Construction AI: diesel hits £2 a litre just as the AI build-out piles on the work, and the data centres risk choking on their own cabling

So this is the one that lands straight on your cost sheet. By 29 September diesel had crossed £2 a litre across most of the UK, and on the plant-hire trade's own figures that's a rise of roughly 42 per cent since January. The Construction Plant-hire Association puts a number on what that does to the industry: about £600m on the annual fuel bill, built off infrastructure alone getting through more than a billion litres a year. Fuel is already around 15 per cent of turnover at earthworks firms, so this isn't a line item you can shrug off. A 20-tonne excavator that gets through 440 to 500 litres a week used to cost £620 to £705 to run. It's now £880 to £1,000. Multiply that across a fleet and you can see the problem (CPA figures, reported by Construction Enquirer on 29 September).

It's worth knowing this isn't a quiet drift either. Diesel has gone above its old record, the 199.09p set back in June 2022, and the £2 mark has now been passed in 559 constituencies, more than one forecourt in four (Transport + Energy, 29 September). Steve Mulholland at the CPA and Dr Jonathan Owens at the University of Salford both flagged the timing, which is the cruel part. The order book is fuller than it's been in a while, a lot of it the AI build-out, the data centres and the grid works and the muck-away that goes with them. And the same boom that's filling it is part of what's bidding the fuel price up. You're being paid by the very thing making the job dearer to deliver. Not a comfortable place to run a civils business.

Here's where the AI actually earns its keep, and it's the least glamorous answer imaginable. The telematics box you already pay for on most of the fleet knows exactly which machines are sitting there idling, engine running, doing nothing. At 70p a litre that was noise. At £2 it's real money, and the modelling tools and idle-detection that come bolted into fleet software now can turn that data into a weekly list of the worst offenders. It won't fix the pump price. It will claw back the burn you're not even using.

The procurement filter: before you sign another fixed-price earthworks package, put a fuel fluctuation clause on the table, and price today's diesel, not last quarter's.

The second story this week is about where a lot of that diesel is being burned: the data centres. A survey of 300 senior decision-makers, commissioned by the cabling firm Onnec, so take the framing with the usual pinch of salt, found that 78 per cent admit they've compromised on cabling quality or specification just to get a facility deployed faster (reported by PBC Today on 29 September). Three-quarters, 77 per cent, say cabling has become a critical bottleneck for AI workloads, and 41 per cent say it's already delaying projects. The backdrop is the UK's roughly £100bn data centre construction pipeline, the number that's been driving half the funding rounds we've covered lately.

But think about what that admission actually describes. You've got a building stuffed with GPUs that cost more than the structure around them, and more than a third of those surveyed, 37 per cent, report post-launch bottlenecks that choke the AI training the whole thing was built for. Onnec's own line is the one I'd keep: an undersized pathway or a poor design decision is much harder to undo once the facility is live. It's the first-fix that nobody puts in the drone footage. Compare it to the wiring behind the plaster in a house. Get it wrong and the kitchen still looks lovely, right up until you try to run everything at once and the board trips. The comparison only goes so far, but the point is you don't see the problem until the load's on it.

What that means on site is a straight fight between the programme and the spec, and the survey says the programme is winning. For the people actually running these jobs, the data centre manager, the M&E lead, the commissioning engineer, the temptation is to treat containment and cabling as the thing you squeeze when the date slips. That's the trap. The structure is the cheap bit to get right and the dear bit to redo.

Worth doing: on any data centre bid, ask to see the cabling and containment design before you're asked to commit to the handover date, not after.

Put the two together and you get the shape of the moment we're in. The AI build-out is the best customer UK construction has had in years, and it's also quietly bidding up the diesel in your tank and leaning on your teams to trim the spec to hit its dates. The work is real and the money is real. So is the squeeze. Neither story needs a clever bit of kit to manage. They need the old disciplines done properly: price the fuel for what it costs today, hold the line on the parts of a build you can't cheaply redo, and use the data you're already collecting to stop the waste you can actually see.

That last bit is where AI in construction actually earns its keep, by the way. Not the render of a robot laying bricks. The idle-time report off the telematics, the clause that reprices the fuel, the design review that catches the undersized containment before the racks go in. Unglamorous, checkable, and worth real money on a job that's already tight. That's what it's about.

Today's action: pull one number before Friday, the idle hours across your fleet last month, and work out what they cost you at £2 a litre. It'll be more than you'd guess.