A provincial health office holds decades of electronic medical records. A transport bureau holds years of vehicle and freight data. A grid operator holds load curves. None of it was collected for training models, and most of it has never been usable by anyone outside the agency that owns it.
Anhui's provincial government has now put a number on how much of that it wants to convert, and a deadline.
The General Office of the Anhui Provincial Government (安徽省人民政府办公厅) issued a three-year action plan on 23 October 2026 covering the development and use of public data in eight priority sectors, with the explicit purpose of feeding what the document calls the "intelligent economy" (智能经济).
Key takeaways
- The plan, Three-Year Action Plan for Developing and Using Public Data in Priority Sectors to Promote the Intelligent Economy (2026–2028), was formally issued on 23 October 2026 — not a draft for comment — and reported publicly in early November.
- It covers eight sectors: advanced manufacturing led by automobiles, finance, logistics, healthcare, science, energy, water resources and emergency response.
- Targets by 2028: about 50 high-value data-use scenarios, about 30 industry datasets, about 200 data products and services, and more than 30 replicable solutions.
- Automobiles go first, with a provincial standard for auto datasets, data property registration rules, a trusted data space and explicit pricing models.
- The plan extends the model to aerospace, biomedicine, new energy storage and robots, and names quantum technology, embodied intelligence (具身智能) and brain-computer interfaces as future users.
Why public data, and why now
The bottleneck for industrial AI in China is no longer model access. It is data that is legally usable, documented and clean enough to train on.
Government agencies sit on a large share of the country's highest-value data — health records, grid load, traffic flows, credit information, water systems — but the data was gathered for administration, not for machine learning, and it is fragmented across systems that were never designed to talk to each other.
Anhui's plan is an attempt to turn that administrative archive into a supply chain. The mechanism is a four-list process: a scenario list, a data list, a governance list and a results-replication list, with the province matching demand from data users against supply from data-holding agencies.
The document also restates a principle that matters more than any target: except where laws, administrative regulations or State Council decisions provide otherwise, departmental public data should in principle be opened, shared back to lower levels of government, or licensed for operation.
Cars first
The advanced manufacturing track is built around the automotive industry, which is Anhui's most important industrial cluster and its most data-hungry.
The plan calls for scenarios in algorithm iteration for intelligent connected vehicles, data-driven R&D and manufacturing, vehicle-road coordination, personalised user operations, supply chain data management and auto insurance. Supporting instruments include a guideline for developing auto data products, a provincial standard for building high-quality auto datasets, provincial rules for registering data property rights, and a trusted data space for the industry.
Two details are more concrete than the usual policy language. First, pricing: the plan explicitly explores "basic service free plus value-added service paid" and "per-use billing plus annual subscription" models, which implies a commercial market for public-derived data rather than one-off open releases. Second, security: it calls for provincial auto data security management rules and a security protection standard for auto data platforms, published alongside the commercial framework rather than after it.
Health, science and the machine scientist
Healthcare is the second track that will draw attention. The plan proposes a provincial residents' health information platform and authorised use of electronic medical records, health archives, examination results, medical insurance information and drug traceability data, with scenarios in insurance claims, tiered diagnosis, chronic disease management, early screening and medical large-model training.
The science track is unusually specific for a provincial document: an "AI-Ready" data flywheel for materials science, plus scenarios for machine-scientist autonomous experiments and intelligent workstation services. That is the same idea as the autonomous laboratory published in Nature this month, written into a provincial plan as a procurement category.
The uncomfortable clause
The plan does not pretend the data is free of risk. It requires a security risk review of public data development and use, and calls for improving compliance governance for large models and agents.
It also permits city governments to explore data-element development funds and allows "centralised procurement with separate contracts" (统采分签) as a way to pay data-holding agencies for governance work — an acknowledgement that cleaning data costs money and that agencies currently have no budget line for it.
What the targets actually mean
Fifty scenarios and thirty datasets are modest numbers for a province of Anhui's size, and that is probably deliberate. The plan is a governance programme, not a spending programme: it names targets, assigns lead agencies, and creates lists and platforms, but discloses no fiscal envelope.
That makes the numbers a schedule rather than a commitment. A "scenario" in this vocabulary is a documented use case with a data supplier and a user, not necessarily a deployed product; "about 50" leaves room for redefinition.
The more meaningful test will be whether the commercial instruments arrive. If auto data can be licensed at a published price, with a registration record and a security standard attached, other sectors will copy it quickly. If the marketplaces fill with datasets nobody can legally use, the plan becomes another shelf document.
Honest limitations
The primary source is the Anhui government's published notice and plan text, dated 23 October 2026, together with the official policy interpretation and coverage by Anhui News (安徽新闻网) and Anhui Business News. I read the Chinese text directly but did not obtain the annexes that assign numeric targets to individual cities or agencies, if such annexes exist. The 50/30/200/30 targets are planning goals for 2028, not appropriations, and no budget is disclosed. "In principle open" is a policy principle with statutory exceptions that the plan does not enumerate. The pricing models are described as explorations, not adopted tariffs. The extension to brain-computer interfaces and embodied intelligence is a stated intention for future work, with no measures attached. Reported implementation will depend on provincial department behaviour that cannot be verified at publication.
What readers can do now
- If you sell AI into Anhui's automotive or health sectors, map your product to a named scenario in the plan — procurement follows the list, not the press release.
- Watch for the auto dataset standard and the data property registration rules; those two documents decide whether licensing is practical.
- For companies outside Anhui, treat this as a template: similar four-list mechanisms are spreading, so the same preparation applies in other provinces.