County-Level AI Procurement Reaches 100m Yuan: How Smaller Suppliers Can Respond

County-level AI procurement is now reaching the hundred-million-yuan scale. For smaller AI suppliers, the decisive advantage is no longer technical capability but the ability to see tender notices early and assemble the paperwork inside a narrow window.

The shift downward is the clearest signal

On September 24, the Development, Industry, Information Technology and Technology Commerce Bureau of Pu County in Linfen, Shanxi, published a government procurement intention for September to October 2026. It included an artificial intelligence project at the hundred-million-yuan scale.

A county.

Anyone doing B2B business understands what that means: the level of procurement is moving downward, and for the first time AI companies at the smaller end have a direct relationship with a county-level budget.

That window was pushed open a crack at the end of September. Who can squeeze through depends on two things — whether you can see the notice, and whether you can pull the materials together inside the window.

Three scattered examples make the pattern visible.

The Pu County intention was published on September 24, 2026, covering procurement arrangements for September and October. In Chengjiang, Yunnan, a tender notice for full-process cost consulting services for the "AI Empowering Chengjiang Cultural Tourism High-Quality Development Project" was recently posted. The tender documents can be obtained from September 29 to October 12, 2026, excluding statutory public holidays and festivals, and only during two daily windows: 9:00 to 11:30 in the morning and 2:00 to 5:00 in the afternoon. In an industry tender information roundup, the AI medical large model foundation and application project of the Xiamen Health and Medical Big Data Center has a budget of 10.8 million yuan.

One is a county in Shanxi, one is a county-level city in Yunnan, one is a municipal unit in Fujian. They do not constitute a statistic, but the direction is clear: AI-related fiscal spending is moving down from provincial capitals to prefecture-level cities and to counties.

Policy language is pushing in the same direction. The "Opinions on Deeply Implementing the 'AI Plus' Action," issued in August 2025, proposed that by 2030 the penetration rate of new-generation intelligent terminals, agents and similar applications should exceed 90 percent. The target is on paper; the money has to come out of budgets at every level.

Local fiscal action is not limited to procurement. Hubei adjusted its subsidy categories for the 2026 consumer goods trade-in program. The new rules take effect on October 1, 2026, and bring embodied robots, cooking robots, exoskeleton robots, and companion robots and robot dogs into the subsidy scope. The program runs until December 31, 2026.

Procurement and consumption are spending at the same time. That is the window.

How big is the pool, and is there enough for smaller firms

On September 28, the Ministry of Finance released a brief on national government procurement in 2025: the full-year procurement scale was 3,321.296 billion yuan, comprising 795.392 billion yuan in goods, 1,304.508 billion yuan in works, and 1,221.396 billion yuan in services, with services accounting for 36.77 percent.

But another figure deserves more attention.

Decentralized procurement reached 2,594.738 billion yuan, accounting for 78.12 percent of the national government procurement scale. Centralized government procurement was 624.443 billion yuan, only 18.8 percent.

What does 78.12 percent mean? It means the vast majority of projects do not go through the centralized procurement channel. They are scattered across the notices published by budget units at every level. For smaller suppliers this is good news — the channel has not closed the door. The bad news is that these notices are scattered across different levels and different platforms, and no one is compiling them for you. With centralized procurement you at least know where to look. With decentralized procurement you are not even sure where to look.

Information cost has become the most practical competitive factor in this round.

The barrier is not technology, but seeing and catching up

The common situation looks like this: a company doing AI applications has no technical problem and a presentable proposal, but it bids on only a few projects a year and always learns about them the day before the bid opens.

There are three bottlenecks.

The first is keywords. What you need to search is not "AI" but the terms actually used in tender documents — "artificial intelligence," "large model," "agent," "computing power," "data foundation" — and they must be cross-combined with your industry and your region. The same term points to completely different projects in Beijing and in Linfen.

The second is the window. The document acquisition window for the Chengjiang project runs from September 29 to October 12, 2026, with weekends and holidays excluded and only two time slots per day. The National Day holiday sits in the middle. Miss it and it is gone, with no connection to your technology at all.

The third is qualification conditions. The same notice states that consortia are not accepted and that bidders must have no record of dishonesty, as determined by queries through "Credit China" or the National Enterprise Credit Information Publicity System, with the query period running from the date the notice is issued to the deadline for submitting bid documents. These conditions are not complicated, but if one item does not match, everything afterward is wasted.

So the starting move should be to turn "finding tenders" into a standing mechanism, rather than something you dig into only when a project appears.

The Pu County "September to October procurement intention" is a signal sent before the formal tender — knowing at the intention stage and knowing at the tender stage are separated by an entire preparation cycle.

How to put this into practice

If you are serious about moving, the sequence is roughly three steps.

First, build a keyword library. Use core terms such as "artificial intelligence," "large model," "computing power" and "agent," overlay the nouns of your industry, and subscribe to both categories in combination. On geography, do not circle only your own city. Include surrounding prefecture-level cities and counties as well — the opportunity in this round lies precisely in lower-tier markets.

Second, turn subscription into a fixed daily action. SHEYU LEIDA (舍予招标雷达) aggregates government procurement and bidding information nationwide, with real-time subscription push by keyword, region and industry, so tender notices are not missed. Beyond its own iOS app, it also has a white-label mini-program channel version that can be filtered on a phone. Its value is not bidding on your behalf; it is putting intention notices in your hands before they become formal tender notices.

That leaves the third step: check qualifications first, then assess technology. When you see a target project, the first thing to do is run through the qualification conditions, the document acquisition window, and whether consortia are accepted. Once you clear that gate, then talk about the proposal.

Final note

Between a policy dividend and a company's books sits a public notice.

Notices have a shelf life. For projects like the one in Chengjiang, the acquisition window runs only about two weeks, counting the National Day holiday.

Seeing it, and catching it — that is the most concrete barrier right now.