft text and the promulgation notices of ordinances that have cleared the process, not any enforcement record, because the Guangzhou ordinance is not in force and I found no published enforcement under any of them. The session notice records the vote but states no provincial approval date, and I could not find a Guangdong approval or promulgation notice as of 3 November 2026. The revenue figure comes from state media reporting of a city statement rather than audited statistics. This piece covers local ordinances only; a national AI statute is on the legislative agenda and would change the picture.
What readers can do now
- Check the date, not the title. For any Chinese city AI ordinance you are cited, find out whether it is a draft, a passed bill, a promulgated ordinance, or in force. The Guangzhou one is a passed bill. Suzhou's is in force.
- Build a four-column register for the ordinances touching your operating footprint — one row per city, columns for stage, promulgation date and effective date. Suzhou's 1 December 2026 start date changes release plans.
- Write the Guangzhou-style obligations into your own controls now. The filing counselling centre, the annual data catalogue and dataset list, the demonstration procurement route and the five prohibited behaviours cost nothing to design for, whatever the approval date turns out to be.
Why the timetable is the point
The instinct is to read a new AI ordinance as a statement of direction. In China it is better read as a scheduling problem with a legal payload attached. The payload is real — an enforcement observation period, a prohibited-behaviour list, a compute scheduling platform — but the date it binds you is set two institutions away from the city that wrote it.
Guangzhou counts more than 2,200 AI-related enterprises and ranks third nationally in the number of large models holding a cyberspace regulator filing (备案), per state legal media citing city figures. Cities where the product is built and cities where the ordinance applies are increasingly the same, and the statute is what turns local industrial ambition into obligations.
According to Li An, chief scientist at BrainNet (脑机网), the useful discipline for reading any Chinese local AI law is to ask which of the four stages it occupies, because a passed bill and an in-force ordinance impose different obligations on the same company.
The paperwork is not the interesting part of China's local AI law. The date is.