The Engine Room Nobody Can Farm
Week of 12 June: the $155 billion at the door, the one objection that's real, and a continent of sun to answer it
A fortnight ago Sam Altman beamed into a Sydney conference and told 800 of CBA’s best customers that Australia could be the “data centre capital of the world.” The numbers back him up. Business investment is growing at its fastest clip since the peak of the mining boom in 2012, and data centres are doing the heavy lifting.
Right on cue, the chorus of objections has arrived. Too much power, too much water, too noisy. Senator David Pocock wants to tax these things before they’ve earned a dollar of profit.
So this week we’re sorting the real problems from the confected ones, because the prize on the table is enormous and we are entirely capable of fumbling it.
The Size of the Prize
Start with the scale. Australia was the second-largest destination for data centre investment on the planet in 2024, and Westpac puts the current pipeline at around $155 billion. Microsoft has committed $25 billion, Amazon another $20 billion.
NEXTDC, a home-grown ASX success story from Bevan Slattery, is building a $7 billion, 650 megawatt facility at Eastern Creek with OpenAI signed up as the anchor tenant. When it was announced in December it was billed as the biggest in the southern hemisphere. It held that title for about six months before IREN’s 800 megawatt South Australian campus blew past it, with a 1.2 gigawatt proposal in western Sydney queued up behind that.
The opportunity is bigger than it appears on the surface. Right now, Australian companies route a huge amount of compute through Singapore because it’s available and relatively cheap. Here’s the thing though, Singapore froze new data centre construction for years because it ran out of power, and it still rations capacity.
Every workload we bring onshore means more subsea cables and capacity landing here, sovereign control over Australian banking and government data, and downward pressure on the cost of compute for every Aussie business. The economy-wide payoff from AI adoption is estimated at $116 billion to GDP over the next decade, with around 400,000 jobs supported through the construction phase.
Here’s the insurance policy for the AI bubble worriers. Firmus started life in 2019 as a bitcoin mining outfit. It pivoted to AI, and it’s now valued at around $6 billion with Nvidia’s backing and an ASX float in the works. IREN ran the same play.
The lesson? This infrastructure is not a one-way bet on AI chatbots. If AI disappoints, the substations, the cooling systems, the fibre and the silicon get redeployed to whatever comes next, exactly as the crypto kit was redeployed to AI. You don’t refuse to build a port because you’re unsure what the ships will carry down the line.
Water
Here’s the viral stat of the moment, and it’s a beauty. California’s almond orchards alone consume more water every year than every data centre in America combined. It totals more than ten Sydney Harbours annually!
The Australian numbers tell much the same story. Every data centre in the country currently uses a fraction of 1% of national water consumption, which is fundamentally akin to a rounding error.
The answer, as always, is technology. Closed-loop heat exchange and immersion cooling, of the kind celebrated in the Firmus IPO prospectus, uses a fraction of the water.
Energy
This is the genuine pinch point, so let’s not dance around it. Data centres use about 2% of the power on the national grid today. AEMO says that triples to 6% within four years, and the Clean Energy Finance Corporation thinks 11% within a decade.
Let’s concede the obvious. You cannot claim a mining-boom-sized opportunity while praying the energy question away. The boom is only as big as the grid that feeds it. But the mining boom didn’t stall because railways and infrastructure didn’t exist. We built the railways. We built the infrastructure. Demand this big doesn’t shrink the opportunity, it tells you exactly what to build next.
Look at what the smart money is doing. Microsoft signed a 20-year deal that is restarting the Three Mile Island nuclear plant, an 819 megawatt reactor due back online next year. Google and Amazon have signed deals for small modular reactors.
Ford, like many automakers, wrote off billions in EV research and development last year. Yet they’ve just repurposed an underused EV battery factory into a grid-storage business selling straight to data centres, which earned Wall Street’s confidence with its best month in 17 years. GM joined in on the power play this week with grid-scale batteries of its own.
It constitutes an extraordinary pivot. When the good people who build Mustangs and Silverados are pivoting into the energy business, the message could not be louder. Energy is the bottleneck of the AI age, and supplying it is the commercial opportunity of the decade. That’s what serious operators do when demand shows up. They build supply.
Now enter Chris Bowen, who declared that if you want a data centre in Australia, you’ll bring your own renewable energy and you’ll be flexible, thanks very much.
