Real schemes, paper schemes, and the state in the room

The government confirmed on Friday 3 July that mandatory pre-application consultation requirements for Nationally Significant Infrastructure Projects will be scrapped from 24 July, delivered through the Planning and Infrastructure Act. Ministers claim the reform cuts up to 12 months from the planning process and saves industry £1bn over this Parliament; both figures come from the government's own press notice, so hold them at arm's length. In place of the statutory duty, developers get earlier technical support from the Planning Inspectorate, and The Register followed on 7 July with the angle everyone was thinking: the projects that gain most are the data centres the government wants built quickly.

The part that matters more than the headline is that consultation doesn't vanish, it becomes discretionary, and the government's own response says fresh statutory guidance will make clear that applications are unlikely to progress to examination if pre-application engagement was inadequate. So the risk moves rather than disappears. A thin consultation used to be a procedural defect you could point at; now it's a judgement call the Examining Authority makes, and a gift to any objector building a judicial review case on fairness grounds. The developers who treat the freed-up year as a reason to engage earlier and better will bank the time saving. The ones who skip the conversation with the parish council will find the year again at examination, with interest.

I've sat through enough pre-application rounds to say the statutory version had become a performance, three rounds of exhibition boards nobody read, so I won't mourn the duty. But the discipline behind it, knowing what the community objection actually is before you design past it, was always worth the fee.

The tender-desk view: If you support DCO applications, rewrite your pre-application offer this month. The sell is no longer getting the client through the statutory stages; it's doing the engagement that keeps their examination short, now that nobody's forcing them.

Two stories this week, read together, spell out how the government intends the real pipeline to be delivered. Clearstone Energy announced the Ebbsfleet AI Data Centre Campus on 2 July: 300MW on 145 acres off New Barn Road in Kent, up to 180,000 sqm across four buildings, a build cost KentOnline puts at around £3bn, a direct connection to National Grid's transmission network, and a Secretary of State direction treating the scheme as a development of national significance, which lifts the decision out of the local committee room. First facility targeted for 2030, community design workshops this autumn, full consultation in 2027.

Then the Telegraph revealed on 6 July, corroborated by Capacity, that Lord Stockwood, the Minister for Investment, wrote to Epping Forest district council's chief executive in November to "emphasise our strong support" for the Nscale and Microsoft data centre at Loughton. The council approved the enlarged scheme in June despite objections from the town council, from its own landscape and urban design teams, and from residents; officers cited the letter and called data centres critical infrastructure of national importance. Planning correspondence from a minister direct to a council chief executive is, as the reporting politely puts it, highly unusual.

I'm of two minds and I'll say so. The UK needs this capacity, and the construction pipeline attached to it is one of the few unambiguous growth stories the sector has. But approvals won under pressure invite judicial review, local resistance, and the programme uncertainty that reaches the contractor eventually. For the supply chain, Ebbsfleet alone is a multi-year civils, MEP and commissioning programme whose pre-construction services work starts long before the 2027 consultation closes. The work is real. So is the politics wrapped around it.

For your board pack: Split your data-centre risk register into technical risk and consent risk, and track the second one per scheme. The planning route tells you the programme risk, and the programme risk tells you when the tender actually lands.

Turner & Townsend published its 17th annual Global Construction Market Intelligence report this week, covered by New Civil Engineer on Wednesday 8 July. The dataset spans 112 markets across 44 countries, and the headline is blunt: AI infrastructure, data centres above all, is squeezing global construction capacity. Data centres are now the most capacity-constrained sector worldwide, with over 70% of surveyed markets reporting tight or overstretched contractor capacity. In the UK the sector has climbed to second most in-demand, behind defence and ahead of industrial and logistics. The report describes a two-speed market, AI-driven work racing ahead while residential and commercial wait for investor confidence, and Bloomberg ran a parallel story the same day on the US labour crunch, so this is a global pattern rather than a British quirk.

