In October 2026 the Digital Ministry told Parliament that the Data Centre Task Force had approved 27 projects worth an estimated RM225.97 billion since it was formed in February 2025. Sixteen are in Johor, ten in Selangor and one in Negeri Sembilan. One is operating. The other 26 are under construction.
Johor alone already has 1,110 MW of live capacity, according to Knight Frank figures reported in August 2026, which makes it the third largest market in Asia Pacific after Tokyo and Singapore. Only 0.7% of its co-location space is vacant.
With that much concrete going up, it is reasonable to assume that a Malaysian company using AI has its data processed in Malaysia. Often it does not. A data centre in Johor and an AI model you can call from Malaysia are different things.
Three questions hiding in one
“Where is our data?” is really three questions.
- Where is it stored? The database, the uploaded files, the logs.
- Where is it processed? The place where the AI model reads your prompt and writes its answer. Engineers call this inference.
- Who operates it, on what terms? How long prompts are kept, and whether they are used for training.
A cloud region in Malaysia answers the first question for whatever you put there. It does not automatically answer the second.
The regions that exist
| Provider | Status in Malaysia | Announced investment |
|---|---|---|
| AWS | Asia Pacific (Malaysia) Region, live since 22 August 2024, three availability zones | About US$6.2 billion (RM29.2 billion) through 2038 |
| Microsoft | Malaysia West, in Greater Kuala Lumpur, live since 28 May 2025, three availability zones. A second region in Johor Bahru was announced in November 2025 with no opening date | US$2.2 billion, announced in 2024 |
| First data centre, at Elmina Business Park in Selangor, is operating. On 4 November 2026 Google still described its Malaysian cloud region as “upcoming” | US$15 billion over 2027 and 2028 | |
| Alibaba Cloud | Public cloud region in Johor launched on 9 June 2026, bringing it to five data centres in the country | Not stated |
| Oracle | First Malaysian public cloud region announced in October 2024. We could not confirm that it has opened | More than US$6.5 billion (RM26.8 billion) |
What you can call from Malaysia today
We read the providers’ own documentation in November 2026. It changes often, so treat this as a snapshot and check before you rely on it.
AWS. Amazon Bedrock offers three ways to call a model. In-Region keeps the request in one region. Geographic cross-Region routes it within a group of regions. Global cross-Region, in AWS’s words, “routes anywhere worldwide”.
For the Malaysia region, the table for Claude Sonnet 5 shows no in-Region option and no geographic option. Global is the only one. The other current Claude models we checked showed the same. Singapore, by contrast, is listed with in-Region support for that model.
Amazon’s own Nova Pro is offered from Malaysia through a geographic profile. Its documented destinations from Malaysia are ten regions: Tokyo, Seoul, Osaka, Mumbai, Hyderabad, Singapore, Sydney, Jakarta, Melbourne and Malaysia itself.
Microsoft. Azure has the same three levels under different names. A global deployment “might be processed in any Azure region where the model is deployed”. A data zone deployment for Asia Pacific means “data processed within any Asia Pacific nation”. A regional deployment is processed in the region you chose. Microsoft’s availability page for the models it sells directly, OpenAI’s included, lists Asia Pacific regions such as Australia East, Japan East, Korea Central, Southeast Asia and South India. Malaysia West does not appear on it.
Google. With the cloud region not yet open, nothing on Vertex AI is processed in Malaysia. Its data residency tables list Singapore, Taiwan, Japan, South Korea, India and Australia for Asia Pacific. Requests to the global endpoint may be processed in any Google Cloud location in the world.
The model company directly. Calling a model maker’s own API is usually further away still. Anthropic’s, for example, has two settings for where inference runs: “global”, the default, under which “inference may run in any available geography”, and “us”. For stored data, its documentation says “us” is the only location currently available.
In short: in the documentation we read, none of these providers offered a frontier model that is processed only in Malaysia.
What cross-region inference means
Your application sends a prompt to an endpoint in one region, the source. The provider forwards it to whichever of its destination regions has capacity. The answer comes back the same way.
