OpenAI has signed a multi-year agreement with Australian infrastructure company Firmus to secure dedicated computing capacity from two data centres planned for Malaysia, strengthening the ChatGPT developer’s access to the hardware needed to operate advanced artificial-intelligence models.
The agreement makes OpenAI an anchor customer for Firmus and marks the infrastructure provider’s entry into Malaysia. Neither company has disclosed the contract’s financial value or the amount of computing capacity reserved specifically for OpenAI.
Firmus said the partnership has lifted its total contracted capacity across all customers to more than 900 megawatts. That figure covers the company’s wider customer portfolio and should not be interpreted as the capacity purchased by OpenAI alone.
The deal reflects the growing competition among leading AI companies to secure processors, electricity, cooling systems and data-centre space. Training large models requires substantial computing power, while serving responses to millions of users creates a continuous demand for infrastructure after those models have been released.
Firmus Expands Its Asia-Pacific Network
Firmus currently has seven planned or operational “AI factories” across Australia, Singapore, Indonesia and Malaysia.
Two facilities, located in Australia and Singapore, are operating. Five additional sites are under development and are expected to become ready for service over the next 24 months, according to the company.
The two Malaysian facilities covered by the OpenAI agreement remain part of this development programme. Firmus has not announced their precise locations, individual power capacities or expected commissioning dates.
OpenAI Vice President of Compute Strategy Sachin Katti said the Malaysian data centres would help the company meet growing demand for its products across Asia-Pacific and other international markets.
Locating computing infrastructure closer to users can reduce delays when requests travel between devices and data centres. It may also provide additional capacity outside the United States, where competition for suitable sites, power connections and specialised equipment has increased sharply.
Malaysia has become one of Southeast Asia’s fastest-growing data-centre markets. Its proximity to Singapore, lower operating costs and available industrial land have attracted investments from cloud providers and infrastructure companies.
However, this rapid expansion is placing additional pressure on electricity and water supplies. Data centres require large amounts of power for computing equipment and cooling, leading Malaysian authorities to apply closer scrutiny to new projects and their resource-efficiency plans.
Nvidia Vera Rubin Systems Will Power the Facilities
Firmus said the Malaysian sites will use Nvidia’s next-generation Vera Rubin accelerated-computing platform.
The planned equipment includes Vera Rubin NVL72 systems, which combine processors, networking and rack-scale infrastructure designed for demanding AI workloads. Firmus intends to deploy the technology across its broader Asia-Pacific network rather than limiting it to the two Malaysian facilities.
The data centres will also use the Nvidia DSX AI Factory Platform and Firmus’ HyperCube infrastructure. HyperCube combines liquid cooling, electrical systems and mechanical equipment within prefabricated modules produced in regional New South Wales, Australia.
Firmus says this standardised approach can reduce the time required to construct new facilities and lower the cost of generating AI output. These remain company claims, and actual performance will depend on factors including energy prices, hardware utilisation, cooling efficiency and the complexity of the workloads being processed.
The use of liquid cooling has become increasingly important as newer AI processors consume more electricity and generate more heat. Traditional air-cooling systems can struggle with high-density server racks, while liquid systems transfer heat more directly from the hardware.
The OpenAI agreement further strengthens Firmus’ relationship with Nvidia, which is both a technology partner and an investor in the Australian company.
In June, Firmus announced a wider agreement involving the planned deployment of 170,000 Nvidia graphics processors between the first quarter of 2027 and early 2028. Those processors are expected to be installed at facilities in Batam, Indonesia, as part of Firmus’ regional expansion.
Firmus Builds Capacity Before Potential IPO
Firmus raised $2 billion in an equity funding round announced on August 7. The transaction placed its post-money valuation above $10.5 billion, nearly twice the valuation attached to an earlier financing completed four months previously.
The round included follow-on investment from Nvidia and Coatue Management. Funds managed by Blackstone and global trading company Jane Street also participated.
Firmus said it would use the capital to accelerate the development of AI factories across Australia and Asia-Pacific. The company has raised more than $3 billion in new equity during the past year as it moves to secure sites and equipment before expected customer demand becomes operational.
Reports have indicated that Firmus may pursue an initial public offering in Australia during 2026. The company has not formally confirmed an IPO timetable, issue size, price range or target valuation, meaning any listing remains prospective.
The agreement also illustrates a wider change in the AI industry. Competition is no longer limited to producing the most capable model. Companies must also secure enough computing capacity to deliver their products reliably and at a commercially sustainable cost.
For Firmus, the next task is turning more than 900 megawatts of contracted capacity into operational infrastructure while managing construction, hardware supply and energy requirements. For OpenAI, the Malaysian facilities add another source of computing power as usage of ChatGPT and its business products expands.
The partnership gives both companies a larger presence in Asia-Pacific, but its commercial impact will depend on when the two Malaysian sites begin operating and how efficiently they can deliver AI computing at scale.
