Malaysia has spent the last three years winning the AI infrastructure race. Billions in data centre commitments. Johor turning into a server farm belt. Hyperscalers queuing up with press releases and groundbreaking ceremonies. And yet, ask the average Malaysian SME boss whether AI has changed how they run the business, and you will get a blank look. Maybe a shrug. Maybe a polite "we tried ChatGPT once lah."
That gap is the subject of a new analysis making the rounds ahead of Budget 2027, and it makes an argument we think every Malaysian business owner should read: the country is pouring money into the bottom layer of the AI stack while almost nobody is funding the layer where value actually shows up.
What the analysis actually says
The core claim is simple. Data centres are landlord infrastructure. They generate construction jobs, electricity demand, property deals, and impressive investment figures for ministerial speeches. What they do not generate, on their own, is productivity in the rest of the economy.
A data centre in Sedenak does not teach a kitchen supplier in Klang how to forecast demand. It does not help a Kuching accounting firm automate its bookkeeping. It does not show a Penang manufacturer how to cut defect rates with computer vision. The compute sits there, rented largely by foreign companies serving foreign users, while local adoption crawls.
The piece points out that SME uptake of AI tools in Malaysia remains low. Not zero, but low enough that the disconnect between investment headlines and ground reality is embarrassing. And SMEs are not a side story here. They are roughly 97 percent of Malaysian business establishments and employ close to half the workforce. If they do not adopt, the national productivity number does not move. Full stop.
Why infrastructure alone does not convert
Here is the part worth understanding properly, because it explains why good intentions keep producing underwhelming results.
The AI stack has roughly four layers. At the bottom sits physical infrastructure: power, cooling, buildings, network. Above that, compute and chips. Above that, models and platforms. At the top, applications and the humans who use them.
Malaysia has invested aggressively in layer one. Some of layer two, through chip access and partnerships. Layers three and four are thin. Very thin. And economic value in AI is captured overwhelmingly at the top, where a specific business process gets faster, cheaper, or more accurate.
Think of it like the highway system. PLUS built beautiful roads. But the economic return came from the logistics companies, the factories, the tourism operators who used those roads to move goods and people. If Malaysia had built the North-South Expressway and then nobody bought lorries, we would have had lovely tarmac and a stagnant economy.
Right now, we are building tarmac. The lorries are the missing piece.
What is actually stopping SMEs
We hear the same blockers again and again from Malaysian business owners, and they are almost never about compute availability.
- Skills. Nobody in the company knows what AI can realistically do. They have heard the hype and seen the nonsense, and cannot tell the difference.
- Time. A 12-person company has no spare capacity for a three-month experiment. If it does not work in week one, it is dead.
- Trust. Worries about data leaking, hallucinations in customer-facing output, and compliance under the PDPA amendments.
- Cash. Subscription costs in USD hit differently when your revenue is in ringgit and your margin is 8 percent.
- Relevance. Most off-the-shelf tools assume English-first, US-centric workflows. Bahasa Malaysia, Manglish, mixed-script invoices, halal certification chains, SST filing: not well served.
Not one of those problems is solved by another 100MW of capacity in Johor. Every one of them is solved by money spent closer to the business.
The roadmap the piece suggests
Three priorities come through, and they are refreshingly unglamorous.
First, AI literacy at scale. Not university degrees in machine learning. Practical, short-form training for owners, managers, and clerks on what these tools do and where they break. The kind of thing that could run through MDEC channels, SME Corp, the chambers of commerce, and TVET institutions. Think two days, not two years.
Second, adoption grants aimed at implementation rather than procurement. The distinction matters enormously. A grant that pays for software licences produces shelfware. A grant that pays for a consultant to sit with the company for six weeks, map one process, and ship one working automation produces a changed business. Singapore learned this with its productivity schemes. The money has to fund the doing, not the buying.
Third, support for local applications. Malaysian-built tools for Malaysian problems: Bahasa Malaysia language handling, local regulatory workflows, agriculture and palm oil use cases, Islamic finance compliance, multilingual customer service across our actual language mix. This is where local firms can build defensible products instead of reselling someone else's API with a markup.
The uncomfortable arithmetic
Let us be fair to the data centre strategy, because it is not stupid. It attracts FDI. It builds digital sovereignty options. It positions Malaysia as a regional hub ahead of neighbours who moved slower. Those are real wins, and walking them back would be silly.
But the returns are narrow and the costs are not. Data centres are capital intensive and labour light. After construction, a large facility might employ a few dozen people. They consume serious electricity and water, which means the rest of us absorb grid pressure and tariff adjustments. Tenaga's capacity planning now bends around them. And the tax incentives that attracted them are money the government chose not to collect.
Compare that to what a fraction of the same forgone revenue could do if pointed at SME adoption. Say a programme reaches 20,000 companies with real implementation support. Even modest productivity gains across that base, compounding annually, outperform the employment and multiplier effects of another cluster of server halls. The ROI case is not close.
The honest framing is not data centres versus adoption. It is proportion. We have gone roughly 95 percent infrastructure, 5 percent everything else. The ratio needs inverting, or at least rebalancing hard.
What this means for you
If you run a Malaysian business, do not wait for Budget 2027 to decide your AI strategy. The tools are already cheap enough to matter, and the gap between companies that use them and companies that do not is widening every quarter.
Start with one process. Something repetitive, text-heavy, and low risk. Quotation drafting. Meeting notes into action items. First-pass customer email responses. Supplier document extraction. Pick the thing your team complains about most, and spend two weeks testing whether a model handles the boring 70 percent while a human checks the rest.
Measure it properly. Hours saved, error rate, turnaround time. Write the numbers down before and after. This matters for two reasons: you will know whether it worked, and you will have a documented case ready when adoption grants do appear. The companies that get funded first will be the ones that can already show a result.
And if you are in a position to talk to policymakers, through a chamber, an industry association, or a ministry consultation, say the quiet part out loud. Ask where the money for the top of the stack is. Ask what the target adoption rate is, and how it will be measured. Investment announcements are easy. Adoption metrics are the ones that hurt to publish, which is exactly why they are the ones that drive behaviour.
The bigger question
There is a version of the next decade where Malaysia hosts the region's AI compute and captures very little of the value created on top of it. Rent collected, margin exported. We become the place where AI happens rather than the place that gets better because of AI. That outcome is not hypothetical. It is the default if nothing changes.
There is another version where the data centres are the foundation and the SME base is genuinely more productive because 200,000 Malaysian businesses learned to use these tools well. Higher wages, better margins, local products that travel. That version requires spending on things that do not come with a ribbon-cutting ceremony.
Budget 2027 is the decision point. The infrastructure commitments are largely locked in and the money is flowing. The open question is whether the next tranche goes into more concrete and cooling towers, or into the far less photogenic work of teaching a nation of small businesses to actually use what we have built.
We would take the second option over teh tarik any day. Let us see what the numbers say when the speech is read.