In 2026, Taiwan SMEs applying for AI-related subsidies most often end up choosing between two programs run by the Ministry of Economic Affairs: a training-and-software subsidy capped at NT$100,000 (about US$3,200) per company for businesses with 30 or fewer employees, and the SBIR small business innovation R&D grant, which pays up to NT$12 million (about US$387,000) depending on the project phase. These serve very different purposes — the first funds staff digital-skills training and basic software purchases, and is a good starting point for companies still building internal capability; the second funds real R&D and prototyping, and suits companies that already have a concrete AI project in mind. This guide covers the main programs available today, their eligibility rules, application windows, and the most common reasons applications get rejected.
What AI and digital transformation subsidies are available to Taiwan SMEs in 2026?
Government support for SME AI adoption generally falls into two categories. The first is "capability-building" — getting staff to a baseline digital skill level and helping companies buy compliant software, with a lower funding cap but also a lower bar to qualify. The second is "R&D funding" — supporting companies that are building or validating actual technology, with higher caps but stricter review. Beyond these two recurring tracks, MOEA units including the Industrial Development Administration, the Department of Industrial Technology, and the Department of Commerce also publish project-specific digital transformation or AI programs on an ad hoc basis; names, amounts, and application windows for these change every year, so the SME Administration's Multi-Track SME Revitalization Platform is the best starting point — it aggregates digital transformation subsidies and advisory programs announced across agencies.
| Program | Administered by | Eligible applicants | Subsidy cap | Funding mechanism | Status as of September 2026 |
|---|---|---|---|---|---|
| SBIR (Small Business Innovation Research) | SME Administration, MOEA | Legally registered SMEs, any industry, project must be innovative | Phase 1: NT$1.5M (~US$48k) / 6 months. Phase 2: NT$6M (~US$194k) / 1 year or NT$12M (~US$387k) / 2 years. Phase 2+: NT$6M (~US$194k) / 1 year | Amount set by document and on-site review | Rolling online applications year-round, reviewed monthly — still accepting applications in September 2026 |
| 30-employees-or-fewer manufacturing digital transformation subsidy | SME Administration, MOEA | Manufacturers with 30 or fewer insured domestic employees | NT$10,000 (~US$320) per employee, NT$100,000 (~US$3,200) per company | Reimbursement; +10% bonus if wage-increase conditions are met | FY115 round closed to applications December 31, 2025; next round TBD per official announcement |
| 30-employees-or-fewer service-sector digital transformation subsidy | Department of Commerce, MOEA | Service businesses with 30 or fewer insured domestic employees | Same as above: NT$10,000 per employee, NT$100,000 per company | Same as above | Same as above — closed, next round TBD |
| Other digital transformation / AI programs (Industrial Development Administration, Dept. of Industrial Technology, etc.) | Varies | Varies by program | Per the administering agency's current-year announcement | Per the administering agency's current-year announcement | Check the SME Revitalization Platform for current listings |
Treat the "subsidy cap" column as a ceiling, not a guaranteed payout. SBIR's approved amount is set through document and on-site review, and the review committee decides based on the plan's feasibility and budget reasonableness — writing the maximum figure into your application doesn't mean you'll receive it. The two 30-employee training subsidies are reimbursement-based, but they're still capped per employee and per company, and each business can only claim them once. The "Status" column is also worth reading carefully: application windows for the same program shift year to year, and a program can close early mid-year if its budget is exhausted before the announced deadline. Always confirm the current window on the administering agency's own site or the SME Revitalization Platform before you start preparing an application — don't assume last year's or last round's dates still apply.
One point that matters for international founders: eligibility for these programs is based on how the applicant entity is registered in Taiwan, not on the nationality of its owners. A foreign-invested company that holds a valid Taiwan company or business registration and meets the headcount and industry tests can generally apply on the same terms as a locally owned SME — the main exclusion to watch for is SBIR's rule against mainland Chinese investment in the applicant, which is a distinct restriction from foreign ownership more broadly. If your Taiwan entity is newly registered, confirm your labor-insurance enrollment history covers the lookback period each program requires before you apply.
SBIR: the first stop if you already have a concrete AI project
If you already know what you want to build — a customer-service AI agent, an internal knowledge-base assistant, a defect-detection model on the production line — SBIR is currently the main R&D-track subsidy. It runs in three phases. Phase 1 (feasibility research) focuses on validating whether an idea is viable, capped at NT$1.5 million (about US$48,000) over six months. Phase 2 (development) funds actual technology build-out and prototyping, capped at NT$6 million (about US$194,000) for a one-year project or NT$12 million (about US$387,000) for a two-year project. Phase 2+ (value-added application) supports projects that have finished R&D and are moving toward commercialization, capped at NT$6 million over one year. SBIR isn't restricted by industry or technology theme — any legally registered SME with a genuinely innovative project can apply, provided there's no mainland Chinese investment in the company and no outstanding tax or compliance violations. Applications are accepted online year-round with monthly review meetings, not on a single annual deadline like most capability-building subsidies. As of the 2026 program year, SBIR has also added a patent-filing subsidy as a complement to funding the underlying R&D.