Here's the bit that matters if you never touch a data hall: the squeeze prices everyone's projects, because the M&E fitter, the commissioning engineer and the high-voltage jointer a hyperscaler poaches were on somebody's hospital or school programme first. What we've found on capacity-constrained bids is that the tender price tells you less than the resourcing plan behind it, and right now the resourcing plans are being quietly rewritten across the south east.

I'll hedge honestly: T&T sells programme management to exactly the clients this report alarms, so the framing serves them. But the survey base is big, the direction matches what the Yorkshire Post pulled from the same data, and it matches what anyone pricing MEP packages this summer already knows. The consultants aren't inventing the shortage; they're measuring it.

Worth doing: Reprice the labour assumptions in any bid running past 2027 against data-centre competition for the same trades, especially MEP and commissioning. A programme built on 2025 availability rates is a claim waiting to happen.

Balfour Beatty announced on Monday 7 July that it's investing £10m in Pi Labs' Fund IV, an early-stage venture fund focused on built environment technology that will back up to 50 companies across AI, data and robotics. The detail worth pausing on: this is the first time Balfour Beatty has ever invested in a venture capital fund. A tier one with a £9bn-plus order book has decided that writing a cheque to a VC is now core business, joining strategic investors including Dar Ventures, the venture arm of Sidara, and Dubai developer Ellington Properties.

Put this next to the last fortnight. NG Bailey created a chief AI officer, Kier hired a head of data, Laing O'Rourke is recruiting AI leadership. That was the org chart responding; this is the balance sheet, and it's a different signal, because a job title costs a salary while a fund commitment costs £10m and locks you in for the fund's life. What Balfour Beatty is actually buying is deal flow: early sight of fifty startups before they're polished enough to cold-call an innovation team, and the option to pilot, partner or acquire while the price is still sensible. The question an editor might cut is whether £10m of LP money buys influence or just information; strategic investors get visibility, not steering rights.

For the mid-tier contractor the lesson isn't to go and find a VC. It's that the biggest player in UK contracting has decided tool selection is too important to do reactively, and your version of that can be much cheaper: a named person, a quarterly scan, a decision log for what you trialled and why you dropped it.

The practical bit: Ask who in your business saw the last three construction AI tools before a salesperson introduced them. If the answer is nobody, that's the gap Balfour Beatty just paid £10m to close, and a one-page scanning process closes most of it for you.

McLaren Construction announced on Monday 6 July a partnership with FieldAI, the California robotics firm, to deploy autonomous quadruped robots across its UK sites. The first units are Boston Dynamics Spot robots running FieldAI's Field Foundation Models, tested on a job with some pedigree: the Passivhaus refurbishment of the LSE's 35 Lincoln's Inn Fields. The robots capture 360-degree imagery and point cloud data, verify progress, compare the as-built against the design model, patrol for safety compliance and run quality checks. What separates this from the Spot pilots we've seen since 2020 is the autonomy claim: no prior maps, no pre-planned routes, the robot adapts as the site changes. FieldAI says its technology already works on hundreds of sites globally, which is vendor-reported, and I'd want the definition of "working" before repeating it in a bid.

A robot that walks your site on its own schedule, recording everything, is a wonderful progress record and a data governance question in a hi-vis vest. Where does the imagery live, who can pull it in a dispute, and what's your position when the footage captures an operative doing something the HSE would rather they didn't? Those are contract and policy questions, and they're easier to answer before the first patrol than after the first incident. The site manager who used to walk the floor with a 360 camera on a stick gets that hour back, which is the point. But someone in the office inherits the archive, and that job needs a name on it.

Today's action: Before any autonomous capture lands on your site, get three things in writing from the vendor: where the data is stored, who holds the retention decision, and whether footage trains their models. A supplier who hesitates on the third one is telling you something.

