On September 9, NVIDIA announced a partnership with eight Australian infrastructure operators to deploy AI factories with a total capacity of up to two gigawatts by 2027. This is significantly more than Australia's current capacity, which is about 1.6 gigawatts according to Data Centres Australia and research company DC Byte (citing Reuters). The project involves Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC, and AirTrunk. NVIDIA will supply computing accelerators, network equipment, software, and technical support, while the Australian partners will develop and manage the infrastructure facilities themselves. The target capacity of two gigawatts is a collective figure, not a single NVIDIA-owned data center.
Sharon AI plans to deploy up to 68,000 NVIDIA graphics processors using DSX infrastructure, connected via NVIDIA Quantum InfiniBand and Spectrum-X Ethernet network solutions. The company is focused on serving Australian startups, enterprises, government agencies, and scientific organizations in need of locally hosted computing power. IREN, which has historically worked with Bitcoin mining infrastructure, is now shifting to AI cloud services and high-performance computing. The company will develop the Bundey campus in South Australia, designed for a capacity of 800 megawatts. CDC already manages more than 550 megawatts of capacity in Australia and New Zealand, with another 800 megawatts under construction. The company also claims that its facilities use renewable electricity and water-free cooling systems; these environmental claims are company statements and have not been independently verified at all proposed sites.
NVIDIA describes DSX as a full-featured AI factory platform covering facility design, computing, network components, software, and reference architecture. The system is compatible with CUDA and is designed to support future generations of NVIDIA hardware without the need to redesign each level of infrastructure. The capacity expansion is expected to provide Australian organizations with better access to open NVIDIA Nemotron models. The company Heidi applies Nemotron tools in clinical applications, while Atlassian uses NVIDIA technology for semantic search and other functions in the Rovo AI platform.
However, NVIDIA has not disclosed the expected cost of the project, individual investment commitments, or the volume of capacities already funded, permitted, and connected to the network. The success of reaching the target capacity will largely depend on the availability of electricity, network connectivity, and cooling capabilities for high-performance GPU clusters. NVIDIA also warned that its infrastructure estimates depend on demand, partner execution, regulation, and technology availability. A general timeline of milestones up to 2027 has not been published. The next measurable steps will be site approvals, energy agreements, construction phases, and GPU delivery schedules.