Walk into any AI lab, creative studio, or marketing team in Malaysia today, and you’ll find the same set of tools quietly humming in the background: OpenAI, Google, Meta, Microsoft, maybe a sprinkle of Stability AI or Anthropic. All foreign. All hosted on global clouds. All brilliant — but none truly ours.
This begs the question:
Can Malaysia build its own AI stack? Or are we destined to stay on the digital sidelines, renting power from someone else’s server farm forever?
What Is an AI Stack, Really?
Before we dive into the deep end, let’s break it down. When we talk about an AI “stack”, we’re talking about everything that makes artificial intelligence run:
- Data infrastructure: storage, governance, processing power
- Model layer: like large language models (LLMs), computer vision systems, or audio transformers
- Deployment layer: APIs, applications, UI/UX where AI shows up in real life
- Regulatory and ethical frameworks
- Talent and tooling to maintain the above
Building the whole stack means owning not just the application — like a chatbot or facial recognition app — but also the foundational models, compute pipelines, and decisions about how data is used, labelled, stored, and protected.
Right now, Malaysia mostly operates at the application layer. We take what’s already built — often by US or Chinese companies — and customise the last mile.
But is that enough?
Where Malaysia Stands Now
Malaysia has made solid progress in setting national AI goals.
- In 2021, the government launched the Malaysia Artificial Intelligence Roadmap (AI-Rmap), aiming for “AI-ready talent, ecosystem and governance” by 2030.
- MIMOS (the national R&D center) has its own local LLM initiative — still under wraps.
- PETRONAS, Maybank, and Telekom Malaysia are all exploring or deploying AI use cases in operations and service.
- Budget 2024 earmarked RM100 million+ for automation and smart tech adoption, including AI.
But the big stuff — training local LLMs, hosting massive compute clusters, or offering an open-source Malaysian-made foundation model — remains out of reach for now.
Here’s how it could look:
- An ASEAN Open Model
Instead of building solo, we lead a regional initiative — an ASEAN LLM, trained on Southeast Asian languages, values, and accents. - AI Infrastructure Push
Just like we built Cyberjaya or MSC Malaysia, we could create a national AI compute cloud, powered by GLCs and green data centers. - Talent Incubators
Fund universities and startups to train engineers at the model level, not just app level. The goal isn’t more chatbot builders — it’s to grow researchers who understand how the black box thinks.
Why It Matters
Without control over our AI stack, we’re forever adapting to others’ models, values, and constraints. We risk:
- Cultural loss, when our languages and contexts are left out of training data
- Economic leakage, when every AI service is paid to a foreign provider
- Security gaps, when foundational tech isn’t locally governed
And perhaps most importantly — we miss out on the chance to define what Southeast Asian intelligence looks like in a digital age.
