Nvidia Taps 8 Australian Partners for 2GW AI Data Centre Capacity
▼ Summary
– Nvidia has partnered with eight Australian data centre firms to build AI factory capacity targeting two gigawatts of compute by 2027.
– The initiative defines an AI factory as a full-stack solution centered on Nvidia’s DSX platform and CUDA ecosystem rather than just physical infrastructure.
– The project emphasizes sovereign AI benefits like local data residency while relying entirely on American technology stacks and private funding without government subsidies.
– Unlike the EU’s state-aided approach, this model relies on private operators to assemble capacity quickly, though it raises questions about power supply independence.
– The arrangement highlights the commercial strategy of converting independent landlords into deployments of a single company’s architecture.
Nvidia has secured agreements with eight Australian entities to develop what it terms AI factory capacity, a move that could deliver up to two gigawatts of artificial intelligence computing power by 2027. The partners involved in this strategic expansion include Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NextDC, and AirTrunk. Notable among them is IREN’s Bundey campus in South Australia, an existing site with an 800MW footprint, while CDC reports operating over 550MW across Australia and New Zealand with another 800MW currently under construction.
The term AI factory represents more than just physical infrastructure; it denotes a full-stack solution encompassing facilities, compute resources, networking, software, and reference designs centered on Nvidia’s DSX platform and CUDA ecosystem. This commercial definition effectively transforms independent Australian data center landlords into nodes within a unified corporate architecture. While the two-gigawatt figure is substantial,roughly equivalent to the output of two large nuclear reactors,the announcement does not specify the source of electricity. This omission is standard for such deals but remains a critical gap given the grid operator’s recent warnings regarding firm capacity in eastern states.
The push for local infrastructure aligns with the concept of Sovereign AI, driven by data residency requirements and industries seeking to keep sensitive information offshore. However, this sovereignty is partial. As noted by industry observers, while the racks are located in Australia, the accelerators, interconnects, software stacks, and reference architectures remain American products. This distinction carries significant weight in Canberra, particularly as Washington exerts pressure on Australia regarding platform regulation. Unlike the European Union’s approach, which relies on state aid and consortium bids, Australia’s model is being assembled through private operators without public subsidy, offering a faster deployment mechanism but raising earlier questions about ownership and energy supply.
Industry leaders have expressed strong support for the initiative. Oliver Curtis of Firmus emphasized building Australia’s AI future, while James Manning of Sharon AI highlighted the country’s energy, connectivity, and ambition. Daniel Roberts of IREN provided a pragmatic perspective, stating: “building AI infrastructure at scale requires integrating every layer, which is precisely the argument for buying the whole stack from one supplier and precisely the concern.”
Despite the enthusiasm, the current landscape reveals a market where capacity is being committed ahead of confirmed demand. Six of the eight partners are landlords or connectivity providers rather than model developers, and the only customer-side participant mentioned is a healthcare startup. The announcement lacks specific timetables, capital commitments, named customers, or details on power generation. These missing elements determine whether the two-gigawatt target is a concrete plan or merely a headline, leaving the actual execution of the project open to interpretation.
(Source: The Next Web)