OpenAI is reserving artificial intelligence computing power in Malaysia before the machines are even switched on. Firmus, an AI infrastructure firm that partners with Nvidia, said on 8 September that OpenAI has signed a multi-year deal to become the anchor customer of two AI data centres it plans to build in the country.
Before the celebration runs ahead of the facts, it is worth being precise about what the deal actually confirms. As TechRepublic reported, OpenAI's allocated capacity, the contract value, the exact sites and the service dates are all undisclosed. The one large figure attached to the announcement, more than 900 megawatts of contracted capacity, covers Firmus's entire customer base, not OpenAI alone. Neither Malaysian site is operating yet, and their exact locations, capacity and timeline have not been released.
What is not in doubt is the scale of the wave OpenAI is joining. The Malaysian Investment Development Authority says data centre and cloud projects drew RM95.8 billion in approved investment in the first half of 2026, about 44 per cent of everything the country approved in that period. Malaysia offers cheaper energy and available land, Inside AI News reported, which is part of why the build-out is spreading here.
Power, water and approvals are the tests a new project now has to pass. Johor alone has a data centre pipeline of 8,542 megawatts, according to Knight Frank, even as vacancy in its existing colocation space sits at just 0.7 per cent. Malaysia's Data Centre Task Force now weighs whether a developer has secured power and water before it approves a project, the same power and water strain we covered when an AI company began naming data-centre backlash as a business risk.
There is a quieter tension underneath the announcement. OpenAI's undisclosed allocation will be dedicated compute, set to run on Nvidia silicon in facilities that Firmus is developing. It arrives in the same fortnight that Malaysia is weighing whether to build its RM2 billion Sovereign AI Cloud on Huawei chips, a supplier choice reported by Bloomberg but not yet officially confirmed, in an initiative the reporting says is meant to give the country more control over national data. Housing a hyperscaler's compute and running your own models are not the same thing, and the distance between them is the whole point of the minister's push for Malaysia to build AI rather than only consume it.
The hardware on the spec sheet is not an afterthought, for the record. Firmus plans to run Nvidia's Vera Rubin NVL72 rack systems, the company's next-generation AI platform, through the DSX AI Factory platform, using its own liquid-cooling design.
So the honest read is this. Firmus has secured a major anchor customer for two planned Malaysian sites, neither of them operating yet. The numbers that will tell us what this deal is really worth, how many megawatts OpenAI has taken, where the two sites will stand, and when they switch on, are exactly the ones Firmus has not released. Those are the disclosures to watch, and until they arrive the announcement is a reservation, not a result.
Image(s) courtesy of Taylor Vick and Tyler on Unsplash.
