The announcement listed eight partners before it got to the number that mattered. Nvidia said on September 9 that it is working with Australian cloud and data-center operators including Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC, and AirTrunk to expand land, power, and building capacity, with a target of up to 2 gigawatts of AI factories online by 2027. The platform underneath is Nvidia’s DSX full-stack offering, and the customers range from startups to universities to enterprises.
The Australian push is a country-level version of a strategy Nvidia is running everywhere: pair its chips with local partners who build the buildings and buy the power, while Nvidia supplies the silicon and the software stack. Heidi and Atlassian are among the companies that will use Nvidia’s open Nemotron models to build local models, an arrangement that keeps data onshore and gives Nvidia a beachhead it does not have to build alone.
Sharon AI’s commitment is the largest on the list, with plans to deploy up to 68,000 GPUs. IREN is planning 800 megawatts at its Bundey campus in South Australia. The individual numbers vary, but together they describe a coordinated buildout sized to the gigawatt, a scale that would have been unthinkable for Australian data centers a few years ago.
The day after the Australia announcement, Nvidia extended the same logic inward. The company said it is working with Palantir to bring what it calls sovereign AI into the critical supply chain, and the first customer is Nvidia itself. Jensen Huang, the chief executive, put the scale in his own terms: a single Vera Rubin rack contains 1.3 million components, and the two companies will use Nvidia’s Nemotron models and Palantir’s Ontology platform to connect the chain from wafer to token.
The phrasing matters. Nvidia is not selling this to a customer first; it is running it on its own supply chain, then holding up the result as a reference. The company that sells picks and shovels to the AI industry is now using its own products to run the factory that makes the shovels, a demonstration that doubles as marketing.
Sovereign AI has become Nvidia’s answer to a question it keeps being asked: what happens when countries want their own AI infrastructure rather than renting compute from American clouds. The pitch is that local models, local data, and local data centers can be assembled from Nvidia’s parts, and Australia is the case study in progress.
The economic logic for Nvidia is layered. Hardware revenue comes from the GPUs. Platform revenue comes from DSX and the software that runs on top. And now, with Palantir, a third layer: the supply-chain intelligence that connects procurement to production. Each layer is a way to make the same customer relationship worth more.
Australia is a deliberate choice. The country has abundant renewable energy, strong data-sovereignty instincts, and a government that has shown willingness to invest in domestic AI capability. For a company trying to prove the sovereign AI model works outside the United States, Australia is a friendly proving ground with room to build at the gigawatt scale.
The Palantir partnership extends Nvidia’s reach into a market where it has not historically played: the industrial software that runs factories. By wiring its own supply chain with Palantir’s ontology, Nvidia is positioning the pair to sell the same system to every other company with a complex physical supply chain, which is most of the Fortune 500.
Analysts said the Australia buildout and the Palantir tie-up share a theme: Nvidia is moving beyond selling chips into selling outcomes, from infrastructure to the intelligence that manages it. The risk is execution. Gigawatt-scale data centers take years and billions to build, and the partners Nvidia has lined up vary widely in their track records.
The 2-gigawatt target is an ambition with a deadline. Whether it is met depends on power availability, construction timelines, and the appetite of Australian enterprises to actually buy the compute once it is built. Nvidia’s role is to make sure that when the capacity arrives, its chips and software are already the default choice inside it.
Australia’s appeal for the sovereign AI pitch includes its data-localization culture and its surplus of renewable power, both of which align with Nvidia’s argument that AI infrastructure can be built outside the American cloud giants. The partnerships spread the capital risk across local operators while Nvidia keeps the technology layer.
Huang’s wafer-to-token framing is the clearest statement of the strategy. Nvidia wants to own the relationship from the moment raw silicon enters the fab to the moment a model produces an answer. The Australia announcement and the Palantir deal are two steps down that same path, one building the factories and the other running the supply chain that feeds them.