The provider does this because the chips that run large models are scarce. Pooling them across regions gives customers higher limits and a lower price. AWS says its global option costs about 10% less than the geographic one.
Your stored data does not move, and AWS notes that its logs continue to be recorded in the source region. But the prompt, with whatever customer details are inside it, was read by a machine somewhere else.
Two consequences are easy to miss.
A region restriction can mislead. A team that has limited its cloud account to Malaysia and Singapore may believe nothing leaves those two places. Global routing is designed to go beyond them. AWS documents a specific policy to block it, and advises that organisations with data residency requirements “should assess whether Global cross-Region inference fits their compliance framework”. We have written about a similar gap, where a generated region policy exempted the AI service altogether.
The label is not a guarantee. AWS’s own warning: “Don’t rely on the geographic prefix in an inference profile ID alone to determine where requests can be routed.”
Does it matter?
Sometimes. Sending a customer’s document to a hosted model discloses personal data to a third party, often in another country, and the amended Personal Data Protection Act changed the rules on cross-border transfers. We covered what that means for engineering teams in our PDPA article. Banks and other regulated customers will also ask where the data goes before they sign. For the wider picture, see our map of Malaysia’s AI rules.
Whether processing abroad is acceptable for your data is a legal and contractual question, and you should take advice on it. For many businesses the answer is yes, provided it is known, written into the contract and disclosed. The failure we see is not processing abroad. It is not knowing.
If it has to stay close
- Store in Malaysia. Databases, files and logs can sit in a Malaysian region today on AWS and Azure.
- Send less. Give the model only the fields it needs. Strip names and identity numbers where the task allows.
- Choose the narrowest routing that offers the model you need. One region, then a geography, then global. Narrower usually means higher prices or lower limits.
- Consider a model you run yourself. For data that must not leave, an open model on infrastructure you control is an option to evaluate. Expect it to be less capable and more work to operate.
The same principle applies as with email and other systems: know where it is, and make sure the account is yours.
What to ask a vendor
- Which model do you call, through which provider, at which region’s endpoint?
- Is that call in-region, geographic or global? Show us the setting.
- Where are prompts and answers stored, for how long, and are they used for training?
- Where are our database, files and logs?
- Which fields of personal data are sent to the model?
- What stops a developer switching to global routing next month?
- If the rules or our customers’ requirements change, how quickly can you change region or model?
A vendor who can answer in one page has thought about it. In the systems we build, we tell the client which provider sees which data, and we can keep processing inside a region the client chooses.
How to check for yourself
- AWS. Each model’s page has a “Regional availability” table, and the
GetInferenceProfilecall lists the destination regions. - Azure. The region availability page is organised by deployment type.
- Google. The Vertex AI data residency page lists regions per model.
- Anthropic. Each API response reports the geography where inference ran.
Sources
- 27 data centre projects worth RM225.97bil approved under task force, says Digital Ministry (The Star)
- Johor data centre boom hits capacity wall (New Straits Times)
- Amazon launches AWS Asia Pacific (Malaysia) Region, to invest over RM29 bln (MIDA)
- Microsoft announces its first cloud region in Malaysia (Microsoft)
- Microsoft to expand cloud region in Johor Bahru (Microsoft)
- Google announces US$15 billion Malaysia investment (Google Cloud Press Corner)
- Alibaba Cloud amplifies Malaysia infrastructure with public cloud region in Johor (Alibaba Cloud)
- Oracle plans nearly RM27b investment on cloud facilities in Malaysia (Malay Mail)
- Claude Sonnet 5 model card, regional availability (Amazon Bedrock User Guide)
- Nova Pro model card, regional availability (Amazon Bedrock User Guide)
- Global cross-Region inference (Amazon Bedrock User Guide)
- Supported Regions and models for inference profiles (Amazon Bedrock User Guide)
- Region availability for Foundry Models sold by Azure (Microsoft Learn)
- Data residency, Generative AI on Vertex AI (Google Cloud)
- Data residency (Claude Platform documentation)
This article is general information and is not legal advice.