OpenAI has signed a multi-year strategic partnership with Firmus Technologies, announced September 8: OpenAI becomes an anchor customer contracting dedicated AI compute from two Firmus AI factory sites in Malaysia. The deal lifts Firmus's total contracted capacity across all customers past 900 megawatts and puts OpenAI's weight behind Nvidia-based infrastructure - Vera Rubin-class processors on Firmus's HyperCube-integrated DSX AI Factory platform - in a market where Chinese open weights are the default alternative.
The megawatts are the headline and the least interesting part. AI factory deals have become the genre's baseline: a partner with land and power, a frontier lab as anchor tenant, a number with a MW in it. What distinguishes the Firmus arrangement is who is supplying the stack: Nvidia-based infrastructure end to end - Vera Rubin-class processors running on Nvidia's DSX AI Factory platform, integrated with Firmus's own HyperCube system. The anchor customer, the landlord, and the chip vendor are all pulling in the same direction, which is another way of saying Nvidia keeps winning every layer of these deals.
For Firmus the deal is a coming-out: the Nvidia-backed operator building out Project Southgate across Tasmania and mainland Australia, now claiming more than 900 megawatts contracted across all customers, with sites in Malaysia and an ASX listing reportedly in the works. OpenAI's anchor tenancy converts that pipeline from speculative capacity into contracted demand.
For OpenAI the placement is strategic. Malaysia sits inside the region where US export rules, Chinese open-weight competition and national sovereignty programs are colliding in real time. Contracting capacity there buys OpenAI a flag on the ground in the exact market where DeepSeek-class open weights are the default alternative.
The open question is the usual one: power. Southeast Asian grids have been under strain from data center growth for two years, and multi-hundred-megawatt campuses are promises measured in years of transmission build-out, not press releases. The deal is real; the electrons have a schedule of their own.
The Template Deal
Expect the pattern to repeat. Anywhere three conditions meet - cheap power, sovereignty pressure, and an open-weight alternative already in the room - the Firmus structure is now the offer on the table: anchor customer, Nvidia-supplied stack, sovereign flag. For OpenAI it converts capex diplomacy into distribution across a region of 680 million people. For the region it is another step in the answer to the compute-localization question - though the electrons still have their own schedule.
The Takeaways
- OpenAI signed a multi-year partnership with Firmus: anchor-customer compute from two Malaysian AI factory sites, lifting Firmus past 900 MW contracted across all customers.
- The stack is Nvidia end to end - Vera Rubin-class processors on the DSX AI Factory platform, integrated with Firmus's HyperCube system.
- Placement is strategic: the exact market where DeepSeek-class open weights are the default alternative - and where grid constraints set the real schedule.
