Scotland has become one of the fronts in the global race to build AI infrastructure. Plans are in place for at least 23 giant data centres, some of which would rank among the biggest in the world. Supporters promise billions in investment and thousands of jobs. Opponents point to the scale, the electricity, the water used for cooling, and the noise of thousands of servers and cooling systems running without pause.
The Scottish Greens and campaign groups have called for a moratorium on large data centres, arguing they could undermine climate targets and strain renewable supply. The Scottish government has refused to impose one.
Read it in megawatts, not acres
Coverage of data centres usually reaches for land area. Scotland's proposals come with the figure that actually decides things.
The largest data centre in Scotland today is thought to be DataVita's site in Airdrie, with a capacity of 12MW. The one proposed at Chapelcross in Dumfries and Galloway would be 1,000MW. At Auchtertool in Fife, a planned 600MW site would cover 60 hectares — about 84 football pitches, roughly the size of the neighbouring village.
Twelve to a thousand is not growth. It is a different category of object. A 1,000MW continuous load is the output of a large power station, drawn by a single customer, in one place, all day and all night. A building that size is best understood as an electricity consumer that happens to compute, and the planning argument that follows is really an argument about who gets the power.
Why Scotland
The attraction is not mysterious: a cool climate that cuts cooling costs, substantial renewable generation, and land. Those are real advantages, and they are also the source of the objection — a renewable grid built to decarbonise existing demand does not automatically have room for a new demand of this size, and every megawatt a data centre takes is a megawatt not displacing a fossil one elsewhere.
That is the moratorium argument in one line, and it does not depend on disliking data centres.
The same arithmetic, applied here
Bangladesh wants this industry. There is national data centre capacity at Kaliakoir, hi-tech parks have been built partly to attract it, and "hosting AI compute" appears in more than one strategy document.
So run the numbers in the same unit. Bangladesh's total installed electricity generation capacity is on the order of 26,000MW, and the country has spent years managing peak-season shortfalls rather than surplus. A single Chapelcross-sized facility would be roughly 4% of national capacity, drawn continuously, by one customer.
That number settles a question that is usually left open. Bangladesh is not in a position to host hyperscale AI training capacity, and no amount of policy enthusiasm changes the generation arithmetic. What it can realistically host is the smaller tier — facilities in the tens of megawatts serving domestic banking, government and telecom workloads, where the value is data staying in the country rather than raw compute scale.
Being clear about which tier is on offer matters, because the two attract different investors and justify very different infrastructure spending.
The jobs figure deserves the same scepticism
"Thousands of jobs" is the standard promise, in Scotland and here. Construction and operation need separating. Building one of these employs a large workforce for a few years. Running it does not: a hyperscale facility is deliberately designed to need very few people per megawatt, because staff are a cost and automation is the product.
The durable employment from a data centre is mostly in what gets built on top of it, not inside it — which is an argument for caring about where the capacity goes, not for refusing it.
Related: Amazon's data centre approved by the M25, the Texas deed fight, and the Finnish town heated by server exhaust.
