Taiwan AI Industry Upgrade Forum Series |... | TAIWANinside 晶片島上的光芒

The Taiwan AI Industry Upgrade Forum Series held two sessions this year, each during a major industry trade show. On August 20, “Bringing Fab-Grade Smart Manufacturing to Small and Medium-Sized Factories” was held at Automation Taipei. On September 4, “Keeping Data On-Site, Securing Compute Autonomy” was held at SEMICON Taiwan. The organizers collected company views both through the registration process for the two forums and through the post-event “Survey on AI Adoption in Taiwan’s Manufacturing Sector,” receiving responses from 132 participants. This article summarizes the main findings and the policy recommendations that follow from them.

Nearly 60% of companies already use AI; only about 10% have scaled it across the factory

Among manufacturing respondents, nearly 60% of companies already use AI, but only about 10% have reached scaled use across multiple processes or multiple plants. Of the companies that have tried AI, about half have discontinued at least some projects.

Among companies that have adopted AI, 70% say the result they have felt is that “individual staff save time on their own work.” About 30% have seen improvements in production metrics such as yield and delivery times; about 10% have seen revenue growth or a clear drop in costs; and about 30% say they still see no concrete results.

There is also a gap between what companies think about AI and what they actually do. More than 90% believe AI will fundamentally change the way work is done in their industry, but fewer than 10% have set a clear budget and timeline, and about half say they are “interested, but have no timeline yet.”

Most companies have never heard of the government’s AI and digital-transformation resources

For each of the five government resources listed in the survey, more than half of the companies had never heard of it. More than 90% had never heard of “Industrial AI Diagnostic Consulting” or the Cloud Marketplace Industrial Pavilion subsidy; about 80% had never heard of the special loan program for micro, small and medium-sized enterprises (MSMEs); and nearly 70% had never heard of the investment tax credit for AI equipment. Even the best-known resource, the Small Business Innovation Research (SBIR) program, was unfamiliar to more than half of the companies.

Among companies that had heard of these resources, the most common answer was “We are not sure whether we qualify.” Fewer than 10% had actually used any of them. The greatest difficulty companies reported when applying was that “preparing the documents takes too much time,” selected by nearly 40%. About 20% said the subsidy is too small to be worth the effort of applying.

Companies also do not treat learning about policy resources as a priority. Among the seven topics registrants could choose when signing up for the forums, “How to use government subsidies, loans and other policy resources” ranked sixth, selected by only about 20% of manufacturing respondents.

What actually makes companies invest is major-customer requirements and peers’ results

Subsidies do raise companies’ willingness to try AI. When the survey described a lightweight tool that “does not require changes to shop-floor operations, with the government covering most of the cost,” about 80% of companies said they would be willing to try it.

When asked what would be most likely to make their company actually take action, however, no company chose “subsidies that cover most of the cost.” The two most common answers were “a major customer has put forward digitalization or audit requirements” and “seeing peers adopt AI with concrete results,” each chosen by about a quarter of companies.

On payment models, about 30% of companies prefer “a fixed monthly or annual fee with a predictable amount,” while fewer than 10% are willing to accept usage-based pricing.

Production scheduling still runs on Excel; data foundations and a person in charge are the bottlenecks

The scenarios in which companies most want to apply AI are production-process improvement and quality inspection. Yet nearly 70% of companies still schedule production with Excel, paper records, or the experience of veteran shop-floor technicians. About a quarter have a manufacturing execution system (MES)—software used to record and manage production on the line—and about 10% have machines that are mostly networked and collect data automatically.

When registering for the forums, nearly half of manufacturing respondents selected “building and governing in-plant data,” which ranked second among the seven topics.

The most common obstacle companies face in adopting AI is “not knowing where to start the evaluation,” selected by more than 40%. Next is “lacking someone who can take charge of this,” selected by nearly 40%. When companies named the obstacle with the greatest impact, these two together accounted for about half of the answers.

Generative AI is already part of everyday office work. At about 80% of companies, staff use tools such as ChatGPT to handle quotations, emails, and reports. At about 30% of the companies that use these tools, staff use only their personal accounts, so company data may leave the firm through private accounts.

Companies that have already gone overseas most need local partners

More than 70% of manufacturing respondents already have overseas operations, mainly in mainland China, Vietnam, and the United States. What companies most lack when expanding overseas are local partners—such as system integrators and agents—selected by more than 60%. Next are local market information and channels, and local operations talent, each selected by more than 40%. Fewer than 10% listed “going overseas as a team with other Taiwanese firms” as a need.

