Hong Kong and Hangzhou are funding SME AI adoption, but the real test is your use case

Hong Kong and Hangzhou have both moved in 2026 to push AI spending down to small and medium-sized firms. The money is real, but it is mostly matching or reimbursement-based, which means the application, not the ambition, decides who gets paid. The harder constraint is being able to describe a concrete use case.

Hong Kong has rewritten the rules this year

According to the Chief Executive's 2026 Policy Address, support falls into two broad categories with very different thresholds.

The first is cash. Under the BUD Fund, the Easy BUD scheme raised its application ceiling to HK$150,000 in June this year, and the BUD Fund itself now offers more targeted funding for AI projects. Also in the cash category is the Digital Transformation Support Pilot Programme (DTSPP), a matching-fund scheme that helps SMEs buy and apply digital solutions. The Policy Address states it will be optimised so more SMEs can adopt AI and cybersecurity tools.

The second is resources. The Hong Kong government has allocated HK$3bn to subsidise companies, universities and R&D institutes using computing power at the Cyberport AI Supercomputing Centre, with the subsidy ratio generally no less than 70%. For manufacturing, engineering and testing, there are the New Industrialisation Funding Scheme and the New Industrialisation Acceleration Scheme, the latter offering up to HK$200m per enterprise.

Supporting infrastructure is being filled in too. The Productivity Council will expand the functions of its "Digital Self-Service" platform, consolidate SME advisory services, and provide advice and technical support for AI application scenarios.

The thresholds can be remembered simply: cash schemes ask whether you are an SME and whether you will put up part of the cost yourself; resource schemes ask whether you will actually consume the computing power. A subsidy is not a certificate of merit. It is a match.

Hangzhou's Yuhang district is worth studying

Hong Kong gives money. Yuhang gives a complete path, and it is worth taking apart.

The background is straightforward: Yuhang wants to build an "AI+OPC" innovation and entrepreneurship hub, and its goal is to pull founders into the district. The approach is not a single grand document but money distributed across district and sub-district levels, with differentiated top-ups.

Future Sci-Tech City issued twelve measures on a district-wide friendly innovation ecosystem, covering funding support, computing power and token subsidies, and talent allowances. Liangzhu New City's "Eight Measures for Digital Habitat" are more specific: model procurement is subsidised at 30%, up to RMB250,000, and computing power procurement at 50%, up to RMB500,000. Cangqian and Wuchang sub-districts have issued ten measures on AI+OPC industrial support, focused on scenario matching, community building and talent housing.

As a result, according to Xinhua, Yuhang district has received 3,560 OPC project applications and 687 projects have moved in. Locals summarise the model in one line: from "people looking for policy" to "policy looking for people".

Two lessons are worth copying. First, do not fixate only on national or provincial documents; the real top-ups are often at district and sub-district level, smaller in amount but more practical in ratio and faster to apply for. Second, most subsidies are paid as a proportion of procurement — figures like 30% and 50% mean you need real purchase contracts and invoices first, and the money comes back afterwards or in proportion.

Timing, and the real bottleneck

On timing, Hong Kong is relatively easy to confirm: the new Easy BUD ceiling took effect in June this year, and the computing power subsidy is an ongoing policy arrangement. In Yuhang, 3,560 applications received shows the door is open, but district-level policies generally roll over annually or in batches, so it is best to check the current batch's terms before preparing materials rather than working from an old version.

The materials do not specify a deadline. Anyone who gives you a definite date on this is probably not reliable.

The real bottleneck is not the paperwork. It is the scenario.

What reviewers want to see is not "I want to use AI" but "this AI solution solved this specific problem for me, saved this much labour, and this is why the purchase was necessary". Many small teams stall here: they get the money, and their first instinct is to buy computing power, but they have neither a technical team nor a use case they can explain clearly. The budget runs out and the paperwork cannot be closed into a loop.

How to get started

For a small team with no technical staff, doing one thing that produces a visible result is more worthwhile than buying computing power first.

Content customer acquisition, for example, has a clear labour cost and is one of the easiest things to explain in terms of input and output. For this step, consider SHEYU AGENT: 16 industry advisers each covering a professional area, and 34 zero-threshold tools covering Xiaohongshu posts, AI posters and copywriting, with one login working across desktop, mobile and web. It addresses the specific problem in this article — being able to say in an application what you actually did with AI, rather than writing "planning to introduce AI". Get the tool running first, and both the cost structure and the results become describable. Then, when you discuss computing power subsidies, the logic holds together.

Subsidy windows open annually and the rules are tweaked each year. Those who can explain their own scenario clearly will always get the money before those who only watch the amount.