Every few months, the same story pops up in the news: a planning row over a data centre. A giant grey box the size of several football pitches, a new substation, an argument about water, and the promise of a few dozen jobs. And every time, nobody seems to ask the obvious question: why does our computing have to arrive in enormous lumps?
We never really had that conversation. So let's have it now.
Small, buried, and close to home
Here's the idea. Picture a precast culvert section, the sort of concrete tube we already bury under roads by the thousand. Now imagine each one holds a couple of server racks. Join a few together, add an access shaft, and bury the lot under a car park, a field, or the corner of a business park.
For want of a better name, let's call this thing a small modular data centre. SMDC, if you like. The energy industry already made "small modular reactor" a household phrase; maybe compute is next.
Why bother? A few metres down, UK ground sits at a steady 10–12°C all year round. That's free, gloriously boring, utterly reliable cooling. There's nothing to look at, not much to object to, and the land above just carries on doing its job: parking cars, growing grass, whatever it was doing before.
And because the units are small, they can go where the users are: a town, a hospital campus, a housing estate. Compute at the edge, owned by the people it actually serves, instead of rented from somebody else's "region" on a map far away.
Heat is a product, not a problem
Big data centres spend serious money throwing heat away. A small one sitting next to buildings can hand it over instead.
A pair of typical racks draws somewhere around 5 to 8 kW each, and essentially every watt of that comes back out as heat (ASHRAE treats the conversion as 100%). So call it 12 to 16 kW of steady warmth, about what a couple of homes need on a cold day. Pipe it into the office above, or the houses along the road, through a low-temperature heat network, and the same electricity gets used twice: once to do the sums, once to keep someone warm. There's something quite pleasing about that.
AI raises the stakes, in a good way. GPU racks run far hotter than ordinary servers and need liquid cooling, which sounds like a headache until you realise that liquid carries heat away at a much more useful temperature. Those workloads might be happiest under a block of flats or an office building, with plenty of people right above them who'd love the warmth.
The awkward bits (let's be honest)
I'm not going to pretend this is solved. A few things genuinely need answering:
- Big is cheap. Hyperscale exists for a reason: cost per rack falls as you grow. Small has to earn its keep through heat sales, cheaper land, and quicker planning.
- Soil fills up. The ground is not a bottomless heat sink. Push heat into it for years and it warms up, so this needs proper modelling, not optimism.
- Summer. Nobody wants heating in July. Without a year-round use for the heat, you still need another way to get rid of it.
- Lots of small sites. Hundreds of buried rooms are harder to maintain and secure than one big shed. Flooding, access, and fire safety underground all need real answers.
- It won't do everything. Nobody is training a frontier AI model in a culvert. Running one is a different matter: inference, like most of the workloads we actually use day to day, belongs close to the people asking the questions. (Though dense GPU racks will want liquid cooling, mind, not just cool soil alone won't cut it.)
Bigger than the building
One buried room changes nothing. A few thousand small modular data centres, joined up? That might.
If every site speaks the same open standards, they stop being isolated server rooms and start behaving like a cloud. A council, a co-op, or a small business could run its own node, lend out spare capacity, and borrow someone else's when it runs short. No single owner, no proprietary lock-in, and no rent flowing to a handful of companies on another continent.
There's money in it too. A landlord, a homeowner, a group of tenants, a business with a spare patch of ground, any of them could host a unit and earn from it: rent, a share of the hosting income, or simply cheaper heat. Dead space under a car park becomes a revenue stream.
That's the part that interests me most. The culvert is only the catalyst.
So… what if?
I think data centre sprawl is a design choice we've mistaken for a law of physics. I might be wrong, and I'd much rather find that out in public.
So I'd genuinely love to hear from you:
- 🔧 Work in district heating, civil engineering, or planning? Tell me where this falls over. You won't hurt my feelings.
- 🌡️ What would you do with a steady supply of low-grade heat in August? (Seriously, someone must have a good answer.)
- 📍 Where would you bury the first one? Under your supermarket? Your school? Your street?
- 🗳️ And who should own it? The council? A co-op? You?
The comments are open. Bring a better idea. I'll be in there.
Top comments (0)