by Tan Aik Keong (AK)
When people talk about sovereign AI, a lot of minds jump straight to foreign big tech's models. In reality, Malaysia already has three genuinely representative local large language models taking root: Mesolitica's MaLLaM, YTL AI Labs' ILMU, and Agmo's Merdeka LLM. What they share is a focus on Malaysian data, local deployment and local teams. What differs is that each has picked a completely different lane — making them complementary rather than competing head-on.
MaLLaM: language and dialect first
The first is MaLLaM, from Mesolitica, a local startup dedicated to language models. MaLLaM was never meant to become a general English-language model — it was trained from the start as a brain specifically for the Malaysian linguistic environment, with training data drawn overwhelmingly from Malaysian context, at a scale approaching nine billion tokens.
According to multiple media reports and AWS, MaLLaM can understand local Bahasa Malaysia, Manglish, Bahasa Rojak, and a range of state dialects and regional languages — sixteen regional languages and local accents in total — built specifically so AI genuinely understands how locals actually speak.
What's more interesting is that Mesolitica isn't just building a text model — it's partnered with RTM to bring its own TTS and voice cloning into broadcasting, demonstrating real-time voice conversion and synthetic anchor voices at RTM events, and signing agreements to explore local AI voice technology within the public broadcasting system.
Mesolitica's distinctive angle is pushing on both language and voice at once, aiming to train the model on accents and phrasing from Johor, Kedah, East Malaysia and beyond, so the AI's sense of language feels as close as possible to real conversation at the kopitiam, not just standard written text.
ILMU: national AI infrastructure
The second is ILMU, built by YTL AI Labs. The name comes from Intelek Luhur Malaysia Untukmu, and it's a multimodal large model developed and operated locally under YTL AI Labs, able to handle text, voice and images — officially built entirely by a Malaysian team, trained on local data and corpora, and kept fully under local control.
ILMU has quickly landed in two very grounded use cases. The first is Ryt Bank, the fully AI-driven digital bank launched jointly by YTL Group and Sea. Multiple reports confirm that Ryt Bank's core conversational assistant, Ryt AI, is built on ILMU, handling natural-language transfers, balance checks and financial literacy for users.
The second is in-car entertainment and voice assistance — YTL AI Labs demonstrated an in-car AI assistant built with ACOTech at a launch event, using ILMU to power voice interaction for navigation, entertainment control and vehicle status updates.
In terms of positioning, ILMU looks more like AI infrastructure: the local large model itself on one side, and the compute, networking and partner ecosystem behind it on the other — telecoms, media, banking, automotive and other industries all connecting to the same underlying local model.
Merdeka LLM: an industry-specialist factory
The third is Merdeka LLM, from Agmo Group. Per the official site and announcements, Merdeka LLM is positioned as AI Sovereignty as a Service — developed by a local team, hosted in Malaysian data centres, trained on local data, with the focus on letting government and enterprise keep control of their data, hosting and model ownership locally.
Unlike the other two, Merdeka LLM was never meant to become a general-purpose chatbot — the goal is helping different industries train their own specialist expert models. The official materials name legal, HR and finance as focus areas, emphasising fine-tuning specialised models — a Malaysia Legal LLM, a Malaysia HR LLM — using Malaysian legal, HR and industry-specific data, for tasks like contract review, compliance checks and policy document organisation.
On education, Agmo has formed a partnership with education publisher Sasbadi. Company announcements and multiple media reports confirm the two are jointly developing Malaysia's first large language model focused specifically on education, with Sasbadi contributing local curriculum content and education expertise, and Agmo handling model design, training and deployment — built around the national curriculum, to give students and teachers localised, bilingual, personalised learning support.
Agmo's annual report and related coverage describe Merdeka LLM as a Malaysian education-focused LLM built on local content aligned with the national curriculum — an important part of the group's ESG and education-equity positioning.
Seen this way, Merdeka LLM's real distinguishing feature isn't whether it can write flowery essays — it's that in highly sensitive domains, it can ingest local Malaysian data that can't simply be sent to an overseas cloud, and train, within a local compliance environment, a genuinely specialist model that actually understands local regulation, institutions and teaching requirements. That kind of data is generally off-limits to global general-purpose models like ChatGPT or DeepSeek — and that's the core value of sovereign AI.
Three complementary lanes, not three competitors
Put the three side by side, and it's clear they're not doing the same job. Mesolitica works more like a language and voice lab, focused on training every state's accent, Bahasa Rojak, local slang and multiple regional languages into the model — laying groundwork for any application that needs a deep understanding of local language.
YTL's ILMU works more like an AI utility company — running local models and compute infrastructure tied to national sovereignty on one side, while plugging that model into banking, in-car systems and media on the other, so the public genuinely uses a local large model through their banking app or car dashboard.
Agmo's Merdeka LLM is a factory purpose-built for vertical specialists — working with experts in law, HR, education and finance to turn industry-specific knowledge and data into individual domain models under a sovereignty-first approach, with the Sasbadi education partnership as a clear example.
Sovereign AI, in the end, isn't shaping up as three companies walling each other off — it's growing into an ecosystem with a clear division of labour: a brain that understands local language and dialect, a base layer providing national-scale infrastructure, and specialist models embedded deep in specific industries, handling sensitive data. On that front, MaLLaM, ILMU and Merdeka LLM represent three mutually reinforcing lanes, not three competing fronts.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Nov 22, 2025.