SHEYU.AI舍予基业
Opportunity Radar

AI procurement is moving downstream, and smaller vendors can catch it

AI government contracts are shifting from provincial mega-tenders to universities, hospitals and county offices, opening a narrow, fast-closing window for smaller suppliers.

·4 min read
AI procurement is moving downstream, and smaller vendors can catch it
AI-generated illustration, not a news photograph

AI-related government procurement notices have been piling up across China over the past three weeks. They rarely appear on provincial mega-platforms; instead they are scattered across universities, vocational colleges, hospitals and county-level departments. Individual contract values range from roughly RMB 160,000 to over RMB 10m. For small and mid-sized software and equipment vendors, this is a window that is opening — and closing quickly.

Smaller lots create room for smaller players

Consider a few specific notices.

On 28 September, the School of Computer Science and Technology (School of Software) at Guangxi University of Science and Technology issued an inquiry-based procurement notice for embedded AI development boards for teaching instruments: 43 sets, each with 8GB LPDDR4x memory, 64GB eMMC storage, a 6TOPS NPU, a 1080P screen, a camera and a smart car module. The procurement ceiling was RMB 162,733.50.

On 22 September, the Xiamen Health and Medical Big Data Centre issued a tender for Phase One of its AI medical large-model foundation and application project, with a budget of RMB 10.8m.

On 29 September, Guangdong Polytechnic of Science and Technology opened a public tender for an AI technology application resource library. A day earlier, Shaanxi Normal University's School of Artificial Intelligence and Computer Science opened a tender for equipment renewal at its future teacher disciplinary experiment centre.

On 24 September, the Development, Industry, Information Technology and Science Bureau of Pu County, Shanxi, published its procurement intentions for September to October 2026. One AI project in that list was flagged by industry media as being of hundred-million-yuan scale.

Put the five together and the outline is clear: AI procurement is moving downstream from "provincial mega-platforms and general contracting" to "university laboratories, vocational training rooms, hospital departments and county-level offices". Contract values are spread wide, from a few hundred thousand to tens of millions, with every tier represented.

What does this mean for smaller firms?

Tenders worth tens of millions are contests of integration and delivery systems. A RMB 160,000 tender is a contest of something else entirely: delivering 43 boards on time, issuing invoices, and having clean qualifications.

Top-tier integrators will not bid for a RMB 160,000 development board order. That slot is empty.

The barrier is not technical specifications

Read the Guangxi University of Science and Technology notice from start to finish and the sticking points are obvious.

On qualifications, bidders cannot be listed as dishonest judgment debtors, parties to major tax violation cases, or in records of serious dishonest conduct in government procurement. They must also meet the qualification conditions under Article 22 of the Government Procurement Law of the People's Republic of China. Consortium bids are not accepted, nor are subcontracting or transfer.

On timing, procurement documents can be collected from 28 September to 9 October during normal working hours. Weekends and public holidays are excluded, and collection is in person — no mail. Response documents are due by 11am on 10 October; late submissions will not be accepted.

On procedure, suppliers that have not collected the procurement documents cannot bid. Submissions must be sealed, the legal representative must appear in person with the original ID card, and incomplete materials are rejected outright.

The technical requirements are detailed enough that they are not the hard part.

What actually deters bidders is this sequence: you first have to know the notice exists, then between 28 September and 9 October, on a working day, take a copy of your ID, business licence and authorisation letter to Room 210A, Science and Education Building, No. 2 Wenchang Road, Chengzhong District, Liuzhou, to collect the documents.

From posting to submission: 12 days. With a long holiday in the middle.

There are two windows, and most people watch only the second

illustration

Another easily overlooked point is the advance publication of procurement intentions. Under the rules, procurement intentions must be disclosed before a formal tender. The Pu County notice for September to October is one such case.

So there are two windows, not one.

The first runs from intention disclosure to formal posting, usually several weeks, used to probe requirements, draft a plan and judge whether you qualify. The second runs from formal posting to submission, often only about ten days, used to write the bid, with almost no room for trial and error.

The first is easy to let slip. The second is easy to miss entirely.

The buyer side is changing too. On 11 September, the Hainan Provincial Government Procurement Centre introduced an AI assistant to help handle challenge responses. Around the same period, information from public resource trading platforms across the country frequently mentioned credit management. Buyers are digitising. If sellers still rely on manually trawling web pages, the rhythm will not match.

How to act

Information is not something you can keep up with by human vigilance alone.

Every day, government procurement and bidding notices nationwide are scattered across provincial and municipal platforms, industry sites, university websites and hospital websites. Keywords range from "AI development board" to "large model foundation" to "training room equipment", and each must be combined with regional filters. A person checking five or six websites a day will start missing things by the third week.

This is the step LEIDA (招标雷达) handles: it aggregates government procurement and bidding information nationwide, with real-time subscription push by keyword, region and industry. Set terms such as "artificial intelligence", "teaching instruments", "large model" and "practical training" alongside the provinces you can actually cover, and you receive the notice the day it appears — rather than discovering on 9 October that documents must be collected in person.

There is no need to start big. First copy ten notice titles that have appeared in your niche over the past month, break them into keywords, subscribe, and watch for two weeks to see whether the push is accurate. If it is, then talk about bidding.

Policy sets the water level; each tender lot is a small boat. Whether you can get on board often depends only on whether you saw the notice the day it was posted.

AI procurementChina government tendersSME supplierspublic sector AItender monitoring

閱讀繁體中文版 →