“It was the best of times, it was the worst of times.”
A Tale of Two Data Centres
One city approved a 225-megawattAI data centre the size of 14 football pitches. Another refused one almost identical in scale — unanimously. Same technology. Same decade. Opposite outcomes.
This is the largest infrastructure build in human history — $1.7 trillion in worldwide data-centre capex projected by 2030. These are the factories of the 21st century. And like every industrial revolution before it, it is not arriving evenly.
The boom
320 MW at full build
NEXTDC’s M3 campus in West Footscray runs at 225 MW operational today, with a full-build target of 320 MW. It holds 41,000 m² of technical space across a campus the size of roughly 14 football pitches — about six pitches of raw computing floor alone.
Put another way: M3 is around 1.5 times the size of the MCG complex. These aren’t server rooms. They’re the factories of the 21st century, and they’re being built at industrial scale on the western edge of Melbourne.
take-up driven by AI
Melbourne’s total data centre supply has nearly tripled to 4.7 gigawatts. And the demand isn’t coming from the cloud era we already know — 95% of new co-location take-up is being driven by AI workloads.
As Sydney runs into land and power constraints, Melbourne is absorbing the overflow and becoming a hyperscale hub in its own right.
of AirTrunk (AUD)
The scale of the money tells the story. Blackstone acquired AirTrunk for $24 billion AUD — the largest-ever Australian digital infrastructure deal.
This is global capital treating Australian compute as a long-term industrial asset. The buildout isn’t speculative froth; it’s being underwritten by some of the largest investors on earth.
24 data-centre projects
Jemena, a local electricity distributor, has 24 data centre projects in its connection pipeline requesting a combined 4,000 MW. That is roughly four times Jemena’s entire current summer peak demand.
To manage the uncertainty, the sector uses a “contingent projects” model: $75.1 million is earmarked for grid upgrades around the Tullamarine and Heidelberg clusters — but only funded once the load actually materialises. The grid is being asked to plan for a demand surge nobody can yet confirm.
— if renewables keep up
Here’s the part that surprises people. Because these facilities consume so much power, they spread the grid’s fixed costs across a far larger volume of energy.
In theory, that could actually lower the average household electricity bill — provided renewable supply scales fast enough to keep up. The boom is genuinely double-edged: real strain on the grid, but also a potential dividend for everyone connected to it.
The backlash
~8 football pitches
The proposal sat on the former Royal Bank of Scotland HQ at 1 Redheughs Avenue, South Gyle — a 5.74-hectare site, around eight football pitches. The proposed capacity: 210 to 213 MW, almost a mirror image of Melbourne’s M3.
The developer was Shelborn Asset Management (Shelborn Drummond Ltd), a private landowner — not one of the household-name hyperscalers.
impact” policy assumed
This is the heart of it. Scotland’s 2022 National Planning Framework 4 (NPF4) classified “green” data centres as having “negligible impact” on greenhouse gas targets.
But that assessment predates ChatGPT. It was written assuming facilities of around 10 MW — not 210 MW hyperscalers. That’s a scale mismatch of roughly 20 times. The site was marketed as a “green data centre” with renewable sourcing claimed, but no verified PUE was ever published.
The policy wasn’t wrong when it was written. The world it described simply stopped existing.
the facility would draw
Action to Protect Rural Scotland (APRS) put numbers to the objection. The facility would use enough electricity to power a quarter of a million homes.
And when its diesel backup generators run, APRS estimated they would emit air pollution equivalent to more than 100,000 idling diesel cars. For a development sold as “green,” the backup story told a very different one.
by the planners
For a build of this scale, the absence is conspicuous: planners did not request an Environmental Impact Assessment. That became a point of real controversy — a 210 MW facility waved through the early gate on a policy framework written for something a fraction of its size.
The transparency gap runs through the whole Edinburgh case: no published PUE, no disclosed rack count, no confirmed cooling specs. The opacity itself became part of the argument against it.
refused outright
In February 2026, Edinburgh City Council refused planning permission unanimously — on climate impact, lifecycle emissions, and conflict with City Plan 2030 (Place 19, the area’s mixed-use housing and leisure vision).
And here Scottish planning law matters. Unlike England, the appeal route is uncertain — the developer may have to return with a redesigned proposal rather than simply appeal the decision. This was one of 16 hyperscale data centres currently in the Scottish planning pipeline; it will not be the last fight of its kind.
Two facilities, compared
Note the asymmetry.Edinburgh discloses far less than Melbourne — no published PUE, no rack count, no confirmed cooling. The shaded cells aren’t missing data; they’re the transparency gap, and it’s part of the story.
The bigger picture
safe margins by 2030
The constraint on AI has quietly shifted. It’s no longer the supply of chips — it’s grid-scale electricity.
Seven of thirteen major US grid regions are projected to operate below critical safety margins by 2030. Real blackout risk is now a factor being modelled into AI deployment decisions.
off-grid by design
A new pattern is emerging: “bring your own power.” Tech giants are financing Small Modular Reactors — nuclear — for reliable, carbon-free baseload independent of public grids.
Others are blunter about it: some operators (Musk among them) are carting in gas turbines and diesel to run facilities off-grid entirely. When the public grid can’t move fast enough, the buildout simply routes around it.
data-centre water demand
Power gets the headlines. Water is the hidden constraint. Australian data centre water demand is projected to more than triple — and it rarely gets airtime in the debate.
The cooling arms race is producing strange experiments: China has trialled an underwater data centre roughly ten miles offshore, cooled directly by seawater. The places we put compute are starting to follow the water.
compute you don’t control
There’s a sovereignty problem hiding underneath the infrastructure. Models can be pulled by government decree — the Fable 5 export-control example is the cautionary one.
Building a national economy on compute and models you don’t ultimately control is its own category of risk. The physical data centre may sit in your country; the thing running on it may not be yours to keep.
these investments
Here’s the question every figure in this episode circles back to. Is this multi-trillion-dollar supercycle about automating jobs, or augmenting them?
The optimistic path is labour augmentation — AI as a cognitive force multiplier. The dark path is the one where the spend only makes sense through job replacement. Realistically, it’ll be some mix. And we’ll find out fast: the payback window on these investments is just three to five years.
The Dickensian question
The best of times and the worst of times isn’t a prediction.
It’s a choice.
Who owns the factories this time?
Privately owned, global in scale, with limited public accountability — the communities host the physical burden: the noise, the grid strain, the diesel. The infrastructure serves a global market, not the local economy.
But it could power local booms too — compute for cancer research, for aged-care robotics, for industries that don’t exist yet.
— Stephen & Lauren