|
Getting your Trinity Audio player ready...
|
Nvidia-Backed Firmus Signs OpenAI Deal for Malaysia AI Data Centers, Passing 900MW Capacity
OpenAI has locked in a significant new source of computing power in Southeast Asia, signing a multi-year agreement with Australian AI infrastructure company Firmus Technologies to secure dedicated capacity from two data centers in Malaysia. The deal makes OpenAI an anchor customer for the Malaysian sites, a designation that carries real weight in the infrastructure business, since it means OpenAI has committed to buying a meaningful share of the facilities’ future capacity well before construction is even finished, giving Firmus the kind of predictable, long-term demand that makes large-scale data center financing possible in the first place.
The agreement pushes Firmus’ total contracted capacity across all its customers past 900 megawatts, a substantial jump for a company that currently operates just two fully functioning AI data centers, located in Australia and Singapore. Firmus has five additional facilities under development across the Asia-Pacific region, and the Malaysian sites represent the company’s latest expansion into a market that has become one of the most closely watched data center growth stories in the world over the past two years.
Malaysia’s rise as a regional AI infrastructure hub did not happen by accident. The country has become Southeast Asia’s fastest-growing data center market, benefiting from proximity to Singapore, comparatively available land, and a regulatory environment that has actively courted large-scale technology investment. Johor, the Malaysian state bordering Singapore, has emerged as a particular hotspot for hyperscale facilities, as companies look for space and power capacity that Singapore itself can no longer easily accommodate given its physical size and existing infrastructure constraints. That regional dynamic has turned Malaysia into something of a release valve for the broader Southeast Asian AI buildout, absorbing demand that would otherwise have nowhere to go.
That growth has not come without friction. The rapid pace of data center construction in Malaysia has drawn increasing scrutiny over electricity and water consumption, concerns that mirror debates playing out in data center hubs worldwide, from Virginia to Ireland to parts of the Middle East. Large AI training and inference facilities consume enormous amounts of power for both computing and cooling, and water-intensive cooling systems in particular have become a flashpoint in communities near major data center developments. Malaysia’s government has so far continued approving new projects, but the tension between attracting AI infrastructure investment and managing its resource footprint is likely to remain a recurring theme as the country’s data center sector keeps expanding.
On the technology side, Firmus plans to deploy Nvidia’s next-generation Vera Rubin processors at scale across its Asia-Pacific facilities, with the Malaysian sites extending that rollout into a new market. Nvidia’s backing of Firmus goes beyond a simple customer relationship. Firmus was valued at more than $10.5 billion in its most recent fundraising round, with Nvidia listed alongside investors including Jane Street, Blackstone funds, and Coatue Management. That kind of investor lineup signals substantial confidence in Firmus’ growth trajectory, and the OpenAI deal arrives just ahead of a rumored initial public offering that could rank among the largest in Australian history this year, adding a financial market dimension to what is otherwise primarily an infrastructure story.
For OpenAI, the Firmus agreement fits into a broader pattern of diversifying where its computing power comes from. Microsoft Azure has long served as OpenAI’s primary compute provider, but the company has been actively expanding relationships with additional infrastructure partners to avoid being overly dependent on any single supplier, a strategy that reduces bottleneck risk as demand for training and running increasingly capable AI models continues to climb. That demand pressure is not abstract. OpenAI unveiled Astra last week, describing it as the company’s most capable model to date, and each new generation of frontier AI models tends to require substantially more computing capacity than the one before it, both for the initial training process and for running inference once the model is deployed to users at scale.
The Firmus deal also reflects a wider geopolitical and competitive dynamic playing out across the AI infrastructure landscape. Both American and Chinese AI companies are racing to secure power and data center capacity wherever they can find it, and Southeast Asia has become one of the more contested regions in that scramble, given its relatively favorable energy costs, government incentives, and geographic proximity to both established Asian tech hubs and emerging markets. Malaysia’s willingness to accommodate large foreign AI infrastructure investment, including a separate $2.36 billion agreement between Nvidia and Malaysian utility company YTL Power announced earlier this year to build AI data centers and support a sovereign Malaysian language model, suggests the country is positioning itself as a long-term player in this competition rather than a one-off beneficiary of overflow demand from Singapore.
Neither Firmus nor OpenAI has disclosed the financial terms of their agreement. Firmus declined to comment on contract value when asked, and OpenAI did not respond to requests for comment on the deal’s specifics. That kind of financial opacity is fairly typical in large-scale infrastructure agreements of this nature, where public disclosure often waits until a company’s IPO filing or a subsequent funding round forces greater transparency around contracted revenue.
What is clear is the trajectory. Firmus has gone from operating two data centers to holding contracted capacity above 900 megawatts in a relatively short window, and OpenAI’s decision to anchor two new Malaysian facilities suggests the company sees continued value in building out a geographically diversified compute footprint rather than concentrating its infrastructure needs in any single country or provider. As AI model capabilities continue to scale and the compute requirements behind them grow correspondingly larger, deals like this one are likely to keep multiplying across Southeast Asia, reinforcing the region’s emergence as a genuine center of gravity in the global AI infrastructure buildout rather than a peripheral market chasing overflow demand.
