Nvidia’s 2GW Australia Push Isn’t a Power Business — It’s a Claim on Other People’s Megawatts

Nvidia announced this week that it is working with eight Australian operators to build out up to 2 gigawatts of AI data centre computing capacity by 2027 — enough, on the company’s own framing, to more than double the country’s current computing load. Read quickly, it sounds like the next chapter in a familiar story: Nvidia, running out of datacentres to sell into, is now moving into the power business itself.

It is not. Decompose the 2GW figure and the round number dissolves into something with a different owner and a different meaning. A gigawatt is a measure of electricity, not revenue, and the electricity here is not Nvidia’s. The eight named partners — Firmus, Sharon AI, IREN, ResetData, Megaport, CDC, NEXTDC and AirTrunk — are the ones investing in the land, the power and the building shells; Sharon AI alone says it will deploy up to 68,000 Nvidia GPUs. Nvidia supplies the DSX full-stack platform, the chips, the networking and the software, then steps back while its partners carry the capital cost.

Nvidia’s own balance sheet confirms the arrangement. Over the trailing twelve months the company’s capital expenditure was roughly $7.4 billion against about $127 billion of free cash flow — not the spending profile of a firm building gigawatt-scale power plants, which analysts now price at $50 billion to $60 billion each. The 2GW sits on other people’s balance sheets, not Nvidia’s.

The scarce input is the watt, not the chip

So if the power isn’t Nvidia’s, how does this become Nvidia revenue? The answer is the structural reason the deal matters at all. Electricity has become the binding constraint on AI compute: U.S. datacentres already draw about 4.4% of the country’s electricity, a share the tracker projects will roughly double by 2028, and Nvidia points to more than 200 GW of projects sitting in interconnection queues. Silicon is abundant; watts are not.

That constraint is precisely what Nvidia’s DSX platform is engineered to monetize. The company has reframed its pitch from “how many GPUs fit in a building” to “how much useful AI work can be pulled from each fixed megawatt” — more tokens per megawatt, which is to say more revenue per gigawatt. Its MaxLPS power-tuning layer combines 45°C liquid cooling with software that redistributes power across racks, and Nvidia claims it lets operators fit up to 40% more GPUs into the same power envelope. If the scarce input is power, then the vendor that makes each watt more productive captures value from every build that uses it.

The recurring cut underneath the headline

The financial mechanism that turns this efficiency story into a durable earnings line arrived in July, when Nvidia launched a two-part model for its AI factory partnerships. Beyond its standard hardware margin, Nvidia now takes a recurring, usage-linked share of the cloud revenue its partners generate on the supported capacity. The terms — the percentage, how usage is measured — have not been disclosed.

This is the real content of the claim that “Nvidia’s next growth leg is power.” Nvidia is not selling electricity. It is selling the productivity of constrained electricity and then collecting a toll on the compute that power enables, while its partners carry the fixed capital. It is an asset-light claim on everyone else’s megawatts.

What the 2GW is, and isn’t

The Australia deal is best read as a forward demand pipeline. Power and land secured today become DSX orders spread across 2026 and 2027 — tangible visibility into what the market can expect Nvidia to sell. But the word that matters in the announcement is “up to.” Two gigawatts by 2027 is an ambition, not a commitment, and power buildouts are prone to slippage: grid-connection queues run long, and deployed capacity is not the same thing as revenue in any given quarter.

None of that means the number is meaningless. It means the 2GW is not a target Nvidia must hit to keep growing — its revenue is already expanding at more than 80% year over year, and this capacity is a rounding consideration next to that base. The figure to watch is narrower. It is whether the recurring usage-linked revenue share starts to show up in Nvidia’s reported economics, and whether the eight partners convert “up to” capacity into powered, connected megawatts on something close to schedule. Until then, the more disciplined reading of “Nvidia’s next leg is power” is this: Nvidia’s growth increasingly depends on other people’s ability to deliver watts on time — and that dependency, not the 2GW headline, is the risk an investor is actually underwriting.