IN Budget 2027, Prime Minister Datuk Seri Anwar Ibrahim announced a RM30 million allocation for the Malaysia Digital Economy Corporation (MDEC) to help 4,000 micro, small and medium enterprises (MSMEs) adopt artificial intelligence (AI), and to train and certify 5,000 AI professionals. Separately, nearly RM15 million will go to AI Malaysia Bhd to strengthen a safe and ethical AI ecosystem, supporting a target of 200,000 skilled AI workers.

But can businesses actually turn access to AI into productivity gains? Funding adoption is a necessary first step, but the harder task is building the capacity to implement the technology, measure its value and scale what works.

A May 2026 discussion paper by the Khazanah Research Institute (KRI), developed with MDEC, found that digitalisation among Malaysian micro and small enterprises remains broad but shallow. Many have a digital presence, but fewer have progressed to integrated, data-driven operations, held back by financing constraints, skills gaps and fragmented support.

AI faces the same problem. Buying a subscription is easy; preparing reliable data, integrating software, training employees and redesigning processes are much harder.

Under the New Industrial Master Plan 2030 (NIMP 2030), Malaysia aims to develop 3,000 smart factories by 2030. Yet on Oct 2, 2026, Deputy Investment, Trade and Industry Minister Sim Tze Tzin said only 93 manufacturers had completed the transformation, with another 161 in progress.

Consider a Malaysian manufacturer supplying components to a multinational electronics company. AI-assisted visual inspection could identify defective parts, and predictive maintenance could reduce unplanned downtime. But both require reliable production records, compatible equipment, and employees who can interpret the results. For smaller suppliers on tight margins, such investments are hard to justify when returns are uncertain, yet failing to upgrade could cost them higher-value orders.

The RM30 million (about US$7 million) allocation is a positive step, but its impact will depend on how the money is distributed and what it enables businesses to achieve. With no details on eligible expenses or support per firm, it would be premature to judge it on the headline figure alone.

The more important question is whether the programme addresses the actual costs of adopting AI productively. If support mainly covers software subscriptions or equipment, businesses may still struggle with data preparation, system integration, training and maintenance, the costs that decide whether an AI project works or sits unused.

Regional benchmarks show where peers are putting their money. Singapore set aside S$150 million (about US$112 million) in 2025 for its Enterprise Compute Initiative, which provides firms with AI tools and cloud computing, as well as training and consultancy to integrate them. Thailand approved 25 billion baht (about US$770 million) for its national AI framework over fiscal years 2026 and 2027. Vietnam's new AI law creates a National AI Development Fund that prioritises startups and SMEs, alongside an AI voucher scheme.

Though not like-for-like, these show neighbours funding implementation and capability, not only access. Malaysia's own experience points the same way: Budget 2026 offered MSMEs an additional 50% tax deduction for AI and cybersecurity training, yet the skills gap persists.

A uniform subsidy also risks ignoring differences in readiness. A microenterprise relying on paper records may need basic digitisation first; a retailer with an online operation may be ready for demand forecasting; a manufacturer may need specialised software and technical support. Identical packages could fund tools some cannot use while giving others too little.

The government should therefore judge the programme by two questions: how many businesses can it realistically help overcome their implementation barriers, and what measurable improvements should they achieve in return? The goal is not to maximise recipients, but to remove the constraints holding back viable AI projects.

Malaysia should build its AI support around specific business problems, not treat adoption as an outcome in itself.

The Malaysian Investment Development Authority (Mida), the Malaysia Productivity Corporation and Intel Malaysia ran the Artificial Intelligence for SMEs (AI4S) programme, giving 100 SMEs AI starter kits and training for tailored pilots. By 2022, Mida reported 31 completed pilots in areas such as quality assurance, monitoring and detection, showing that businesses benefit when technology is applied to defined needs.

Budget 2027 should adopt a tiered model, starting with an assessment of each firm's digital maturity. Support could then range from basic digitisation to advanced AI integration, with matching grants or vouchers covering data preparation, system integration, training and cybersecurity, and low-interest financing for larger investments. Payments should be tied to implementation milestones, not software purchases.

Projects should target clear business value, such as demand forecasting to cut spoilage for food producers. Each should set a baseline and measure results, so successful pilots can be copied by similar firms and failed ones still yield recorded lessons. Public funding should support the successful application of technology, not merely its purchase.

If Malaysia wants AI to strengthen its economy, programme evaluation must go beyond adoption statistics. The number of businesses receiving grants, employees completing training and AI tools deployed can indicate implementation progress, but cannot establish whether firms have become more productive or competitive.

Penang-based ViTrox, which makes machine-vision inspection systems for the semiconductor and electronics industries, presents the value of its AI tools in operational terms: fewer defects escaping the line, higher detection accuracy and more reliable equipment. A manufacturer receiving public support for AI-based inspection should likewise report its defect escape rate, downtime and yield before and after deployment.

MDEC should publish a regular performance scorecard covering participating businesses, implementation costs, continued usage and changes in business performance. Relevant indicators could include labour productivity, defect rates, processing times, operating costs and inventory losses, depending on the application. It should also track whether firms keep using AI after support ends, since a tool abandoned once funding expires signals an unsuitable solution or an incomplete implementation.

Where feasible, the government should compare supported businesses with similar firms that did not receive assistance. This would help distinguish improvements attributable to the programme from changes caused by wider economic conditions.

Talent development requires the same scrutiny. Budget 2027 sets a target of 200,000 skilled AI workers, while MDEC's programme will train and certify 5,000 AI professionals. These targets may cover different groups and initiatives, but the government should clarify how they relate to one another and judge training by practical competencies, workplace application and employment outcomes, not enrolments and certificates.

Such reporting would allow policymakers to identify which interventions deliver results, redirect funding away from ineffective approaches and expand programmes that demonstrate measurable benefits.

Targeted funding and honest measurement need one more foundation: clear rules and clear owners.

Malaysia still has no law that specifically governs AI. The 2024 National Guidelines on AI Governance and Ethics are voluntary, and the MY-AI Standards platform is only at the first of three stages. Digital Minister Gobind Singh Deo told the Dewan Rakyat on Oct 5 that the finalised AI Governance Bill will be tabled early next year, with risk-based obligations and a new Central AI Authority. But it must still pass Parliament before any obligations take effect.

Of Budget 2027's eight AI-related measures, only one mentions safety or ethics. If the scam-check app wrongly flags a legitimate business account, nothing requires prior testing or an appeal route.

Responsibility is also scattered across at least seven government bodies, and some roles conflict: AI Malaysia Bhd drives the national AI action plan while also running the Malaysia AI Safety Institute, which judges whether AI is safe.

The Bill should set three clear levels. Builders, meaning the ministries, agencies and AI Malaysia Bhd, would promote AI through one reporting line. A checker, the Central AI Authority working with Nacsa and regulators such as Bank Negara Malaysia, would clear high-risk AI in hospitals, banks and public services, with lighter rules for low-risk uses. A referee, a Cabinet-level AI steering committee, would settle disputes, set missing deadlines and report to Parliament through the performance scorecard.

The government need not wait for the law. It should table the Bill on schedule, fund the future regulator, and require recipients of public money who deploy high-risk AI to follow recognised standards and report serious incidents now.

Malaysia does not need businesses merely to say they use AI. It needs them to produce more, waste less and work better, with AI that is safe and accountable. Funding the tools is not enough; funding the difficult work of making them useful, and being clear about who is responsible, will decide whether the investment pays off.

Tan Jia Xuen and Saravuth Maneerat are policy analysts on AI policy at the Higher Education Malaysia Association (Heya).