The pattern has repeated across the industry for two years: an American AI lab with enormous computing needs, and a data-center developer willing to build fast where power is cheap. On Tuesday the Australian developer Firmus announced the latest version of that arrangement, saying OpenAI has signed a multiyear strategic agreement to buy dedicated AI computing capacity from two of its facilities in Malaysia.
Firmus said OpenAI will serve as the anchor customer for the Malaysian sites, and that the deal pushes the developer’s total contracted capacity past 900 megawatts. The agreement is the kind of commitment that data-center builders live on, giving Firmus the revenue certainty to finance construction while handing OpenAI a fixed supply of computing power in a region where new capacity is scarce.
The two facilities are part of a wider portfolio. Firmus operates seven AI factories across Australia, Singapore, Indonesia and Malaysia, with two already running and five under construction. The company said the remaining sites will come into service over the next 24 months, a schedule that reflects both the demand it sees and the constraints of building power-hungry buildings in crowded electricity markets.
The technical specification of the Malaysian sites tells the story of the industry’s current hardware cycle. The factories will deploy Nvidia’s Vera Rubin NVL72 systems, the rack-scale machines that pair the company’s newest accelerator generation with its networking and cooling components, running on Nvidia’s DSX platform for AI factory deployment. Cooling comes from HyperCube, Firmus’s own liquid-cooling system, an acknowledgment that air conditioning can no longer handle the density of modern accelerator clusters.
The investor list behind Firmus reads like a map of the AI economy’s financial backers. Nvidia itself holds a stake, alongside Jane Street, funds managed by Blackstone and the investment firm Coatue. Firmus said its valuation after its most recent funding round exceeds $10.5 billion, and people familiar with the company’s plans say an initial public offering could come this year, a listing that would rank among the largest Australia has seen.
The Malaysia agreement fits a pattern that has defined the AI infrastructure buildout: compute capacity is flowing to places that can deliver power and speed rather than to the countries where the demand originates. Southeast Asia has become a favored destination, with Malaysia in particular drawing billions in data-center investment on the strength of its electricity supply, its land and its proximity to the undersea cables that carry traffic between Asia and the rest of the world.
OpenAI framed the arrangement in terms of demand rather than geopolitics. The company said the Malaysian data centers will help it meet growing needs in the Asia-Pacific region and around the world, language that emphasizes the practical reasons for the deal. But the choice of location carries implications European policymakers have begun to notice: while governments debate sovereign AI strategies, the actual computing capacity is being built where construction is fastest and power cheapest.
The economics of the arrangement are straightforward for OpenAI. The company has committed to enormous computing capacity across multiple continents as it races to train and run frontier models, and contracts like the one with Firmus let it lock in supply without owning the buildings. For Firmus, the anchor contract validates the 900-megawatt portfolio and gives lenders confidence that the remaining five factories will find tenants.
There is a competitive dimension as well. The global supply of AI-ready data centers remains far smaller than the demand from American labs, cloud providers and national programs, and developers with contracted capacity have become sought-after partners. Firmus’s lineup of investors, its operational sites and now its anchor agreement with OpenAI put it among a small group of builders that the largest AI companies treat as essential suppliers.
The agreement also illustrates how the industry’s procurement has changed. In the early days of the AI boom, labs rented computing capacity by the hour from cloud providers, treating GPUs as a commodity they could buy as needed. The scale of frontier-model development ended that flexibility: companies now sign contracts measured in years and megawatts, committing to capacity before the buildings that house it exist. Arrangements like the one with Firmus give developers the financial foundation to build, and they give OpenAI the assurance that the machines will be there when its next model needs to train.
Firmus, for its part, has built its business around the Nvidia vision of the AI factory, a facility designed from the ground up for accelerator computing rather than a repurposed enterprise data center. The company’s Malaysian sites are part of a wave of such projects across Southeast Asia, where developers have found both land and electricity in quantities that are increasingly difficult to secure in the regions where the largest AI companies are based.
The regulatory backdrop adds a layer of complexity. Malaysia has welcomed the investment wave while working through questions about power pricing, water use and the environmental cost of data centers, and the government has signaled it wants the country to capture more of the value chain than raw hosting. For Firmus and OpenAI, those are questions for later; Tuesday’s announcement is about the simpler matter of securing capacity while it is still available, in a market where the queue for power grows longer every quarter.
