To The Moon

🇦🇺 🇦🇺 AUSSIE INCOMING 🇦🇺 🇦🇺 

Australia has spent generations exporting resources.

Iron ore.

Coal.

LNG.

Gold.

We extract them, process them, put them on ships and sell them to the world.

But NVIDIA’s latest Australian AI infrastructure announcement made me wonder whether the next Australian resource boom could work very differently.

Because this time, the resource may never physically leave the country.

It might leave as compute.

NVIDIA has announced it is working with Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC and AirTrunk on up to 2 GW of Australian AI-factory infrastructure by 2027.

That is a huge number.

Australia’s current data-centre capacity is estimated at roughly 1.6 GW, meaning the planned NVIDIA ecosystem buildout alone could be comparable with the country’s existing installed capacity.

But the 2 GW number is not actually the part I find most interesting.

NVIDIA says this infrastructure could support not only Australian users, but also regional and global demand for AI compute.

And that changes the way I think about the entire Australian AI infrastructure story.

Because an AI factory effectively takes:

⚡ Electricity

🏗️ Land and physical infrastructure

🧠 Semiconductors

🌐 Networking

❄️ Cooling

and converts them into something that can be sold almost anywhere:

COMPUTE.

Australia traditionally monetises energy by exporting the physical resource.

LNG goes on a ship.

Coal goes on a ship.

But AI infrastructure creates another possibility.

Keep the energy here.

Use it here.

Convert it into computation.

Then export the economic output digitally.

Australia could potentially begin exporting intelligence per megawatt.

That sounds dramatic.

But follow the economics.

AEMO says data-centre electricity consumption in the National Electricity Market is forecast to rise from around 5 TWh in 2025-26 to 34 TWh by 2035-36.

That would increase data centres from roughly 3% to about 13% of electricity supplied through the grid.

And the connection pipeline is already enormous.

In Q1 2026, AEMO was tracking 11 data-centre projects representing 5.4 GW of maximum load through the transmission connection process.

Of that, 4.1 GW was still at application stage and 1.3 GW had moved into implementation. AEMO also makes an important qualification: data-centre demand ramps gradually and not every proposed megawatt should be treated as inevitable future consumption.

That distinction matters.

Because this is not a prediction that every proposed data centre gets built.

It is evidence that AI infrastructure is beginning to collide with something much more physical than software.

THE GRID.

And this is where I think the AI thesis changes.

For the last few years investors have obsessed over the semiconductor bottleneck.

First GPUs.

Then HBM.

Then networking.

Then advanced packaging.

But eventually, the bottleneck moves outside the server rack.

You can manufacture another accelerator.

You can order another rack.

You can raise another billion dollars.

But you cannot instantly manufacture:

500 MW of reliable electricity.

Or the transmission line carrying it.

Or the substation connecting it.

Or properly zoned land beside that connection.

Or the cooling system.

Or the fibre routes linking the facility to customers thousands of kilometres away.

That means the AI supply chain may eventually become constrained by something much older than AI itself:

PHYSICAL INFRASTRUCTURE.

And Australia is unusually interesting here.

The Australian Government itself identifies abundant renewable energy potential, available land, access to advanced chips, political stability and international submarine-cable connectivity as reasons Australia could become a major Indo-Pacific data-centre market.

Between 2023 and 2025, announced Australian data-centre investments could potentially scale beyond A$100 billion.

Australia also attracted around A$10 billion of data-centre investment in 2024, making it the second largest global destination for the asset class that year according to figures cited in the National AI Plan.

So this is no longer a niche infrastructure story.

It is becoming industrial policy.

And government policy is beginning to reflect that.

In July, Canberra announced plans for rules requiring the next generation of large data centres to underwrite new power supply, pay their full share of grid-connection costs and reduce electricity consumption when necessary to support the grid.

Water efficiency is also explicitly part of the framework.

Think about what that implies.

The government is effectively saying:

If you want to build enormous AI infrastructure in Australia…

bring power with you.

That tells me the megawatt itself is becoming part of the AI value chain.

