Construction AI: the levy clock hits 1 October and everyone's shoving applications through the door, while Ofgem puts a price on holding up the grid

Here's the date that's actually moving desks this fortnight. From 1 October 2026 the Building Safety Levy applies in England to new residential schemes of ten dwellings or more, purpose-built student accommodation included, and it only spares you if your building control approval application went in before that day. Get the file in on 30 September and you're out of it, even if you don't break ground for a year. File on 2 October and it's a cost that, in practice, lands on the buyer. The Treasury expects it to raise around £3.5bn over ten years to fund cladding remediation.

You can watch the scramble in the Building Safety Regulator's own numbers. The June-to-August data shows 1,654 Gateway 2 applications in progress, and August alone added more than a hundred, after the caseload had barely shifted from 1,509 in June to 1,556 in July. New-build submissions picked up too, 34 in August. The regulator isn't coy about why: it reckons developers are front-loading applications to get ahead of the levy, and it expects the numbers to climb again through September as the line approaches. Approval rates are actually holding up under the load, 84 per cent over the twelve weeks to 31 August, up from 82, covering something like 18,029 homes.

And here's the aside I'd keep in. The rush is rational, but it's also where mistakes get made. A Gateway 2 application exists to prove a building is safe and the information behind it hangs together. Shove a thin one through the door purely to dodge the levy and you risk an invalidation, which doesn't just lose you the saving, it loses you your slot in a queue that still runs to roughly 22 weeks at the median. So the discipline is dull and it matters: file early, but file complete. The people carrying this are the ones assembling the building control file over the next two weeks, and they'd much rather do it once.

Today's action: for every residential scheme of ten units or more, confirm this week whether the building control application can realistically go in before 1 October, and if it can't, price the levy into the scheme now rather than discovering it at determination.

While the money chases data centres, the regulator is quietly deciding who's even allowed to queue for the power to run one. Ofgem's data-centre connection reform, out for consultation until 16 September, proposes a Data Centre Commitment Fee: a returnable financial security a large-load data centre would post when it accepts a connection offer, refunded when the site actually energises and forfeited if the project bails out of the queue early. The number is the eye-catching bit, between £237,500 and £712,500 per megawatt, which Ofgem reckons is about 2.5 to 7.5 per cent of a typical project's cost.

The reason is a queue that's stopped meaning anything. Demand connection applications have gone from 41GW to 125GW in under a year, and roughly 80GW of that is data centres, most of which will never get built. What the fee does is make the speculative ones pay to hold a place they may not use, so the credible projects can move up. Think of it like a returnable deposit on a table at a restaurant that keeps taking bookings from people who never turn up. The comparison only goes so far, but the point is that a free queue fills with no-shows, and a queue full of no-shows is why real schemes wait 7 to 13 years for a connection in London against 3 to 7 in the rest of Europe.

I think this is the more important data-centre story than any funding round, and it's the one the funding rounds don't mention. A nine-figure raise buys the servers and the steel. It doesn't buy a grid connection, and it can't shorten a queue. So when a data centre lands on your programme, the question that decides your dates isn't the developer's balance sheet, it's whether they've a real connection or a place in a line that Ofgem is about to charge them to keep. That's the bit the site team needs to know before anyone commits to a date.

For your board pack: treat a confirmed grid connection as the go/no-go on any data-centre programme, and ask for it in writing before you resource the job, because the power, not the planning or the money, is now the long pole.

Early September gave us a run of frontier launches, and tucked in behind the model names was the part that actually matters if you're a contractor buying this stuff. On 1 and 2 September, Google, Anthropic and OpenAI all put out cyber-focused models alongside new enterprise safeguards. Google's Gemini 3.8 Flash Cyber goes only to vetted defenders through a programme it calls Fairwind, with 650-plus partners. Anthropic paired Claude Fable 5.1 with Enterprise Frontier Safeguards, which bolts zero data retention onto its misuse detection. OpenAI said its new Astra model hit the Critical cybersecurity threshold under its own preparedness framework and runs a similar Private Safety Processing setup.

Read as cyber news, that's a story for the security team. But look at where UK construction actually sits. Microsoft said on 2 September that about 40 per cent of Kier's staff are on Copilot, and construction ERP outfits like SIS are now shipping dozens of purpose-built agents wired straight into project cost data. What that means is your people are already feeding drawings, RAMS, cost files and safety photos into these models. So the safeguards announcements aren't abstract. Zero data retention is the difference between an agent that reads your golden-thread information and forgets it, and one that quietly keeps it.

The comparison I'd draw is a subcontractor's confidentiality clause. You don't hand a firm your commercially sensitive drawings without knowing what they're allowed to do with them afterwards, and an AI agent is no different, it's just faster and it doesn't sign anything. So the practical move isn't to ban the tools, plenty of good work is getting done with them. It's to ask the vendor the boring question in the contract: where does our data go, how long is it kept, and what does the model refuse to do with it. That question lands on the person signing off the tool, and right now that's often nobody in particular.

A practical step: before you widen any AI agent across a project, get zero-data-retention terms and a data-handling answer in writing from the vendor, and name one person who owns that sign-off.

Line the three up and they rhyme. The levy rewards the firm whose building control file is ready to go before 1 October and punishes the one still scrambling on the 2nd. The grid reform rewards the data centre that can prove a real connection and clears out the ones just holding a place. And the AI safeguards reward the firm that knows what happens to its data and catch out the one that never asked. Different subjects, same test: are your records and your commitments in order before the question gets asked, or are you assembling them in a panic once it has?

That's the standing discipline, and it doesn't come from software. It comes from keeping the file straight as you go, so that when a deadline lands, or a connection needs proving, or a vendor needs pinning down, the answer already exists and you're not building it backwards under pressure. The tools help. They don't do that bit for you.

The takeaway: pick the one deadline that hits your projects first, levy, grid or governance, and spend an hour this week making sure the paperwork behind it is already true.