A dated update to last week's roundup, which carried the BSR's 75% approval rate for the period to 30 May. The regulator's figures published on 3 July, covered by Construction Management, show the improvement holding for a second period: 368 Gateway 2 decisions at a 77% approval rate in the 12 weeks to 28 June, representing 6,544 residential units, with 138 new-build and conversion applications live in the system totalling 30,393 units. Acting chief executive Charlie Pugsley says existing-building remediation approvals reached 85% over the last 12 weeks, already past the 65% minimum target for the year. And for the first time there's a Gateway 3 number: 277 applications at the final hurdle before residents move in, with the regulator saying it's working with applicants towards approvals rather than rejections.

I hedged a fortnight ago about whether one good quarter was a trend; I'll commit now. The remediation plan launched in April is working, at least on throughput. What it hasn't fixed is determination time on older complex cases, and an 18-week target for non-complex applications is still pencilled for March 2027. But if you shelved a higher-risk building scheme in 2025 because the gateway felt like a lottery, your programme risk register should be built on these numbers, not last year's horror stories. Gateway 3 is where the golden thread either exists as structured, retrievable information or it doesn't, and 277 schemes are about to find out which.

A practical step: If you have a Gateway 3 submission in the next six months, pull a sample of your handover information this week and test whether someone outside the project can actually find the fire-stopping records in it.

The reply to last week's distillation story arrived. From 10 July, Alibaba is banning employees from Anthropic's Claude Code, classifying it as high-risk software, after researchers found code that read time zones and proxy settings and scanned for keywords tied to Alibaba, ByteDance and Baidu (TechCrunch and SCMP, 4 July). Anthropic says it was a March experiment against unauthorised resellers and distillation. Last week the provenance question was whether you could trust a Chinese open-weights model; this week a flagship American coding tool was found quietly fingerprinting its users for four months. The trust question doesn't have a nationality. It has an audit trail, or it doesn't.

Practical bit: Add one line to your AI tooling register: what does this tool transmit, and who checked? If nobody can answer, that's the finding.

On 2 July, the day Fable 5 returned to general availability, Anthropic published the detail of its cybersecurity safeguards: a four-category classifier that always blocks ransomware and malware development but routes dual-use work like penetration testing to authorisation controls, plus a draft Cyber Jailbreak Severity scale (CJS-0 to CJS-4) and a HackerOne programme paying researchers to find cyber jailbreaks. It's the first time a frontier lab has published, in checkable form, the decision logic for a dual-use domain, and dual-use is where construction's awkward AI questions live: BMS hardening, building services vulnerability reviews, anything touching operational technology. Clients' security questionnaires are starting to ask what the model underneath will refuse to do, and until now the honest answer was a shrug.

Worth doing: Send the safeguards page to whoever owns your security questionnaire responses, and add one line to your vendor assessments: which published safeguards policy covers the underlying model.

Google's Gemini 3.5 Pro entered the second week of July still in limited Vertex AI preview, with leaks pointing at a 17 July target and the slip attributed to token efficiency concerns flagged by early testers, alongside coding performance short of flagship standard. None of it is confirmed by Google, so file it under reported. It's the same story as Sonnet 5's tokenizer in last week's roundup: across the industry, cost per completed task is overtaking benchmark scores as the number enterprise buyers actually evaluate, and Google appears to have decided that shipping a model that burns tokens faster than expected costs more than a fortnight's delay.

A practical step: Take one real workflow, run it through the model you use now, and log tokens in and out against the result. That single number is your comparison baseline for whatever ships on the 17th.

A Construction Wave piece from 2 July shows how far the org-chart response has spread. Laing O'Rourke's AI subsidiary Akordi is advertising for an AI construction optimisation lead, and Turner & Townsend has appointed Emma Gilthorpe, the former Royal Mail chief executive, as chief transformation officer with an explicit AI remit. The titles matter less than the reporting lines: a chief AI officer who reports to the chief executive can stop a bad deployment, while an innovation lead buried three layers down mostly writes newsletters. Smaller firms don't need the salary line, they need the name.

Worth doing: Write one sentence in your management meeting minutes this month: "X owns AI adoption and governance." If you can't fill in X, that's the gap, not the tooling budget.