Policy recommendations: attach government resources to the way companies actually decide

Government support for SME adoption of AI has been put in place over the past two years. In October 2025, the Industrial Development Administration (IDA) of the Ministry of Economic Affairs (MOEA) set up an industrial competitiveness advisory team that provides diagnostics, on-site consulting, and support for introducing AI tools. As of March 2026, 197 companies had received subsidies totaling about NT$900 million; in July 2026, the MOEA announced that the team had advised a cumulative 9,249 companies on introducing AI applications. The MOEA’s Small and Medium Enterprise and Startup Administration (SMEA) runs a program to improve the smart-operation performance of SMEs, offering company diagnostics, AI application courses, and implementation consulting; its program for AI innovation and application in MSMEs also promotes an approach in which large firms bring smaller ones along. On tax policy, an amendment to Article 10-1 of the Statute for Industrial Innovation added equipment and services related to AI and to energy saving and carbon reduction to the investment tax credit; licensed software may also qualify. In September, the Executive Yuan approved, at its cabinet meeting, a plan to support and strengthen MSMEs, which will invest NT$100 billion over eight years in a transformation and upgrading development fund. The MOEA is also drafting the Act on the Transformation, Upgrading and Development of Micro, Small and Medium Enterprises, which it expects to send to the Legislative Yuan for review in October.

Set against the survey, these policies show that the government already offers diagnostics, consulting, subsidies, and tax incentives. Where companies get stuck is mainly in how the resources reach them. To go from learning that a government resource exists to actually using AI to improve operations, a company must pass through five stages: awareness, eligibility assessment, application, adoption, and continued use. Respondents reported getting stuck at every stage.

First, the resources already exist, but companies do not know about them. The advisory team has already advised more than 9,000 companies, and the Cloud Marketplace Industrial Pavilion also offers point-based subsidies for AI tools. Yet more than 90% of respondents had never heard of the diagnostic consulting service or the Cloud Marketplace Industrial Pavilion subsidy listed in the survey. The investment tax credit now covers AI-related equipment, services, and licensed software, but nearly 70% of companies had never heard of it, and most of those that had were unsure whether they qualified. The government could therefore build a single cross-ministry search portal: by entering its industry, size, and the problem it wants to solve, a company could see which resources it can apply for and assess its own eligibility. Small subsidies could also be tiered by amount, so that the burden of preparing documents is proportionate to the size of the subsidy.

Second, having large firms bring smaller ones along matches the moment when companies actually act. This survey shows that companies invest in response to major customers’ digitalization or audit requirements, and to concrete results achieved by their peers; no company named subsidies as the most likely reason to act. SMEA has already proposed having large firms bring smaller ones along. The government could expand this into its primary outreach channel: when large manufacturers set data and audit requirements for suppliers, the government could deliver diagnostic and subsidy resources to those suppliers at the same time, and turn peers’ adoption results into case studies that companies can readily understand.

Third, the form of subsidies should match how AI is paid for. About 30% of companies prefer a predictable fixed monthly or annual fee; fewer than 10% are willing to pay by usage. The Cloud Marketplace Industrial Pavilion’s point-based subsidies for AI tools already help companies buy AI tools. The government could go further and study ways for subsidies to cover the service fees that continue after adoption, and help companies understand and manage the cost of usage-based AI services.

Fourth, lay the data foundation before adoption, and fill the gap in who is in charge. Nearly 70% of companies still schedule production with Excel, paper records, or technicians’ experience, and “not knowing where to start” and “having no one in charge” are the obstacles companies encounter most often. The government could make the building of factory data a priority for support before AI adoption, and use external consultants who accompany companies through adoption—or shared-talent arrangements—to fill the gap where no one inside the company is responsible. Because staff at about 30% of the companies that use generative AI rely only on personal accounts, the government could also issue data-security guidelines for companies using generative AI.

Fifth, judge policy results by continued use. About half of the companies that have tried AI have discontinued at least some projects. Published policy results currently focus mainly on the number of companies advised and the number subsidized. Once the Act on the Transformation, Upgrading and Development of Micro, Small and Medium Enterprises enters the legislative process, the government could include “whether companies continue to use AI, and whether they extend it to other processes” among its performance indicators.

Sixth, shift overseas-expansion resources toward matching local partners. More than 60% of companies with overseas operations most need local system integrators and agents, while fewer than 10% want to go overseas as a team with other Taiwanese firms. The government’s overseas-expansion resources could focus on matching companies with local system integrators and agents.

Survey methodology

The data come from registrations for the two Taiwan AI Industry Upgrade Forum Series events held on August 20 and September 4, 2026, and from the Survey on AI Adoption in Taiwan’s Manufacturing Sector conducted from August to September 2026, with a total of 132 respondents. Each percentage is calculated from the respondents who answered that question and is expressed as a range. All respondents were company representatives who took part in the forums or the survey voluntarily; the percentages reflect only the views of participating companies and do not represent Taiwan’s manufacturing sector as a whole.