NVIDIA is working with eight Australian cloud and data-centre providers on AI infrastructure that it says could support a buildout of up to 2GW by 2027. The important bit is local access: the planned capacity is intended to let Australian companies, researchers and public bodies build and run AI without sending every demanding workload offshore.
NVIDIA Watch analysis
What happened
The company announced the collaboration on 10 September with Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk. Those providers will operate the facilities, while NVIDIA supplies its DSX platform, accelerated computing, networking, software and technical support.
DSX spans facility design, compute, networking and software, with infrastructure intended to accommodate multiple generations of NVIDIA hardware. NVIDIA also wants customers to use the resulting capacity with its Nemotron open models for locally developed models, applications and agents.
Why it matters
AI capability increasingly depends on rather unglamorous things: land, electricity, cooling and network links. Bringing more of that stack onshore could reduce latency, support data-residency requirements and give Australian teams better access to the compute needed for training and inference.
The 2GW figure is still a buildout ambition assembled across several providers, not a single finished NVIDIA facility. Delivery will depend on projects securing power, construction and customers over the next year, which is where the glossy diagram meets the electrical substation.
Our read
This is a substantial infrastructure play, not another ceremonial partnership selfie. If the announced projects arrive on schedule, Australia gains a more credible domestic AI-compute market and NVIDIA gains a wider runway for its full stack. Prospective customers should follow the individual providers, because availability, location, residency guarantees and pricing will matter far more than the aggregate headline.
What to watch
- Which projects contribute to the 2GW target and when their capacity becomes operational.
- Whether customers receive competitive access or hyperscalers absorb most of the supply.
- How much new electricity generation is secured for the buildout.
- Whether Australian organisations train significant local models or mainly use the capacity for inference.
Discussion spark: Would more domestic AI capacity materially change what Australian organisations can build, or will access and electricity costs remain the real bottlenecks?
Sources and evidence
- NVIDIA Expands AI Infrastructure Capacity in Partnership With Australia’s Data Center Ecosystem (10 September 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.