Energy researchers like UNSW’s Dylan McConnell have pointed out the maths simply doesn’t add up. Forecast data centre demand has ballooned by around 10 terawatt hours since 2023, more than the last expansion of the government’s own underwriting scheme added, which means Australia can build the data centres or hit Bowen’s signature 82% renewables target, but not both without more generation actually getting built.
Bowen’s response was that he doesn’t accept it’s a choice. Of course he doesn’t. This is classic Bowen. Announce the target, bask in the photo op, then hold industry to ransom at the expense of jobs and investment. We need cheap abundant energy, and that’s a fact. The answer is gas and the brown coal fleet we already have underpinning renewables, funded by industry, with small modular reactors on the table for private enterprise the second Canberra lifts the ban.
Tasmania is the preview of what happens without a plan. The state’s energy minister confirmed in Budget Estimates that Firmus’s three sites will draw around 400 megawatts, making one AI company the largest power consumer in the state within four years. The first Launceston stage alone is reported to need roughly 15% of Tasmania’s entire electricity supply, in a state where other industrial customers have already been told there isn’t enough energy for their needs. The water in the dams is finite. Nobody planned for this.
Concede the sceptics their strongest case, though, because it’s better than the confected stuff. The grid connection queue is already years long, so every data centre that jumps it pushes somebody else’s project further down the list. Land near existing transmission is scarce, and a boom this size drives up the price of every hectare of it. Then there is the tax take, because the construction deductions keeping today’s bills near zero run off eventually, and unless the profits are still booked here when they do, Australia hosts the depreciation while someone else banks the dividend. All three are fair worries. None is an argument against the boom. Each is an argument about what to build, zone and police next.
So here’s the division of labour, and it’s not complicated. Industry pays. New demand funds new generation, a simple principle that even the union/green bloc agitating on all of this seems to agree with, delivered through the same long-term contracts that just brought an ignominious American nuclear plant back from the dead.
Government does the things that only government can do, which happen to be the things it currently does worst. Approvals must be measured in months rather than years, transmission built ahead of demand, market rules that let data centres share firming infrastructure at the big substation hubs instead of every build deploying its own diesel generation, and genuine technology neutrality so the cheapest reliable energy wins.
The public-private partnership worth having isn’t government picking winners. It’s government building the poles, wires and pipes no single company can, and providing planning and regulatory certainty, then standing back while private capital does what it does best in return.
Noise
Credit where due. Pocock says noise and air quality concerns shouldn’t be dismissed as NIMBYism, and he’s right. That proposed 1.2 gigawatt Mamre Road monster in Western Sydney would ultimately put 852 diesel backup generators and 14.4 million litres of diesel storage near two schools.
In Virginia, the heart of the American data centre industry, voter comfort with local data centres has collapsed from 69% to 35% in two years. This is complete with measured noise hitting 90 decibels at residential boundaries, which is the point at which your Apple Watch declares it’s too damn loud and you might need hearing protection! That backlash is now blocking tens of billions of investment in new projects. We should learn from it, not import it.
The thing to realise is that the problem isn’t data centres. The problem is wedging gigawatt-scale industrial facilities, replete with noisy diesel generation attached, into suburbs full of families. Which brings us to the obvious fix…
Put Them Where the Sun Is
Here’s the thought experiment. Why are we fighting over Western Sydney’s water, power and patience when we have a continent of non-arable land drenched in the best solar resource on earth?
Crypto has already proved that compute migrates to wherever power is cheapest, and AI training especially works much the same way. So build the big ones in the regions, on land nobody farms, next to the sun, the wind and the transmission lines, far from anyone’s bedroom window let alone school.
This isn’t a thought bubble either, it’s already happening. IREN’s record-setting $10 billion campus near Bundey in South Australia sits 125 km from Adelaide on a major substation in the heart of renewables country, serving Asia through high-capacity fibre.
GreenSquareDC has bought 3,100 hectares in regional WA to build its own 300 megawatt wind and solar farm to power its facility. Even the Clean Energy Council crowd is pushing the same idea, proposing data centres be sited inside Renewable Energy Zones as “Digital Energy Zones” that anchor new generation in the regions.
Are there limits to building out there? Of course. Remote builds cost more, the nearest fibre can be hundreds of kilometres away, and summer heat is brutal. But building big industrial projects in the middle of nowhere is the thing Australia does best.
