AI Infrastructure / AI news for Malaysia
OpenAI contracts AI compute from two Firmus sites in Malaysia
Australian infrastructure company Firmus says OpenAI has become an anchor customer and will take dedicated capacity from two Malaysian AI factory sites. Neither site is named, and neither is built yet.
In brief
- Firmus Technologies said on 8 September 2026 that OpenAI will contract dedicated AI compute capacity from two of its AI Factory sites in Malaysia, becoming an anchor customer.[1]
- The company says the deal pushes its contracted capacity across all customers past 900 MW, with seven AI factories now spanning Australia, Singapore, Indonesia and Malaysia.[1]
- Only two of those seven sites operate today, both outside Malaysia, and the release gives no location, capacity, value or completion date for the Malaysian pair.[1]
OpenAI signed a multi-year deal for capacity at two Firmus sites in Malaysia
OpenAI has agreed to buy dedicated AI computing capacity from two data centre sites in Malaysia. Firmus Technologies, the Sydney-headquartered infrastructure company behind them, announced the multi-year partnership on 8 September 2026 and called OpenAI an anchor customer for its next phase of building across the region.[1]
The announcement is short on the details a Malaysian reader would want first. It confirms two AI Factory sites in Malaysia and that OpenAI will take capacity from them. It names no state, no megawatt figure, no ringgit value and no opening date.[1]
An anchor customer, and a number that covers everything
The single hard figure in the release is a portfolio total. Firmus says that with OpenAI added, its contracted capacity across all customers now stands at more than 900 MW. That is the whole book, not the Malaysian share, and it is not broken down by country or site.[1]
Sachin Katti, OpenAI's vice president of compute strategy, said the new data centres in Malaysia will help serve growing demand for the company's products across the region and around the world. That places Malaysia inside a global network. It promises no local routing, data residency or Malaysian product launch.[1]
What is actually going into the buildings
Firmus says the AI factories built under the partnership will use NVIDIA's DSX AI Factory Platform to combine Vera Rubin NVL72 rack-scale systems with its own HyperCube platform. HyperCube bundles liquid cooling, mechanical systems and electrification into a repeatable unit prefabricated in regional New South Wales.[1]
Vera Rubin NVL72 is NVIDIA's newest rack-scale inference platform. In its own 24 August 2026 post, NVIDIA claims up to 30 times higher throughput per megawatt than the GB300 NVL72 on agentic workloads, and 35 times lower token costs. Those are vendor figures from a vendor benchmark.[2]
Seven AI factories, two of them switched on
The portfolio behind the headline is younger than the headline suggests. Firmus counts seven AI factories across Australia, Singapore, Indonesia and Malaysia. Two are operational today, in Australia and Singapore. The other five are under development, targeting ready-for-service over the next 24 months rather than on dated schedules.[1]
Malaysia is described as an extension of that footprint, not an existing operation. Co-founder and co-chief executive Tim Rosenfield framed the deal as the moment Asia-Pacific becomes a producer of intelligence rather than only a consumer of it. The two Malaysian sites still have to be delivered.[1]
How Malaysia should read it
For anyone tracking Malaysian data centre investment, this is a demand signal rather than a capital announcement. No investment quantum, incentive, state authority or utility agreement appears in the release, and no Malaysian ministry is quoted. Nothing tells a local supplier which corridor to plan around.[1]
The practical question for Malaysian businesses is narrower than the headline. Compute sited in Malaysia does not by itself change what a local company pays for OpenAI services, where its data is stored or which regulator supervises it. Assume nothing about pricing or residency changed today.[1]
Why Malaysia should care
Malaysia has been named in a compute contract signed by the most visible AI company in the world, and that is the whole of what is confirmed. The release does not say which state the sites sit in, how many megawatts they carry, what the contract is worth, or when they open.
Malaysian SMEs buying AI tools
Capacity located in Malaysia is not a Malaysian service, price or data-residency guarantee, and none were announced.[1]
Practical move: Ask your vendor in writing where your data is processed before assuming your contract changed.
Contractors, suppliers and local councils
Two sites are committed on paper, but no state, timeline or utility partner is published, so there is nothing to bid into yet.[1]
Practical move: Track Firmus and its named partners directly rather than planning around an unidentified corridor.
Policymakers and energy planners
A 900 MW contracted book across four countries says demand is locked in faster than the sites serving it are built.[1]
Practical move: Ask for the Malaysian megawatt figure and the grid connection basis before treating this as settled load.
What Malaysians can do now
- Treat the two Malaysian sites as announced and not yet operating: only two of the seven Firmus AI factories are live, both outside Malaysia.
- Do not repeat 900 MW as a Malaysian number. It is the total contracted across every customer and country.
- Attribute the 30x throughput and 35x token cost claims to NVIDIA's own August benchmark, not to Malaysia.
What we still do not know
A signed contract is not a running data centre
- Where the two Malaysian AI Factory sites are, down to state and district.
- How many of the 900 MW contracted megawatts sit in Malaysia, and how much is committed to OpenAI.
- What the multi-year agreement is worth and how many years it runs.
- When either Malaysian site reaches ready-for-service, and whether any Malaysian ministry, agency or utility has approved or supported it.
Sources
- 1.Firmus surpasses 900 MW contracted capacity, adds OpenAI as anchor customer and expands into Malaysia Firmus Technologies, 8 September 2026
- 2.Up to 30x More Work Per Watt: NVIDIA Vera Rubin NVL72 Sets a New Efficiency Standard for AI Agents NVIDIA, 24 August 2026
- 3.Press Kit Firmus Technologies
More AI News
View all