And once that happens, the investment question changes.

Instead of only asking:

Who makes the best GPU?

We should probably also be asking:

Who owns energised land?

Who can secure grid connections?

Who can finance the substations?

Who operates high-density cooling?

Who connects these facilities to other clouds and data centres?

Who controls international fibre routes?

Who can build capacity quickly enough to turn contracted demand into actual revenue

Take NEXTDC.

It reported 740.1 MW of contracted utilisation for FY26 versus 175 MW of billing utilisation.

That gap shows just how much future demand can be contracted before capacity is fully energised and earning revenue.

And this week NEXTDC announced another A$1.1 billion capital raise to fund AI infrastructure.

It was its third fundraising initiative in a little over four months.

There is another second-order thesis hiding inside that.

AI demand may be enormous.

But converting demand into operating infrastructure requires enormous amounts of capital.

So after chips and electricity, another bottleneck could eventually emerge:

THE BALANCE SHEET.

That is a story for another day.

For now, I think the Australian opportunity is simpler.

NVIDIA may supply the accelerated computing.

But Australian infrastructure companies will still need to provide the surrounding system:

Power.

Land.

Buildings.

Cooling.

Networking.

Connectivity.

Capital.

And fibre matters more than it first appears.

Megaport is one of NVIDIA’s Australian infrastructure partners and describes its role as connecting compute, networking and storage across its global automated infrastructure platform.

Which brings us back to the export thesis.

Physical commodities need ports.

Compute needs networks.

Australia’s old resource economy moved value through bulk carriers.

The AI economy could increasingly move value through fibre.

BUT THERE IS A BIG CATCH.

This can very easily become an infrastructure bubble.

A 2 GW announcement does not mean 2 GW of profitable utilisation magically appears.

A connection application does not guarantee construction.

A signed customer does not eliminate execution risk.

And enormous capital expenditure does not automatically create attractive shareholder returns.

Hardware generations move quickly.

Electricity prices matter.

Financing costs matter.

Customers can shift workloads.

Regional competition matters.

Malaysia, Singapore, Japan, Korea and other markets also want to become AI infrastructure hubs.

And there is an uncomfortable historical parallel.

Periods of explosive technological demand have often produced genuine infrastructure needs alongside enormous overbuilding.

The internet really did change the world.

That did not mean every fibre company built during the telecom boom became a great investment.

AI could be similar.

So I am not arguing:

“Buy every Australian data-centre company.”

I am arguing something different.

IF AUSTRALIA BECOMES A REGIONAL AI COMPUTE HUB, WHERE DOES THE SCARCITY MOVE?

Because scarcity usually determines who gets pricing power.

Maybe it remains with NVIDIA.

Maybe it shifts toward power generation.

Maybe grid access becomes more valuable.

Maybe data-centre campuses with secured electricity become strategic assets.

Maybe networking and fibre become increasingly important.

Maybe capital itself becomes the constraint.

Or perhaps the winner is whoever can assemble all of those pieces into one functioning system.

That is why NVIDIA’s Australian announcement caught my attention.

Not because Australia is suddenly going to replace its commodity exports with GPUs.

But because AI could create a genuinely new way for this country to monetise some of the things it already possesses:

Land.

Energy.

Infrastructure.

Geography.

Political stability.

Connectivity into Asia-Pacific.

Instead of digging up a resource and shipping it overseas…

Australia could increasingly turn energy into computation and sell the output globally.

For generations, Australia exported the resources beneath its feet.

What happens if one of the next valuable Australian resources is the ability to turn a megawatt into intelligence?

And if regional demand for AI compute keeps growing…

WHO CAPTURES THE ECONOMICS?

A) GPU suppliers

B) Data-centre operators

C) Energy and grid infrastructure

D) Fibre and connectivity providers

E) The whole stack becomes one giant AI infrastructure ecosystem

My view?

The biggest mistake may be assuming the AI infrastructure boom ends at the chip.

The chip might only be where it begins.

$NVDA $NXT $MP1 $IREN

Personal market analysis only. Not financial advice. Always do your own research.