AI infrastructure /
The next resource boom in Australia is compute
Australia's next resource boom may not be iron ore, gas or lithium.
It may be compute.
That sounds abstract until you remember what compute is becoming: electricity, land, chips, cooling, cables and software capability turned into intelligence on demand.
The question is not whether Australia should build data centres.
We probably should.
The better question is what Australia gets in return.
I watched David Pocock's conversation with Dr Sue Kay on the cost of compute, AI, data centres, energy and Australia's future. The useful part of the discussion is that it avoids the lazy version of the debate. It is not anti-AI. It is not a blanket argument against infrastructure. It is a much sharper question.
If Australia supplies the land, water, energy, grid capacity and political stability for the AI economy, do Australians end up with real capability, or do we become a hosting layer for someone else's models?
That is the line worth paying attention to.
Compute is becoming a physical resource
AI can feel like software floating above the world.
It is not.
Large models run on chips. Chips sit in racks. Racks sit in buildings. Buildings need power, cooling, land, water, cables, planning approvals and grid connections.
The old cloud data centre was already physical infrastructure, but the AI version changes the scale. Dr Kay makes the simple distinction well: cloud services may have used tens of megawatts. AI training and inference can push facilities into hundreds of megawatts, with gigawatt-scale projects now part of the conversation.
At that scale, compute starts to look less like an app subscription and more like industrial infrastructure.
That should suit Australia.
We know how to build difficult things in difficult places. We have mining automation, port automation, energy engineering, renewables, battery storage, robotics talent, submarine cable connections, political stability and a long history of solving problems under physical constraints.
The optimistic version is obvious enough.
Australia could become a clean compute nation.
We could turn renewable energy into useful AI infrastructure. We could support local research, edge AI, robotics, scientific computing and operational AI for physical-world industries. We could sell sustainable compute to the region and still reserve enough capacity for Australian companies, universities and public-interest work.
That is a serious national opportunity.
It is also easy to waste.
Hosting is not capability
There is a difference between a country hosting data centres and a country building AI capability.
Hosting means the buildings are here.
Capability means Australians can use the compute, tax the value, build companies on top of it, train people for the jobs around it and keep enough control to switch vendors when the market changes.
That distinction matters because the tenant inside a data centre may control the real economic activity. The operator may own the facility, but the hyperscaler or customer inside it decides what compute runs, where the revenue goes and who gets access.
Australia could end up with the visible infrastructure while the valuable layer sits elsewhere.
That is the gas lesson in a new form.
We know what it looks like to export a strategic resource, underprice public return and later discover that locals still pay high prices. Compute is not gas, but the pattern is close enough to make the question uncomfortable.
Who owns the valuable layer?
Who gets access when demand is tight?
Who pays for the grid upgrades?
Who receives the tax revenue?
Who builds the local companies, tools and skills that remain after construction ends?
These are not anti-technology questions. They are pro-capability questions.
Free tokens are not sovereignty
One of the more useful points in the video is the difference between real access and vendor credit.
If Australia trades public resources for 'free tokens', that is not national capability. It is a coupon.
Tokens can be useful. They are not neutral. They tie users to the provider's model, platform, pricing and terms. They can create the appearance of access while deepening dependency.
A better deal would look more like compute reservation, offtake agreements or interoperable access. Australian universities, startups, public agencies and critical industries should have a defined path to use the compute built here, preferably without locking themselves into one vendor's whole stack.
This is the same rule I keep coming back to at company level.
Use vendors. Do not let one vendor own your workflow, data layer, evaluation system and operating model.
At national scale, the same logic applies. A compute boom should increase Australia's optionality, not reduce it.
The grid should be part of the deal
The other hard question is energy.
If AI data centres need large amounts of power, the answer cannot be a vague promise that renewables will appear in time.
Australia's opportunity depends on clean energy. That should make us ambitious, not casual. New data centres should add to the energy system rather than quietly competing with households and businesses for constrained supply.
That means project approvals should ask more than whether a building can be built.
They should ask:
- What new generation comes with it?
- What storage backs it?
- What transmission or grid support does it fund?
- How does it handle water use?
- Can waste heat be reused?
- Does the facility help stabilise the grid, or only draw from it?
- What happens when the anchor tenant changes?
Those requirements need to flow through to tenants, not only operators. Otherwise the public regulates the shell while the real behaviour sits inside it.
Good infrastructure deals make the surrounding system stronger.
Weak ones externalise the hard parts.
Australia should build the layer above the buildings
The most interesting upside is not the concrete.
It is the layer above it.
If Australia has clean compute, then we should use it to build things Australia is unusually suited to build: AI for mines, ports, farms, reefs, energy assets, hospitals, logistics, defence-adjacent industry, robotics and field operations.
This is where the optimistic story becomes more than a tax argument.
Australia has a long track record in physical-world automation because distance, labour scarcity, safety and harsh operating environments forced the issue early. That experience matters. The next generation of AI will not only live in chat windows. It will show up in robots, sensors, drones, maintenance systems, planning tools, compliance workflows and field-support systems.
Those systems need more than model access.
They need operational knowledge, safety boundaries, audit trails, human approval, source citations and enough local context to work in the real world.
That is a better national strategy than hoping foreign model access trickles down into productivity.
Use the compute boom to build the operating layer around AI.
Jobs are a design choice
The video also gets into work, and this part needs a careful middle position.
AI will replace some tasks. Some companies will use it to cut headcount. Some roles will shrink. Pretending otherwise makes the optimistic case less credible.
But job outcomes are not determined by the technology alone.
They depend on incentives, regulation, management choices, workforce training and whether organisations use automation to expand useful work or only remove labour cost.
Dr Kay's robotics comparison is useful here. In physical automation, safety standards and accountability are normal. If a robot injures someone on a factory floor, responsibility does not disappear into a terms-of-service document.
Non-physical AI should not be allowed to escape responsibility because it arrives as software.
That does not mean freezing adoption. It means designing the transition properly: retraining, accountability, human approval for high-risk decisions and incentives for companies to use AI as leverage rather than only labour removal.
A serious compute boom should include a serious workforce plan.
The opportunity is real
There is a negative version of this article that would be easy to write.
Big tech wants our power. Data centres use water. AI takes jobs. Australia will get played again.
Parts of that may be true if the deals are weak.
But it is not the version worth leading with.
The stronger version is that Australia has a genuine opening. Compute is becoming one of the strategic resources of the next economy, and Australia has many of the ingredients needed to supply it well.
The country should be ambitious about that.
Build the data centres.
Build the energy that powers them.
Build local AI companies around them.
Reserve compute for Australian research, startups and public-interest uses.
Tax the value created here.
Train people for the work around it.
Set standards for water, grid impact, accountability and interoperability before the boom hardens into dependency.
The prize is not a few large sheds full of chips.
The prize is national capability.
The rule
Australia should treat compute like a strategic resource, not a property deal.
The next resource boom may be compute. We should build it in a way that leaves Australia with capability, not coupons.