Firmus Makes OpenAI Anchor Customer for Two Malaysia AI Factories
Sydney-based Firmus announced in September 2026 a multi-year strategic partnership under which OpenAI will contract dedicated artificial-intelligence compute from two Firmus AI Factory sites in Malaysia, making OpenAI an anchor customer as Firmus’s total contracted capacity across customers rises above 900 megawatts. The Malaysia sites extend Firmus’s Asia-Pacific footprint beyond Australia, Singapore and Indonesia, with five of seven planned factories still under development and targeting ready-for-service within about twenty-four months. Firmus said the factories will use NVIDIA DSX AI Factory Platform with Vera Rubin NVL72 rack-scale systems integrated into its HyperCube liquid-cooling and electrification stack prefabricated in regional New South Wales.
Filed under Infrastructure and dated October 2, 2026, this AI4Malaysia briefing treats the OpenAI anchor deal as Malaysian AI-factory commercial news distinct from yesterday’s PETRONAS myPROdata agentic upgrade and from Firmus’s earlier Meta compute agreements. OpenAI compute strategy vice president Sachin Katti said the Malaysian centres will help serve growing regional demand; NVIDIA’s Nico Caprez cast HyperCube-based factories as a path to faster capacity and stronger token economics, while exact megawatt allocations, contract values and go-live dates for the two Malaysian campuses remain undisclosed.
Why it matters: Malaysia is racing to host frontier compute without becoming a pure landlord. Anchor hyperscaler contracts can bankroll factories—but only if power, cooling and human commissioning authority travel with every contracted megawatt.
What it means in practice
Malaysian infra, energy and counsel leads should inventory which domestic AI workloads could buy capacity beside OpenAI’s reserved halls; demand named owners for state grid and water commitments to the two campuses; assign an owner for NVIDIA Vera Rubin supply and HyperCube localisation claims; run time-boxed reviews of data-residency and export-control exposure; and prefer contracts that keep humans on energisation gates. Place the partnership beside Firmus’s Meta AI compute deals and Malaysia’s National AI Compute Exchange plan.
Caveats come first. Anchor status is not a published site address; 900 MW is contracted capacity across countries; and Vera Rubin timelines can slip. AI4Malaysia therefore presents the deal as directional infrastructure context until published Malaysian campus milestones appear.
What to watch next: disclosed Malaysian locations and first energisation dates; OpenAI allocated megawatts; and how Putrajaya ties factory anchors to the National AI Compute Exchange. Readers can continue on the AI4Malaysia homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Malaysian AI factories are pairing OpenAI demand with Firmus–NVIDIA stacks and remain early. Organizations that benefit most will demand site evidence, keep humans on commissioning gates, and refuse to confuse an anchor announcement with finished local capacity.