Australia becomes a target for gigawatt-scale AI capacity
NVIDIA has announced partnerships with eight Australian cloud and infrastructure providers aimed at supporting up to 2 gigawatts of AI data-center capacity by 2027. The participating companies are Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk.
The facilities are intended to host multiple generations of NVIDIA DSX systems. NVIDIA will provide compute, networking, software and reference architecture, while its partners develop and operate the underlying land, power, cooling and data-center capacity.
The headline number is a plan, not delivered capacity
The 2-gigawatt figure describes a potential buildout across an ecosystem of projects. It should not be read as 2 gigawatts of installed NVIDIA hardware already available to customers. Construction timelines, power connections, financing, permits and equipment deliveries will determine how much capacity comes online by 2027.
Several partners disclosed more specific components. Sharon AI says it is deploying up to 68,000 NVIDIA GPUs, while IREN points to its 800-megawatt Bundey campus in South Australia. Those figures still describe plans and portfolios whose utilization will develop over time.
DSX links facilities to the software stack
NVIDIA presents DSX as a full-stack AI factory platform spanning facility design, computing, networking and software. The goal is to make sites compatible with successive hardware generations and a range of training and inference workloads.
That standardization can shorten deployment planning, but buyers should examine the full operating model: accelerator availability, networking topology, storage, scheduler efficiency, power usage, cooling, geographic latency and the contractual terms attached to capacity.
Sovereign compute is part of the pitch
The announcement emphasizes local access for Australian startups, universities, researchers, government and enterprises. ResetData specifically highlights Australian data residency, while NVIDIA says its Nemotron open models can support locally relevant applications and agents.
Data residency and domestic facilities can improve control, yet sovereignty depends on more than location. Hardware supply, software dependencies, model licensing, operational ownership and the ability to move workloads all influence how independently an organization can run its systems.
Power and cooling remain the binding constraints
Gigawatt-scale AI infrastructure requires major generation, transmission and cooling capacity. Providers named in the announcement describe liquid cooling, high-density facilities and renewable electricity, but aggregate environmental and grid impacts will depend on when projects connect and how continuously they operate.
Policymakers and customers should ask for project-level disclosure of energy sources, water use, grid agreements, construction milestones and effective compute delivered. Capacity measured in power is not the same as useful model throughput.
What to watch through 2027
The most important milestones will be energized capacity, installed systems, customer availability and measured utilization—not additional headline commitments. Local pricing and network access will determine whether the buildout materially broadens compute access or mainly serves a small number of large tenants.
Readers can follow the AINewsInu homepage and our Industry coverage for verified updates on Australian facilities, NVIDIA DSX deployments and the global race to finance AI infrastructure.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.