• Nvidia and eight Australian partners are preparing up to 2GW of land, power and data-centre space for AI systems by 2027
  • The target covers infrastructure needed to host AI equipment, not 2GW of GPUs already installed or available to customers

The fact

Nvidia said on 9 September that it is working with Australian cloud and infrastructure providers to prepare up to 2GW of land, power and data-centre space for AI systems by 2027. Eight partners are named in the programme: Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk. They will operate the facilities, while Nvidia provides its DSX platform, including computing, networking, software and data-centre reference designs.

Nvidia said the infrastructure is being designed to support several generations of its AI systems. The 2GW target refers to land, power and building shells prepared for those systems. It does not mean that 2GW of fully equipped computing capacity has already been installed or made available to customers.

Some partners have provided more detail about their own projects. Sharon AI says it is deploying up to 68,000 Nvidia GPUs, while IREN plans to use the DSX approach at its proposed 800MW Bundey campus in South Australia. Other partners are expanding AI-ready data-centre, cloud and network infrastructure at different stages of development.

The assessment

Preparing land, electricity, and data-centre buildings removes some of the work needed before AI systems can be installed, but it does not produce usable computing capacity on its own. A site still needs cooling, networking, racks, GPUs and commissioning before customers can run workloads there.

That distinction matters because the eight partners are not all at the same stage. Sharon AI has given a GPU target, while IREN has linked the programme to its planned Bundey campus. Buyers comparing providers will need to look at each project separately and ask what hardware will be installed, where it will be located and when it will actually be available.

For BTW readers, the 2GW target points to a much larger pipeline of Australian infrastructure capable of hosting AI systems. How much of that becomes usable computing capacity by 2027 will depend on the fit-out and equipment installed at individual sites. Named projects, installed GPU numbers and customer-ready dates will provide a clearer picture than the headline 2GW figure alone.

What to watch

Watch for individual partners to publish site locations, power capacity, construction milestones and GPU deployment schedules. Customer-facing offers that specify the hardware available and when workloads can begin running will show how quickly the planned 2GW of underlying infrastructure is becoming usable AI computing capacity.