Malaysia Lands OpenAI as Its AI Factory Anchor Tenant
2 min readMalaysia just picked up one of the most valuable tenants in AI infrastructure. Firmus, the Nvidia-backed data center builder headquartered in Sydney, announced on September 8 that OpenAI has signed a multi-year deal for dedicated compute capacity at two AI factory sites in Malaysia, becoming the company’s anchor customer.
Who Firmus is
Firmus builds what the industry now calls AI factories: purpose-built facilities designed around dense GPU clusters rather than general-purpose servers. Its portfolio spans seven sites across four countries, Australia, Singapore, Indonesia and Malaysia, with two already operational in Australia and Singapore and five more under construction targeting service over the next two years.
The company’s differentiator is a prefabricated unit it calls HyperCube, which packages liquid cooling, mechanical systems and electrical distribution into a single deployable block. That matters in Southeast Asia, where heat and humidity make conventional air cooling expensive to run.
What the deal covers
The Malaysian sites will run Nvidia’s Vera Rubin generation processors on Nvidia’s DSX AI Factory platform, integrated with Firmus HyperCube infrastructure. The OpenAI contract pushes Firmus past 900 MW of total contracted capacity across all customers, a threshold that puts it in the same conversation as far older and better funded operators.
Financial terms were not disclosed. Firmus described the arrangement as a multi-year strategic partnership and framed the Malaysian expansion as its first move beyond Australia, Singapore and Indonesia.
Why it matters for Asia
Singapore has been the default landing point for AI compute in Southeast Asia, but its moratorium-era power constraints have pushed new capacity outward. Malaysia, and Johor in particular, has absorbed much of that spillover. This deal moves the country from overflow destination to a primary AI hosting address in its own right, with a frontier lab as the named tenant.
The wider lesson is about the binding constraint. Chips are no longer the only bottleneck on where compute gets built. Power availability, land and cooling economics increasingly decide the map, and that is a contest Malaysia can win. Watch whether other frontier labs follow OpenAI south, and whether Malaysian grid planning keeps pace with what is now being contracted.