The Real Journey of AI Commercialisation — and the Model Malaysia Needs to Run It

AI innovation is not the same thing as AI research. Research produces knowledge — a new architecture, a more efficient training method, a novel application of an existing model to a new domain. Innovation is what happens when that knowledge is translated into something that changes how a task gets done, at a cost or quality point that makes adoption rational. A transformer variant published in a journal is research. The same idea embedded in a diagnostic tool that a hospital pays for every month is innovation. The distance between the two is the entire subject of this piece.

AI R&D itself is typically conducted in three overlapping tracks. Basic research happens mostly in universities and national labs, and asks foundational questions about learning algorithms, model architectures, and theoretical limits — work that is rarely commercial on its own but underwrites everything downstream. Applied research takes those foundations and points them at a defined problem — fraud detection, protein folding, crop yield prediction — usually inside a corporate lab, a university-industry consortium, or a government-funded mission programme. Applied engineering is where the model becomes a system: data pipelines, inference infrastructure, monitoring, and the unglamorous plumbing that determines whether something can run in production at 2 a.m. without a researcher on call. Countries that commercialise AI well tend to fund all three tracks deliberately, rather than assuming that basic research will spill over into products on its own.

The six-stage commercialisation journey

Stripped of jargon, the path from a lab result to a functioning market offering runs through six stages. Skipping any one of them is usually where "promising AI project" turns into "another pilot that never scaled."

How four regions run this system

No two governments are approaching AI commercialisation the same way, but each of the following four has a distinct, identifiable model — and each shows what happens when that model is executed with discipline versus when it is announced but under-resourced.

China — market-dominant, state-directed China's 15th Five-Year Plan (2026–2030) embeds AI as a priority on par with defence and treats commercialisation as a numeric target rather than an aspiration: 70% AI adoption across key sectors by 2027, 90% by 2030, with the AI industry expected to exceed 10 trillion yuan (roughly $1.4 trillion) in value by 2030.

• Government model: centralised five-year planning with explicit sector-level penetration targets, executed through provincial governments competing to attract AI investment.
• De-risking tool: AI pilot zones and demonstration zones that let companies validate commercial models with state backing before nationwide rollout.
• Compute strategy: subsidised electricity for data centres — provinces like Gansu, Guizhou, and Inner Mongolia cut cloud providers' power costs by up to 50%.
• Distribution strategy: aggressive open-source model releases to seize developer mindshare, mirroring the platform-lock-in logic Nvidia achieved with CUDA in hardware.

South Korea — mission-driven, centrally coordinated Korea's AI Basic Act, effective January 2026, created a National AI Committee under the president with legal authority to approve a triennial AI Master Plan and direct R&D strategy. The 2026 national AI budget is 10.1 trillion won (about $7.3 billion) — nearly triple 2025's 3.3 trillion won, and 28.6% of the entire national R&D budget.

• Government model: a presidential committee with eight domain subcommittees, giving AI policy the institutional weight normally reserved for economic or security matters.
• Capital strategy: a roughly $735 billion sovereign AI push blending Samsung's own $230 billion commitment with government R&D, industrial transformation, and infrastructure spending.
• Commercialisation mechanism: a staged consortium model — five industry teams (Naver, SK Telecom, LG, NCSoft, Upstage) compete for sovereign AI development funding, with the field narrowed in stages down to two survivors by 2027.
• Talent pipeline: ₩1.4 trillion (about $1 billion) committed in 2026 alone to train 11,000 advanced AI specialists.

United States — deregulation-led, private-capital-driven America's AI Action Plan, released July 2025, sets out more than 90 policy actions across three pillars: accelerating innovation, building infrastructure, and leading international AI diplomacy. Unlike China or Korea, the US model relies on private venture capital as the primary commercialisation engine, with the government acting mainly to clear regulatory friction.

• Government model: federal deregulation, streamlined permitting for data centres and semiconductor plants, and funding conditioned on states avoiding new AI-specific regulation.
• Commercialisation mechanism: public-private partnerships to build "full-stack AI export packages" — bundled software, hardware, and technical support sold as sovereign AI offerings to other countries.
• Capital depth: the US and China together account for the vast majority of global AI venture capital activity, according to OECD analysis of Preqin data — a scale advantage most countries cannot replicate and must instead work around.

Germany & the EU — funded but process-heavy Germany's High-Tech Agenda, adopted in July 2025, channels roughly €5.5 billion into AI, quantum, and adjacent deep-tech fields over the legislative period, rising from €500 million in 2025 to €1 billion annually from 2026. Germany's own analysts describe the country as "funded but stuck" — strong on capital allocation, slower on translating it into deployed products.

