Key takeaways

  • NVIDIA says it is working with eight Australian infrastructure partners on up to a 2 GW AI-factory buildout by 2027.
  • The announcement covers a portfolio, not one 2 GW campus. It does not allocate the target among partners or say that 2 GW is operating, permitted or under construction.
  • The few disclosed figures measure different things: Sharon AI cites a GPU count, IREN names an 800 MW campus, and CDC describes its wider Australia-New Zealand portfolio. They should not be added together as confirmed capacity for this program.
  • NVIDIA supplies its DSX platform, accelerated computing, networking and software; the participating providers will operate the AI factories.
  • Site-level power sources, budgets, permitting status and delivery schedules remain undisclosed in the announcement.

What NVIDIA announced

On September 9, 2026, NVIDIA announced a collaboration with Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk. The company described the program as an expansion of the land, power and building-shell capacity intended to host multiple generations of NVIDIA DSX AI factories, with up to a 2 GW buildout by 2027.

That wording matters. NVIDIA announced a forward-looking portfolio target. It did not announce a single 2 GW data center, and the release does not say 2 GW of computing load is online today. It also does not divide the target among the eight partners.

NVIDIA says the participating providers will operate the AI factories. NVIDIA's role is to deliver the DSX platform, accelerated computing, networking, software and ecosystem support.

What the 2 GW figure does—and does not—measure

The announcement links the 2 GW target to expanded land, power and shell capacity, but it does not define a common measurement boundary for every project. For example, it does not state whether each partner's contribution is measured as utility service, total campus electrical capacity or critical IT load.

The release also mixes several kinds of scale indicators:

  • Sharon AI plans to deploy up to 68,000 NVIDIA GPUs.
  • IREN says it is applying the DSX blueprint to its 800 MW Bundey campus in South Australia.
  • CDC says it operates more than 550 MW across Australia and New Zealand and has another 800 MW under construction.

Those figures are useful context, but they are not interchangeable. A GPU count is not a power figure, a campus rating is not proof of energized IT load, and CDC's portfolio numbers are not presented as an allocation to NVIDIA's 2 GW target. The release provides no partner-by-partner reconciliation to the headline total.

What each partner brings

PartnerRole described by NVIDIADisclosed scale
FirmusExpanding Project Southgate for hyperscale and AI-native demand in Australia and Asia-PacificNo megawatt allocation disclosed
Sharon AIDeploying DSX-based GPU infrastructure with Quantum InfiniBand and Spectrum-X EthernetUp to 68,000 NVIDIA GPUs
IRENCombining its power, land, data-center and operating platform with the DSX reference architecture800 MW Bundey campus named; no program allocation disclosed
MegaportExpanding access to accelerated computing through Latitude.sh and Megaport's networkNo megawatt allocation disclosed
ResetDataProviding Australian-resident accelerated computing for enterprise, government and research usersNo megawatt allocation disclosed
CDCProviding high-density facilities with direct liquid-to-chip coolingMore than 550 MW operating across Australia and New Zealand; another 800 MW under construction across its portfolio
NEXTDCDeveloping liquid-cooled, high-density infrastructure, including DSX-based AI factoriesNo megawatt allocation disclosed
AirTrunkExpanding AI-ready powered shells with direct-to-chip liquid coolingNo megawatt allocation disclosed

The table shows why the headline should be read as a program ambition rather than a construction ledger. Some partners name existing portfolios, some describe a specific campus or GPU plan, and others describe capabilities without a capacity figure.

Where DSX fits

NVIDIA describes DSX as a full-stack AI-factory platform spanning facilities infrastructure, computing, networking, software and reference designs. It is compatible with the CUDA ecosystem and is meant to support hardware and software upgrades over the life of the facility.

That makes DSX the common architecture across the announcement, but it does not make the eight partner projects identical. Each provider still controls the physical delivery problem: securing a site, electricity, cooling, connectivity and operations for its own deployment.

What remains unanswered

NVIDIA says the expansion will support additional power-generation projects, but the release does not name their owners, technologies, locations or capacities. It also does not provide:

  • a site-by-site share of the 2 GW target;
  • a common definition of the capacity included in that total;
  • construction budgets or financing terms;
  • permitting or grid-connection status for each site; or
  • individual completion and energization dates.

Those omissions do not negate the announcement. They define what evidence readers should look for next: partner-level capacity disclosures, utility or regulatory records, construction milestones and commissioning dates.

Bottom line

NVIDIA's Australia announcement brings eight named operators under a DSX-based infrastructure program targeting up to 2 GW by 2027. The disclosed partner assets show that the group has real data-center, networking and operating capabilities. But the release does not establish that 2 GW is already financed, permitted, being built or energized. Until site-level records appear, the accurate description is a multi-partner target—not 2 GW of completed Australian AI capacity.

Sources

This article was researched and fact-checked against the following sources: