What Is Sovereign AI? Malaysia's Push, Explained

Sovereign AI means a country, or a company, can build and run AI on its own terms: its own infrastructure, its own data, its own people and its own rules, instead of depending entirely on someone else's. Malaysia has moved fast on this since 2024, with a national AI agency, a homegrown Malay language model, an RM2 billion sovereign AI cloud and a new AI law on the way.

Full disclosure before we start: my company, Kognosys, builds sovereign enterprise AI, so I have a stake in this topic. I've stuck to facts and linked every source so you can check them.

What is sovereign AI?

NVIDIA, which made the term popular, defines sovereign AI as a nation's ability to produce AI using its own infrastructure, data, workforce and business networks.

In plain English, it's about control. Who owns the computers the AI runs on? Where does your data go? Whose laws apply to it? Can you keep going if a foreign supplier changes its prices, its rules or its mind?

It helps to think of sovereign AI as five layers:

• Compute. The data centres and AI chips (GPUs) the models run on.
• Data. Where data is stored, who can reach it, and under which country's laws.
• Models. The AI itself. Some are built locally, some are open models you can run yourself, some are rented from global providers.
• People. Local engineers and researchers who can build, run and fix it.
• Rules. Laws and standards that decide how AI may be used.

Sovereign AI doesn't mean building everything yourself. Almost no country can. It means having real choices at each layer, and keeping the sensitive parts under your own control.

Why sovereign AI matters

There are four big reasons, and they apply to companies as much as countries.

1. Whose laws reach your data. Data stored in Malaysia can still be reachable from abroad. The US CLOUD Act of 2018 lets US authorities compel US tech companies to hand over data they hold, wherever it is stored. Prime Minister Datuk Seri Anwar Ibrahim named exactly this as an open question at the launch of AI Malaysia in July 2026.

2. Language and culture. Global models are trained mostly on English. A model trained on Bahasa Melayu, Manglish and local dialects understands Malaysians better. It also reflects local values and context.

3. Resilience. If one foreign provider controls your AI, a price rise, an outage or a policy change can stop your business overnight. Having options keeps you running.

4. Value stays home. Building AI locally creates jobs, skills and companies here, not just bills paid overseas.

Sovereign AI is a spectrum, not a switch

Most organisations sit somewhere along this range:

Level	What it looks like	Control
Rent	Use a global AI app or API as-is	Low
Local data	Global AI, but data stored in a Malaysian region	Some
Local cloud, open models	Open models run on Malaysian infrastructure	High
Fully owned	Your own models on hardware you control	Highest

Moving up the table costs more and needs more skill. The smart move is usually to match the level to the data. Public marketing copy can use a global tool. Customer records, health data and trade secrets deserve more control. I go deeper on the hardware side in my AI infrastructure guide.

What Malaysia is doing about sovereign AI

Malaysia is working on several layers at once. Here's the timeline at a glance:

When	What happened
Dec 2024	National AI Office (NAIO) set up
Aug 2025	ILMU, Malaysia's first homegrown multimodal AI model, launched
Oct 2025	Budget 2026 sets aside RM2 billion for a Sovereign AI Cloud
Oct 2025	YTL announces a completed AI data centre with NVIDIA in Kulai, Johor
Jul 2026	Public consultation on a national AI Governance Bill
Jul 2026	NAIO becomes AI Malaysia Berhad; National AI Action Plan 2026–2030 launched

Here's what each step means, grouped by the layer it builds.

Rules and leadership: a national AI agency and an AI law

The government set up the National AI Office (NAIO) on 12 December 2024 to coordinate AI work across ministries.

On 28 July 2026, it became a permanent agency, AI Malaysia Berhad, under the Ministry of Digital. The same day, the government launched the National AI Action Plan 2026–2030. It has 28 initiatives: 14 to use AI in key sectors, and 14 to build talent, innovation, infrastructure, governance and financing. The target is for Malaysia to be an "AI Nation" by 2030.

AI Malaysia also houses the Malaysia AI Safety Institute (MY-AISafe), which will test and evaluate AI models.

A new law is on the way too. From 10 to 31 July 2026, the government held a public consultation on an AI Governance Bill. It takes a risk-based approach, meaning stricter rules for riskier uses of AI. It would apply to AI systems used by organisations in Malaysia, no matter where those systems are hosted. Before this, Malaysia only had voluntary ethics guidelines, which I cover in my AI governance starter guide.

