Australia's entire data center fleet draws about 1.6 gigawatts today. Nvidia has now lined up eight local operators to add as much as 2 gigawatts of AI factory capacity by the end of 2027 — a buildout that, if delivered, would more than double the nation's computing load in roughly fifteen months.
The company named the partners on September 9 in Santa Clara, September 10 in Sydney: Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk. Each will own and operate its own facilities. Nvidia supplies the DSX platform — accelerated computing, Quantum InfiniBand and Spectrum-X Ethernet networking, reference designs and software — and the brand. The 1.6GW baseline comes from lobby group Data Centres Australia, citing analysts DC Byte, and was confirmed by Reuters.
The scale is concentrated in a handful of campuses. IREN's Bundey site in South Australia is an 800-megawatt development, among the largest AI infrastructure projects on the continent. CDC already runs more than 550 megawatts across Australia and New Zealand with a further 800 megawatts under construction. Firmus is expanding Project Southgate, targeting up to 1.6GW nationally by 2028, backed by a $10 billion debt facility from Blackstone-led funds with Coatue and a deal covering roughly 18,400 Nvidia GB300 GPUs. Sharon AI is deploying up to 68,000 Nvidia GPUs on DSX; its June compute collaboration with Nvidia carries a contract value of up to $4.88 billion, and it expects more than 55,000 GPUs installed by mid-2027.
The rest fill in the stack. Megaport is routing GPU capacity through its subsidiary Latitude.sh for inference close to end users. ResetData, the group's only immersion-cooling operator, serves customers needing Australian data residency. NEXTDC and AirTrunk provide liquid-cooled halls and AI-ready powered shells.
Nvidia's framing was explicitly economic. “AI factories turn energy into intelligence — the essential resource of the AI economy,” said Raj Mirpuri, the company's vice president of global AI clouds and infrastructure ecosystem, describing DSX as making such facilities “more productive, fungible and durable — and a new investable asset class.” No Jensen Huang quote accompanied the release, and no aggregate investment figure was disclosed.
The operators pitched sovereignty. “Australia has the energy, connectivity and ambition to lead in AI,” said James Manning, cofounder and chief executive of Sharon AI. “That potential only becomes real capability when organizations can access world-class compute without compromising on security or sovereignty.” IREN co-chief executive Daniel Roberts was more operational: “Combining IREN's vertically integrated platform with the NVIDIA DSX AI factory reference architecture provides a repeatable blueprint to bring capacity online at scale.”
Two launch customers were named: Heidi, a clinical voice-AI company post-training on Nvidia's Nemotron 3 Ultra open models, and Atlassian, which is using Nvidia infrastructure for fine-tuning and reinforcement learning behind Rovo's semantic search.
The playbook: name a country, attach a number
Very little here is a new transaction. Sharon AI's Nvidia deal dates to June. Bundey was already under development. CDC's 550 and 800 megawatts predate the announcement. What is new is the aggregation — Nvidia declaring Australia a single coordinated program with one headline capacity figure attached.
That is the same move Nvidia made in Korea, where it named NAVER, SK Telecom and Brookfield under a 200-megawatt target for 2028, and in Japan, where it backed a 27,500-GPU national AI factory. Gartner expects worldwide sovereign cloud infrastructure spending to reach $80 billion in 2026, up 35.6 percent, with Asia-Pacific growing 87 percent. Australia is now a named entry in that race, which is the point: a program-level label pulls policymakers, lenders and enterprise buyers toward a pipeline the operators are financing themselves while Nvidia captures the silicon demand. It is demand seeding with almost no balance-sheet exposure.
The sovereignty argument underneath it is real. The US CLOUD Act lets American authorities compel disclosure from US-incorporated providers regardless of where servers sit, which is why data residency on a hyperscaler's Sydney region is not legal sovereignty.
The constraint is electricity, not chips
Nvidia cannot ship power. The Australian Energy Market Operator's Electricity Statement of Opportunities, published August 25, recorded 66.9 gigawatts of prospective data center grid connection requests across the National Electricity Market as of June — against roughly 2.2 gigawatts of rated connection capacity nationally. More than 40 percent of data center projects since 2025 have dropped out or regressed. Connection queues for large industrial loads routinely run two to five years, meaning an application filed today would not realistically energize before 2028.
AEMO projects data center consumption rising from about 5 terawatt-hours a year in 2025-26 to roughly 34 terawatt-hours by 2035-36, a sevenfold increase; Transgrid puts New South Wales and ACT demand at 3.5 gigawatts by 2035. Policy is tightening in parallel. Prime Minister Anthony Albanese announced in July that large-scale AI data centers would be required to operate as net generators, returning at least as much power to the grid as they draw. National Cabinet endorsed nationally consistent mandatory standards on energy, water and land use on August 26, with legislation targeted for early 2027.
This splits the coalition cleanly. CDC runs on 100 percent renewable electricity with zero-water cooling and is already connected. ResetData secured 72 megawatts of dedicated on-site generation through a Centuria arrangement in August, bypassing the queue entirely. Bundey sits in South Australia, outside the eastern seaboard bottleneck. Firmus, AirTrunk, NEXTDC, Megaport and Sharon AI have disclosed no equivalent dedicated generation. For them, the load-bearing words in the announcement are the qualifier: up to 2 gigawatts.
Three things to watch. Whether Nvidia or its partners publish a split between permitted, financed capacity and aspirational pipeline — that gap is where the 2027 date lives or dies. Whether the net-generator legislation grandfathers projects already in construction, because the mandate and the interconnection queue arriving together would be punishing. And Sharon AI's next disclosure, the only listed pure-play in the group and the first hard read on whether Australian sovereign compute is being contracted as fast as it is being built.
“AI factories turn energy into intelligence, the essential resource of the AI economy.”— Raj Mirpuri, VP of global AI clouds and infrastructure ecosystem, Nvidia