• Government model: ministry-led funding agendas paired with EU-level frameworks — the AI Continent Action Plan and Apply AI Strategy — that add coordination but also compliance overhead.
• Regulatory posture: the EU AI Act, in force since August 2025, imposes risk-tiered obligations that Germany is now embedding domestically through the KI-MIG Act, with the Federal Network Agency as market surveillance authority.
• Commercialisation signal: despite the regulatory weight, momentum is visible — German AI companies raised €1.7 billion in Q1 2026 alone, and the domestic AI market is forecast to grow from roughly €9 billion in 2025 to €37 billion by 2031.

Three constraints no country escapes

Regardless of government model, every country commercialising AI runs into the same three bottlenecks — they just hit them at different points on the journey.

Infrastructure

Compute is the most visible constraint and the most capital-intensive to solve. China's public and commercial compute capacity reached 246 EFLOP/s by mid-2024 against a 300 EFLOP/s target for 2025; Korea is racing to secure 260,000 high-performance GPUs by 2030; Germany's National Data Centre Strategy, adopted March 2026, aims to quadruple AI and high-performance computing capacity by 2030. The pattern is consistent: infrastructure investment now routinely runs into the hundreds of billions, and countries without sovereign compute risk becoming permanently dependent on foreign hyperscalers for a capability that increasingly defines economic competitiveness.

Data

Model quality is downstream of data quality, and most countries — including those with strong compute — struggle with the same issue: obtaining usable, sector-specific, rights-cleared data at scale. Generic internet-scale data is exhausted as a competitive differentiator; the frontier has shifted to proprietary, application-specific datasets that require industry cooperation, clear data-sharing law, and trust between the private sector and government to assemble. This is arguably the least visible constraint and the hardest to solve with money alone, because it depends on institutional trust as much as infrastructure.

Talent

Every country in this comparison is short of AI talent relative to its ambitions, which is a useful reminder that talent scarcity is a structural feature of this technology cycle, not a symptom of any one country's policy failure. Korea is spending ₩1.4 trillion in a single year specifically to close its gap. The difference between countries is not whether they face a shortage, but how directly their funding, immigration, and education policy respond to it.

Where Malaysia stands

Malaysia ranked 34th among 139 economies in WIPO's Global Innovation Index 2025, an improvement from 36th in 2023, and continues to perform better on innovation inputs than on the outputs that convert those inputs into commercial value — a gap that is, in effect, this entire article's thesis playing out at the national level. The government's own targets acknowledge this: gross expenditure on R&D was around 0.95% of GDP in 2020 against a stated ambition of 2.5% by 2025 and 3.5% by 2030, and Malaysia has separately set a goal of commercialising 500 products and solutions through its National Innovation and Technology Sandbox (NTIS).

2026 marks a genuine inflexion point in policy architecture. The National AI Action Plan 2026–2030 ("AI Nation 2030"), developed by the National AI Office and formally institutionalised in July 2026 as AI Malaysia Berhad under the Ministry of Digital, explicitly shifts the country's posture from the readiness-building focus of the 2021–2025 AI Roadmap to execution and commercialisation. The plan is structured around 14 sectoral initiatives and 14 enabler initiatives across five foundational enablers, and it sets numeric targets: a top-10 position in the Global AI Index by 2030, an additional 1.2 percentage points of GDP growth attributable to AI, and 300,000 AI-related jobs.

Budget 2026 backs this with real capital: RM5.9 billion for cross-ministerial R&D and Commercialisation & Innovation projects, a 50% tax deduction for AI training certified by the National AI Office, and a RM2 billion Sovereign AI Cloud. The commercialisation infrastructure already exists in pieces — Cradle Fund has backed more than 1,200 tech startups since 2003 with what it describes as the highest commercialisation rate among government grant schemes; MRANTI runs an AI Sandbox in partnership with Nvidia and hosts more than 150 companies and 30,000 knowledge workers at MRANTI Park; MDEC, MAVCAP, and the KL20 Action Plan target a top-20 global startup ecosystem by 2030. What has been missing is not the institutions — it is the connective architecture between them.

A feasible model: Bridge-and-Scale

Malaysia does not need to import Korea's presidential committee wholesale or Germany's compliance-heavy funding architecture. It needs a model calibrated to what it already has — strong institutional building blocks, a mid-sized economy, and a specific execution gap — rather than what larger economies can afford to build from scratch. The following four-layer model, which I'll call Bridge-and-Scale, draws on elements from each of the regions reviewed above but is sized to Malaysia's actual capacity to execute.

Layer 1 — Coordination AI Malaysia Berhad serves as Malaysia's answer to Korea's National AI Committee: a single body with real budget authority over the AI Nation 2030 targets, empowered to arbitrate among MOSTI, MITI, and the Ministry of Digital, rather than adding a fourth voice to an already crowded room.

