Taiwan AI Subsidies: SBIR, Cloud Market and Local Vouchers
Funding AI adoption via SBIR, Cloud Market or local vouchers? This guide compares the three, flags receipt and proof pitfalls, and gives a spec sample.
Disclaimer: Subsidy eligibility, reimbursable expense items, and acceptance criteria are determined by the organizing agency's announcements and your accountant's opinion. This article is a general overview and does not constitute a guarantee of approval or tax advice.
Why "adopting AI with a government subsidy" sounds simple but often gets stuck
When small and medium-sized enterprises want to adopt AI, the two most common questions are "Is there a subsidy I can apply for?" and "How can I get the application approved faster?" But owners and IT leads who have gone through the process often find that what slows things down is usually not "can't find a subsidy," but two easily overlooked practical issues. This article first explains which channels are available in Taiwan today, then the two bottlenecks, and finally how to describe "adopting AI" as a technical item that review committee members can understand and that passes acceptance.
Three common channels in Taiwan today
SBIR (Ministry of Economic Affairs Small Business Innovation Research Program)
SBIR has already included "digital transformation" (covering artificial intelligence, cloud, and big data) in its general program application areas, and it accepts applications on a rolling basis, so there is no need to wait for a fixed application window. It is an R&D-type subsidy that requires a complete written proposal explaining what you will do, how you will do it, and what results you expect. The application threshold is relatively high, but the funding scale and project period are usually larger.
Specific details such as subsidy amounts, subsidy ratios, and eligibility requirements are not clearly stated with concrete figures on the official announcement page, and this article will not estimate or claim them on its own. Please contact the organizing agency directly or an advisor familiar with the SBIR application process. Consultation hotline: 0800-888-968; email: sbir1@admail.csd.org.tw.
Cloud Market (Ministry of Economic Affairs SME Digital Transformation Tool Procurement)
Cloud Market is a digital tool procurement subsidy provided by the Ministry of Economic Affairs for small and medium-sized enterprises, and it uses a points system: after receiving points, companies purchase approved digital tools from listed service providers. The currently open program is the "ROC year 115 AI Tool Library" under the Industrial Sector, targeting small and medium-sized manufacturers and requiring that the tools solve specific problems such as "bottlenecks in production manufacturing and operations management."
Note that the points available to enterprises for ROC year 115 have already been fully allocated. If you are an enterprise looking to purchase tools, this round may already be closed. If you are a provider looking to list services and become a qualified supplier, the listing application deadline for service providers is October 20, ROC year 116, and the contact phone number is 02-2707-5055. Compared with SBIR, Cloud Market involves "buying ready-made tools" rather than "writing an R&D proposal," so the process is usually faster, but it is limited by the points quota and eligibility requirements.
Local subsidies (county and city digital transformation / digital vouchers)
In addition to the central SBIR and Cloud Market programs, many counties and cities also have their own SME digital transformation subsidies or digital voucher programs. These are usually smaller in scale, and conditions vary by region. This article has not verified the latest announcements for each county or city one by one. The only reminder is this: if the first two channels do not apply to you, it is worth checking with the industrial development bureau or SME office of your county or city to see whether there is a corresponding local program. Please refer to each county's or city's current announcements for details.
One-sentence comparison of the three
| Channel | Nature | Characteristics |
|---|---|---|
| SBIR | R&D-type | Requires a written proposal, large funding scale, longer period, rolling applications |
| Cloud Market | Procurement-type | Buy ready-made tools, faster, but limited by points quota and eligibility |
| Local | Tailored to local needs | Smaller scale, depends on each county's or city's announcements, requires individual inquiry |
The two practical bottlenecks most often overlooked before applying
Compared with "finding a subsidy," what usually determines whether an application can be smoothly reimbursed and accepted is these two issues:
Bottleneck 1: Receipts
If a company pays a foreign AI service provider directly (for example, by subscribing with a credit card), it usually receives an overseas receipt rather than a Taiwan uniform invoice. Such receipts often get stuck during expense reimbursement: accountants are unsure whether they can be used, and review committee members find it difficult to directly recognize them as eligible outsourced technical service expenses. The most direct solution to this problem is to purchase AI services from a supplier with a Taiwan company entity that can issue formal uniform invoices.