Models: ILMU, an AI that speaks Malaysian

On 12 August 2025, the Prime Minister, Datuk Seri Anwar Ibrahim, launched ILMU at the ASEAN AI Malaysia Summit 2025 in Kuala Lumpur. It was built by YTL AI Labs together with Universiti Malaya.

ILMU is Malaysia's first homegrown multimodal model, which means it handles text, voice and images. It understands Bahasa Melayu, Manglish and dialects such as Kelantanese.

YTL says ILMU matches leading global models on Malay-language tests. Those results come from the company itself, not from an independent test, so treat them as a promising start.

Computing power: a sovereign AI cloud and local data centres

Budget 2026, tabled in October 2025, set aside RM2 billion for a Sovereign AI Cloud, to be built by the Malaysian Communications and Multimedia Commission (MCMC). The budget didn't give a timeline or say exactly who will use it, so that's one to watch.

Private companies are adding capacity too. In October 2025, YTL Power announced it had completed an AI data centre with NVIDIA at its 600MW Green Data Center Park in Kulai, Johor, running liquid-cooled NVIDIA GB200 NVL72 (Grace Blackwell) systems. Computing power on Malaysian soil is the foundation every other layer sits on.

People: AI skills for young Malaysians

Sovereign AI also needs people who can use and build it. Alongside AI Malaysia, the government announced AI Untuk Rakyat, a programme that gives up to 100,000 young Malaysians aged 18 to 30 a free three-month subscription to a leading AI app after they complete digital skills courses.

The hard questions Malaysia still has to answer

Sovereignty and foreign investment pull in different directions. AWS, Microsoft, Google and Oracle have committed a combined US$16.9 billion to Malaysian cloud and data centre infrastructure through 2038. That brings jobs and capacity. But data held by US companies may still fall under US law.

At the AI Malaysia launch, Datuk Seri Anwar asked how far Malaysia should go with a sovereign cloud and how to deal with the CLOUD Act. As of the launch, the government hadn't said which agency would own those two questions.

Other open questions:

• Access. Will small businesses, not just government and big companies, be able to use the sovereign cloud at a fair price?
• Openness. Will local models be open for others to build on, or locked to one provider?
• Skills. Malaysia needs many more engineers who can run AI safely, not just use it.

What sovereign AI means for Malaysian businesses

You don't need to wait for the government to act on this. A few practical steps:

1. Know where your data lives. List the AI tools your staff use, and where each one stores and processes data.
2. Sort data by sensitivity. Keep customer, health, financial and confidential data on infrastructure you control, or with providers you trust under Malaysian law.
3. Ask vendors hard questions. Which country's laws apply? Who holds the encryption keys? Can they be forced to hand over your data?
4. Avoid lock-in. Build so you can swap AI models later. My enterprise AI roadmap explains how.
5. Consider local or vertical models. A smaller model trained on your industry or language can beat a giant general one. See vertical vs horizontal LLMs.
6. Watch the AI Governance Bill. It will likely set new duties for organisations that deploy AI in Malaysia, including incident reporting.

Conclusion

Sovereign AI is about keeping control: of your data, your infrastructure and your future choices. Malaysia has made real moves since 2024, with ILMU, the RM2 billion sovereign AI cloud, AI Malaysia Berhad and a coming AI law. The biggest questions, like how to stay open to global investment while protecting data from foreign laws, are still being worked out. For businesses, the lesson is simple: decide which data must stay under your control, and build your AI so it can.

Quick answers

What is sovereign AI in simple terms?

It's the ability to build and run AI on your own terms, using your own infrastructure, data, people and rules, so you aren't fully dependent on a foreign provider.

Does Malaysia have its own AI model?

Yes. ILMU, built by YTL AI Labs with Universiti Malaya, launched in August 2025. It understands Bahasa Melayu, Manglish and dialects like Kelantanese, in text, voice and images.

What is AI Malaysia Berhad?

It's Malaysia's national AI agency under the Ministry of Digital, launched on 28 July 2026. It replaced the National AI Office and coordinates the National AI Action Plan 2026–2030.

Is data stored in Malaysia automatically sovereign?

No. If a foreign company holds it, that company's home laws may still reach it. The US CLOUD Act is the best-known example. Control depends on who operates the system and who holds the keys, not just on where the server sits.