The 30-employee training subsidy: a low-barrier way to build capability before you commit to AI
If your company is still deciding whether to invest in AI and your team's digital skills aren't there yet, the 30-employees-or-fewer digital transformation subsidy is a better starting point. It actually exists as two parallel regulations with nearly identical terms: the manufacturing version, published by the SME Administration as the 30-employees-or-fewer manufacturing digital transformation subsidy guidelines, and the services version, published by the Department of Commerce as the 30-employees-or-fewer service-sector digital transformation subsidy guidelines. Both cover businesses with 30 or fewer insured domestic employees, and both fund the same thing: at least 12 hours of digital-skills training per employee, optionally paired with the purchase of software from the officially published, security-compliant list. Funding is reimbursement-based — up to NT$10,000 (about US$320) per employee who completes training, capped at NT$100,000 (about US$3,200) per company, and available only once. Companies that raise participating employees' wages by a set margin within the funding period can receive a further 10% bonus on top of their approved amount. One important caveat: the FY115 round closed to new applications on December 31, 2025, with training required to be completed and documentation submitted by March 31, 2026. There is no newer round announced as of this writing — check the SME Administration's or Department of Commerce's website directly for whether a follow-up round opens.
Other subsidy channels — and where to check for current announcements
Beyond these two recurring tracks, MOEA units including the Industrial Development Administration, the Department of Industrial Technology, and the Department of Commerce publish additional project-based digital transformation or AI programs each year, covering topics like smart manufacturing, service-sector digital upgrades, or AI adoption advisory support. These typically come with fixed application windows and limited slots, and the eligible applicants, amounts, and cost-share ratios change year to year — don't assume last year's terms still apply. The more reliable approach is to check the "Digital Transformation" section of the SME Multi-Track Revitalization Platform regularly; it's run by the SME Administration and aggregates programs announced by other agencies, and its hotline (0800-280-280) can help confirm eligibility and required documents for a specific program.
Eligibility self-check before you apply
Before submitting, run through the following checklist to avoid the most common causes of rejection:
| Item | What to check |
|---|---|
| Registration status | Valid company, business, or tax registration, not in dissolution or closure |
| Headcount test | Based on labor-insurance enrollment; the 30-employee tracks require meeting the headcount test in at least one of the six months before applying |
| Industry classification | Service-sector applicants must match their tax registration code against the published eligible-industry list |
| Ownership and compliance history | SBIR excludes companies with mainland Chinese investment, and requires no outstanding suspension or tax violations |
| Double-funding restriction | The same expense cannot be claimed under more than one government subsidy |
| Documentation timing | Receipts, invoices, and training or execution dates must fall within the announced subsidy period — retroactive documentation is typically not accepted |
Preparing your documents and syncing the timeline with your project plan
For both capability-building and R&D subsidies, start pulling documentation together before the application window opens, not after the announcement drops. Documents worth preparing in advance typically include: company or business registration proof; a recent labor-insurance enrollment record (used to verify headcount against the eligibility threshold — for the 30-employee tracks, you need to meet the headcount test in at least one of the six months before applying); financial or tax records; a project or training plan; quotes or purchase documentation if the application includes software purchases; and a self-declaration of whether you've applied for a similar subsidy before. The labor-insurance record in particular usually has to be requested separately from the Bureau of Labor Insurance and can take several business days to arrive — request it before the window opens so you're not stuck waiting on paperwork once submissions are open.
The subsidy review timeline should be built into your overall project plan rather than treated as something you wait on before deciding what's next. A steadier approach is to split your AI adoption project into a part that can start regardless of subsidy approval and a part that depends on it — for example, funding an internal digital-skills assessment and basic process mapping out of your own budget first, since that work isn't gated by any review timeline, and holding the higher-cost prototype or system build until SBIR or another R&D subsidy comes back with a decision. That way, even if review takes longer than expected or the approved amount comes in below what you applied for, the first phase of the project still moves on schedule instead of the whole plan stalling on an uncertain outcome. If you're planning to apply for more than one program in sequence — training subsidy first to get staff up to speed, then SBIR for prototyping once the use case is validated — lay out each program's application window, execution deadline, and sign-off requirements on one timeline early, so the execution periods for different programs don't end up overlapping or conflicting.
Common mistakes and rejection reasons
Based on the disqualifying conditions listed in the official guidelines, these are the most frequent — and most avoidable — reasons applications get rejected:
- Failing to submit corrections within the notified deadline, or submitting incomplete corrections
- Receipt, invoice, or training/execution dates falling outside the announced subsidy period
- Delivered work that differs from what was described in the approved application
- Applying after the announced deadline, or after funds have been exhausted and applications closed early
- Claiming the same expense under more than one government program, or having already received a similar subsidy
- Failing to retain accounting records properly, leaving no documentation to support the claim during audit
Next Steps
Government subsidies can offset part of the cost of AI adoption, but the review timeline for these programs and your own project plan need to be aligned early — capability-building subsidies suit staff training and basic tooling, while R&D subsidies require a concrete technical plan and schedule to apply for. Neither track is designed to fund 100% of a real AI adoption project: they lower the entry cost of building internal capability or validating an idea, and the rest of the budget still needs to come from the business. Treat the subsidy application timeline as one input into your project plan rather than the driver of it — a strong technical scope with a realistic schedule tends to fare better in SBIR review than a proposal written primarily to match the funding cap.
If you're working out how to connect available subsidy funding to an actual AI adoption project, see Noise & Signal's Industry Digital Transformation services for how we help companies plan an AI adoption path, from scoping a fundable pilot through to production rollout. For a side-by-side view of process and pricing across our services, see our Process & Pricing page.