NVIDIA says it is working with a group of Australian cloud and infrastructure providers to expand the country’s AI-factory footprint. The September 9 announcement points to up to 2 gigawatts of planned capacity by 2027, combining land, power, data-center shells, cooling, networking, accelerated computing, and software around NVIDIA’s DSX platform.
The scarce input is becoming an integrated site
The release frames an AI factory as more than a room full of GPUs. Operators need a site with enough electricity, high-density cooling, connectivity, mechanical systems, and a supply chain that can keep the installation productive across multiple generations of compute. NVIDIA says its Australian partners will operate the facilities, while NVIDIA provides the DSX reference platform, accelerated systems, networking, software, and ecosystem support.
The partner list spans different layers of the buildout. Sharon AI says it is deploying up to 68,000 NVIDIA GPUs, IREN brings power, land, data-center development, and GPU operations, Megaport adds inference capacity and networking through Latitude.sh, and ResetData is targeting customers that require Australian data residency. CDC, NEXTDC, and AirTrunk describe liquid-cooled or AI-ready facilities across Australia and the wider region.
Local compute, regional ambition
NVIDIA’s stated goal is not only to host global workloads. The company says the expanded capacity should give Australian startups, universities, researchers, enterprises, developers, and AI-native companies greater access to accelerated computing and Nemotron open models. It also points to local use cases from healthcare company Heidi and enterprise-software company Atlassian.
That makes the announcement partly an adoption story. More regional capacity can reduce latency, support local data-residency requirements, and let Australian organizations customize models and agents closer to their operations. It also connects infrastructure investment to a broader policy and sovereignty question: which countries can train and deploy useful systems without relying entirely on distant clouds?
What the headline number does—and does not—say
The up-to-2-gigawatt figure is a future buildout target, not a report that 2 gigawatts of AI compute is already operating. The release includes partner plans and forward-looking statements, so the practical test will be how quickly power, permits, cooling, and hardware move from announced capacity to commissioned capacity. The difference matters because AI-factory economics depend on utilization and delivered power, not just a reservation of land or a powered shell.
Even with that caveat, the announcement is a useful signal about where the infrastructure race is moving. The competitive advantage is shifting from buying an accelerator to integrating the entire stack around it. Australia’s buildout shows how energy, data residency, network access, open models, and enterprise demand are being assembled into a regional AI platform.