Mining sorted out the camps, the logistics and the engineering decades ago, and plenty of mines already run on their own off-grid solar, wind and batteries, 24/7, with no grid in sight. Liquid cooling handles the heat. Fibre is the cheapest infrastructure we lay, and it follows the highway. Nobody is proposing a greenfield build at Birdsville anyway. The model is Bundey, regional sites on existing transmission and fibre, an hour or two from a city, cheap land, no neighbours.
No, it’s not a jobs play, any more than an iron ore mine is. A small crew runs an enormous asset, and the payoff is the asset itself and everything built to serve it. The fibre, the transmission and the generation outlast any single tenant, the same way the mining boom’s railways and ports outlasted the peak boom.
Water barely rates a mention out there either. A closed-loop system gets filled once, then it’s serviced, treated and topped up, and the fill can come from bore water, not anyone’s drinking supply. Government’s contribution is expediting the fibre routes and transmission links, private capital does the rest.
The tech to put data centres in orbit isn’t there yet but it’s coming, so the bush is the next best thing. Empty, sunny, stable and ours. Cheaper and available right now. We could turn land nobody can farm into the engine room of the Asia-Pacific’s digital economy.
Pocock’s Facepalm Moment
Which brings us to the senator for the ACT, who is a bit one step forward and two steps back a lot of the time. Faux pas, fumbles and follies aside, he retains the title of being the only tolerable teal.
This week he branded data centres the next “Great Gas Scam” and pointed, scandalised, at the tax data. Amazon Web Services booking $3.4 billion of income in Australia but paying only $61 million in tax, Microsoft’s local data centre arm reporting $1.5 billion of income and paying none.
Someone needs to walk the good senator through a profit and loss statement. “Total income” in the ATO transparency data is revenue, not profit. A data centre is one of the most capital-intensive assets on earth.
NEXTDC’s Eastern Creek build is $7 billion for the facility alone before billions more in silicon that depreciates and is superseded within a few years. Of course taxable income is minimal in the build-out years. The deductions are the depreciation on the very investment we’re celebrating. That’s no scam, it’s accounting, and it’s how every airport, port and mine in this country was financed too.
The gas analogy collapses on contact. Gas is a finite resource Australians own, dug up once and shipped offshore. Hence there’s a legitimate debate about resource rent and the obligation to tax it accordingly. It’s hard to argue with.
A data centre’s key inputs are imported chips and imported capital. What Australians supply is land, energy, infrastructure and wherewithal, which operators pay for at full market rates, on top of Labor’s new expectations to underwrite renewables and fund their own grid connections. But if they actually execute that part, we could all stand to benefit too.
Tax the revenue of a business with massive upfront capex and razor-thin early margins and you don’t capture super-profits, you simply tell the next $45 billion from Microsoft and Amazon et al to land in Texas or Tokyo instead. If profits genuinely get shifted offshore down the track, the answer is enforcing transfer pricing rules that already exist, not inventing a turnover tax that punishes investment itself.
On his jobs jibe, that NEXTDC employs only 381 permanent staff, you don’t measure a port by the number of blokes standing on the wharf. You measure it by the trade flowing through it. The payoff from data centres is the productivity of every business that runs on them, plus the 400,000 construction jobs during the build.
Weekly Wrap
So where does that leave us? The data centre boom is real, it’s the biggest investment wave since the mining boom.
Everyone’s talking about water use that’s a rounding error without looking at the next-gen tech on the horizon, about noise that’s a siting failure, and about a turnover tax from a senator who made a hell of a rugby player.
We should welcome the capital, all $155 billion of it.
Build the energy, because the boom is only as big as the grid that feeds it.
Make industry pay for it through demand, the way Microsoft just paid to restart a nuclear plant and Ford just became a battery company.
Australia benefits through data sovereignty, cheaper compute, more cables landing on our shores, nation-building infrastructure projects on land nobody can farm, and a second engine for the economy if mining ever cools.
Where is government? Well it’s in the way of course. Rationing power, mandating technologies, lecturing the very investors it’s courting.
The job from here is simple. Approve fast, build the poles and wires, let private enterprise deploy small modular reactors and solar and storage at scale, and otherwise get the hell out of the way.
Singapore ran out of power. America is restarting reactors and turning car factories into battery plants.
The door is only open exactly as long as we keep it open. Get it right, and we can be truly world leading in this space.