NVIDIA Australia AI Infrastructure Plan Targets Up to 2 GW by 2027
NVIDIA and eight Australian infrastructure partners are targeting up to 2 GW of AI-factory capacity by 2027, although project-level power and permitting details remain undisclosed.
Published · 3 min read
NVIDIA Australia AI infrastructure partnerships are targeting up to 2 GW of data center buildout by 2027, linking the chipmaker’s DSX computing platform with land, power and facilities being developed by eight Australian cloud and data center companies.
In its September 9 announcement, NVIDIA named Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk as participating partners. Those companies will operate the AI facilities, while NVIDIA will supply its DSX platform, accelerated computing, networking, software and ecosystem support.
The 2 GW figure is a target across a multi-company portfolio rather than evidence that 2 GW of capacity is operating or fully permitted. NVIDIA did not provide a site-by-site allocation, construction budget, energization schedule or list of committed power-supply projects.
How the NVIDIA Australia AI infrastructure plan is structured
NVIDIA describes DSX as a full-stack reference platform for what it calls AI factories. The partners are expected to develop the land, utility power, building shells and operating infrastructure needed to host successive generations of NVIDIA computing systems.
Firmus is expanding Project Southgate for Australian and Asia-Pacific demand. Sharon AI is developing AI factories for enterprise and government workloads, while IREN plans facilities built around renewable-powered data center operations. CDC, NEXTDC and AirTrunk add established Australian data center portfolios; Megaport supplies connectivity, and ResetData is developing liquid-cooled sovereign AI infrastructure.
The announcement groups projects at different stages under one national target. It does not say every partner will deliver the same amount of capacity or that all projects have secured grid connections. That makes the portfolio materially different from a single-site development such as the planned 1 GW HyperVault campus in Hyderabad, where one developer has secured a defined 264-acre site but has not disclosed its detailed power plan.
The 2 GW target would be large relative to today’s market
Reuters reported on September 10 that Australia’s current computing capacity is about 1.6 GW, citing Data Centres Australia research from DC Byte. On that comparison, an additional 2 GW would exceed the country’s existing base, although the two figures may not use identical definitions or represent simultaneous electrical demand.
Data Center Dynamics independently confirmed the partner list and the 2027 target. Its coverage of the Australian buildout also describes the initiative as an expansion of land, power and shell capacity designed to host NVIDIA systems.
The scale is comparable with some of the largest proposed computing loads now entering grid-planning processes. Microgrid Media recently reported that five Galaxy projects totaling about 4.2 GW received different conditional classifications in Texas. The Australian program has no equivalent public classification in the announcement, so its 2 GW target should not be treated as approved grid demand.
Power supply remains the central unresolved issue
NVIDIA said the expansion would support additional power-generation projects, but it did not identify their technologies, owners, capacities or locations. The company also did not disclose how much of the portfolio would use existing utility service, new grid-connected generation, batteries or behind-the-meter resources.
Those details will determine the initiative’s effect on Australian electricity systems. A 2 GW computing portfolio requires more than annual energy procurement: data centers also need high-quality electricity, network capacity, cooling and redundancy at each operating site.
Microgrid Media’s guide to data center microgrids and battery storage explains why large AI campuses are increasingly evaluating several power architectures when utility interconnections cannot keep pace with computing construction.
Execution will be measured project by project
The NVIDIA Australia AI infrastructure plan is significant because it connects a national computing target with a named group of facility, cloud and network operators. Its practical power impact, however, will emerge only as individual projects obtain sites, contracts, permits, grid connections and generation.
The next credible milestones are partner-specific capacity announcements, utility or regulatory approvals and evidence that construction and energization remain on schedule for 2027. Until those records appear, the strongest supported description is a targeted multi-partner buildout of up to 2 GW—not 2 GW already under construction or online.