Bottleneck 2: Proof of results
"Asking employees to use ChatGPT" sounds like adopting AI, but it does not constitute an acceptable technical adoption item. Review committee members look for a concrete system architecture, quantifiable usage data, and deliverable technical specifications, not a statement that "employees are using it."
How to write "adopting AI" as a technical item that passes review
Below are technical specification directions that can be specifically cited in a proposal (adjust based on the actual adoption content):
- Multi-model API endpoint: connect to multiple model providers through a single OpenAI-compatible endpoint, without applying for a separate account with each one
- Project-level keys and spending caps: create individual API keys by department or sub-project, set spending caps to avoid overspending, and make future cost allocation easier
- Exportable usage reports: records of usage and cost broken down by model, project, and time period, which can serve as concrete evidence for project progress and expense reimbursement
- Data processing commitment: confirm whether the service provider has a written commitment not to use your data for model training, which is especially important for scenarios involving internal company documents or customer data
Supporting materials that can be prepared during the acceptance stage usually include: account activation records, project key creation records, and exported usage report files for a specified period. These are concrete, verifiable execution evidence, and are more likely to pass review than subjective descriptions such as "employees reported it was helpful."
If you need an AI technology supplier that can issue uniform invoices
Take BazaarLink as an example: it is operated by a Taiwan company (Jilian Technology Co., Ltd.), and subsidy applicants can use it as a supplier for the AI technology service item in their project. The electronic uniform invoice line item can be issued as "software service fee." Pricing is billed in USD and quoted in NT$, and companies can negotiate monthly billing terms (Net-30, negotiable). For high usage volumes, pricing can also be negotiated separately based on scale. The usage report export function mentioned earlier can likewise be used directly as supporting evidence for project execution and expense reimbursement.
Below are some model unit prices as of the query date, September 8, 2026 (per 1M tokens, in USD). They are for rough reference only; please refer to the real-time prices shown on the models page for actual figures:
| Model | Input $/M | Output $/M |
|---|---|---|
| openai/gpt-5.6-luna | $0.20 (promotional price) | $1.20 (promotional price) |
| anthropic/claude-sonnet-4.6 | $3.00 | $15.00 |
| google/gemini-3.7-flash | $0.75 (promotional price) | $3.75 (promotional price) |
| deepseek/deepseek-v4-flash | $0.20 | $0.40 |
Conclusion
Whether a government subsidy can be obtained and whether it can be smoothly reimbursed and accepted often depends not on "whether the subsidy exists," but on whether receipts and proof of results have been thought through before applying. Rather than waiting for review committee members to return the application and only then discovering the problems, it is better to prepare these questions before writing the proposal: how AI will be adopted, which supplier will be used, and how you will prove it is being used. That way, whether you ultimately go with SBIR, Cloud Market, or a local subsidy, you can avoid some unnecessary detours.
FAQ
What government subsidies can SMEs apply for to adopt AI?
There are three common channels: SBIR (R&D-type, requires a written proposal, accepts applications on a rolling basis), Cloud Market (procurement-type, buy approved tools with points; the enterprise points for ROC year 115 have already been allocated), and county- or city-level digital transformation or digital voucher programs (smaller in scale, according to each county's or city's announcements). For subsidy amounts and eligibility, please contact the organizing agency.
Can a receipt from a credit card purchase of ChatGPT be used for subsidy reimbursement?
Overseas receipts often get stuck during expense claims, and accountants and review committee members find it difficult to recognize them as eligible outsourced technical service expenses. The most direct solution is to purchase AI services from a supplier with a Taiwan company entity that can issue uniform invoices.
Does asking employees to use ChatGPT count as adopting AI?
Review committee members look for a concrete system architecture, quantifiable usage data, and deliverable technical specifications, such as a multi-model API endpoint, project-level keys with spending caps, exportable usage reports, and data-processing commitments. Simply saying "employees are using it" is not enough.
What should I prepare as acceptance evidence?
Account activation records, project key creation records, and exported usage report files for a specified period. Specific, verifiable execution evidence is more likely to pass review than subjective descriptions.
TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API