About 714,000 Malaysians have never been online. In a country of 36.1 million, that is the entire remaining market for anyone whose business plan depends on connecting somebody new, and it is a rounding error. Malaysia crossed into near-total internet penetration without much ceremony, and in doing so it retired the growth strategy most consumer digital businesses in the region still run on. The next increment of value has to come from the same people doing more. Increasingly, the lever that moves them is Bahasa Melayu.
That is a harder sell than it sounds, because Malaysia is the one market in Southeast Asia where the English shortcut genuinely works. It just does not work as well as running both, and the gap is invisible unless somebody measures it.
A Market With No New Users Left
The scale is easiest to see through the data. DataReportal’s Digital 2026 report for Malaysia counted 35.4 million internet users at the end of 2025, or 98.0 percent of the population, alongside 30.7 million active social media identities and 44.0 million mobile connections in a country of 36.1 million people. Social identities grew by 5.6 million over the year, a jump of 22.2 percent, which sounds like expansion until you notice the user base underneath it barely moved. What grew was accounts, not people.
That has already reshaped the economy. The Department of Statistics Malaysia put ICT and e-commerce at 23.4 percent of the economy in 2024, worth RM451.3 billion, split between 13.9 percent from ICT gross value added and 9.5 percent from e-commerce in non-ICT industries. Close to a quarter of national output now rests on a user base that cannot get materially larger. Every point of growth from here is a question about intensity rather than reach.
With a median age of 31 and mobile connections running at 122 percent of the population, the country is short of neither appetite nor hardware. It is short of headroom.
The Language Premium Survives High English Proficiency
This is where market-entry assumptions about Malaysia tend to go wrong. Malaysians are good at English, genuinely and measurably. In the 2025 EF English Proficiency Index, drawn from 2.2 million test-takers across 123 countries, Malaysia ranked first in Asia and 24th in the world with a score of 581, up fifteen points on the previous year and comfortably inside the high proficiency band. Kuala Lumpur scored 588.
The geography of that proficiency is also less predictable than the cliche allows. Kuantan, not Kuala Lumpur, posted the highest city score in the country at 601, and Penang led the regions on 589. English competence in Malaysia is not a Klang Valley phenomenon that thins out with distance from the capital, which removes the usual excuse for treating English as the default and Malay as a regional accommodation.
The temptation, reading all that, is to treat the language question as settled and ship in English. But proficiency and preference are different variables, and the commercial one is preference. CSA Research’s Can’t Read, Won’t Buy study, run with Kantar across 8,709 consumers in 29 countries, found that 76 percent preferred to buy products with information in their own language and 40 percent would not buy from a website in another language at all. Understanding a page is not the same as trusting it enough to hand over a card number.
There is now a machine-readable version of the same effect. Analysis of 1.3 million AI search citations found translated sites picking up as much as 327 percent more visibility in AI-generated results than their English-only equivalents, which is why this publication argued in June that localisation has stopped being a late-stage marketing task. Human preference and the retrieval layer have started pointing the same way at the same time.
What Malay-First Actually Looks Like
The useful detail is that Bahasa Melayu is not one language but two registers, and the split is visible on any commercial page that has to convert rather than merely inform. Dewan Bahasa dan Pustaka’s sanctioned term for online is dalam talian. What Malaysians actually type into a search box is online. A page of casino reviews written in Malay will carry both, sometimes within a few sentences of each other, because the formal coinage satisfies the institution and the loanword satisfies the reader. Neither one is wrong. Choosing only one of them is.
The same split runs through the vocabulary underneath. Free spins arrives as pusingan percuma in one section and putaran percuma in another, depending on who wrote it. Real money becomes wang sebenar. But software, live dealer, jackpot and e-wallet stay in English, untranslated, because translating them would make the page harder to use rather than easier. Figures are quoted in ringgit, and where they are not, the dollar amounts sit awkwardly next to the ringgit ones and quietly hand the reader a mental conversion to do before deciding anything.
None of that is decoration. It is the difference between a page written for Malaysians and a page written at them, and no translation memory arrives at it unprompted.
Sovereign AI Cut the Price of the Formal Half
Meanwhile the cost side has moved quickly. On 28 July the government launched AI Malaysia Bhd under the Ministry of Digital, replacing the National AI Office that had held the portfolio since December 2024, and paired it with a National AI Action Plan for 2026 to 2030 built around 28 initiatives, fourteen sector-specific and fourteen covering talent, infrastructure, governance and financing. The stated destination is AI Nation status by 2030. A consumer-facing programme, AI Untuk Rakyat, opens on 31 August aiming to reach 100,000 Malaysians aged 18 to 30.
Underneath the policy sits a more consequential piece of infrastructure. ILMU, built by YTL AI Labs with Universiti Malaya and developed further with NVIDIA, reports 87.20 percent on MalayMMLU, the benchmark for Bahasa Melayu comprehension, which the lab says places it ahead of every frontier model on that measure. Astro, Media Prima, Carsome and RytBank are among the businesses already building on it.
For a company selling into Malaysia, the practical effect is that producing formally correct Malay at volume has stopped being a procurement problem. Which is precisely why it has stopped being a moat. MalayMMLU measures the register found in textbooks, government circulars and corporate announcements. It does not measure whether a page knows when to write online instead of dalam talian, and that judgement is now the scarce input rather than the fluency itself.
Which brings the argument back to those 714,000 Malaysians still offline. They are not the opportunity. The opportunity is the 35.4 million already there, most of them reachable in two languages, who have been telling anyone bothering to measure which one they would rather be sold to in.