Layer 2 — Bridge Funding A formally stage-gated fund pooling MOSTI, MRANTI's commercialisation grants, Cradle's CIP Spark/Sprint tracks, and MAVCAP, with explicit graduation criteria between POC, validation, and Series A — closing the exact valley-of-death gap that Germany's SPRIND challenge model and Korea's consortium funding are each designed to solve, but as one coordinated pipeline instead of parallel, disconnected schemes.

Layer 3 — Demand Anchoring Borrowing China's demonstration-zone logic and Germany's Agentic AI Hub model of pairing municipalities with start-ups, Malaysia anchors AI commercialisation in its own strongest sectors — semiconductor packaging and testing, palm oil and plantation, and NIMP 2030's manufacturing targets — using GLCs and government agencies as the first paying customer that de-risks a start-up before it has to win the open market.

Layer 4 — Capital & Talent KWAP's Catalyst Fund and the Strategic Co-Investment Fund provide growth-stage capital once Layer 2 has done its de-risking work, while NAIO-certified training tied to the 50% tax deduction and RPL pathways builds the mid-to-senior AI talent pool that 81% of employers currently cannot find — treating talent as a funded pipeline, the way Korea's ₩1.4 trillion commitment does, rather than a general aspiration.

The core design principle is that none of these four layers is new. Malaysia's task is not invention — it is integration, with clear ownership and clear graduation criteria at each handoff, so that a promising POC at MRANTI Park has a known, funded, time-bound path to a GLC pilot and then to institutional capital, instead of restarting the search for support at every stage.

Key action items

Key Item 1: Formalise the bridge fund with public graduation criteria. Publish the specific technical and commercial milestones that move a project from Cradle-stage funding to MRANTI-stage commercialisation support to MAVCAP or KWAP-stage growth capital, so founders and researchers can plan against a known pipeline rather than a maze of separate applications.

Key Item 2: Convert data centre investment into local commercialisation value. Attach conditions to major hyperscaler approvals — local compute credits for Malaysian AI start-ups and researchers, similar in spirit to the UK's Sovereign AI Fund GPU-hour allocation — so the RM144.4 billion infrastructure pipeline builds domestic capability, not just foreign capacity hosted on Malaysian soil.

Key Item 3: Anchor demand through GLCs in NIMP 2030 priority sectors. Mandate or incentivise GLCs in manufacturing, plantation, and semiconductors to serve as first-customer pilot sites for Malaysian AI vendors, following the demonstration-zone logic that China and Germany both use to de-risk early commercial deployment.

Key Item 4: Close the GERD-to-BERD gap with accountability, not just targets. The 2.5%-by-2025 and 3.5%-by-2030 R&D spending targets need a public tracking mechanism and a business-expenditure sub-target, since business-led R&D — not government spending alone — has historically correlated with commercialisation speed in top-performing GII economies.

Key Item 5: Fund AI talent as a pipeline. Extend the RM600 million AI skills allocation into a multi-year, budgeted commitment tied to NAIO certification and RPL pathways, with explicit retention incentives to compete with remote-work offers from Singapore and global tech firms that are currently pulling Malaysian AI engineers out of the domestic pipeline.

Key Item 6: Operationalise the AI Governance Bill ahead of mass deployment Finalise the AIGE-based regulatory framework with clear risk tiers before AI adoption scales past the pilot stage, giving investors and industry partners regulatory certainty without importing the EU AI Act's full compliance burden wholesale.

Key Item 7: Publish a public commercialisation scorecard Track and report GII rank, BERD/GERD ratio, MRANTI and Cradle graduation rates, and AI-attributable GDP contribution on a fixed cadence — mirroring the accountability structure Korea's triennial Master Plan builds into law — so AI Nation 2030's targets are auditable, not aspirational.

What this means for builders and decision-makers

For founders and researchers, the practical takeaway is that Malaysia's commercialisation infrastructure — Cradle, MRANTI, MDEC, MAVCAP — is more developed than the "talent gap" headlines suggest; the constraint is knowing which door to knock on at which stage, and that is a coordination problem the new National AI Office is explicitly meant to solve. For business leaders and GLCs, the opportunity is to become the first customer for a Malaysian AI vendor rather than defaulting to an established foreign platform — that single decision is what China, Korea, and Germany's models all treat as the pivot point between a promising prototype and a scaled product. For policymakers, the lesson from all four regions studied here is the same: coordination without budget authority is theatre, and capital without a formal bridge mechanism just funds more pilots that never leave the lab. Malaysia has the pieces. What happens next is whether they get assembled into a system, or left to operate as four good institutions solving the same problem in parallel.