Construction is Spending Billions on the Wrong Problem.
This is best enjoyed using audio. To listen, click here: https://open.spotify.com/episode/7gqHcMHUOrAXALdJT2gs8Z?si=eV4xzjf5Se-J4S8PN0__qw
Someone interviewed fifty senior leaders across the AEC industry about how they're using AI. Every single one was bullish. Budgets going up. ROI landing. Moving from pilots to real deployment. And every single one was using AI to optimise the same broken system that has been underperforming for decades.
Not one of them was rethinking the operating model itself.
That finding, from Matt Gough's Cogital report, sat at the centre of everything that happened this week. A market splitting into two realities on both sides of the Atlantic. A billion-dollar bet on AI that understands physics. And a question the industry still can't answer: what are we actually building towards?
Four threads. All connected.
The industry is automating the past
Matt Gough is not a theorist. Former Innovation Director at Mace. Three years at Neom. Now running Cogital. His report drew on interviews with clients, investors, contractors, consultants, and startups. The headline is that AI adoption is accelerating and returns are being generated. The uncomfortable finding underneath is that everyone is applying it to the same thing: speeding up existing tasks inside existing silos.
Quantity takeoffs. PMO reporting. Back-office admin. All valid. But none of it touches the value chain itself, the way risk is transferred through fragmented contracts, the way information degrades at every handoff, the way a scope misalignment in design compounds by the time it reaches the field.
Matt's argument is that when the cost of intelligence is heading towards zero (GPT-3.5 level inference costs dropped 280 times between late 2022 and end of 2024), the friction is no longer information. It's how the industry organises people around it. And right now, most firms are using AI to do the same tasks slightly faster rather than asking whether those tasks should exist in their current form.
Watch out for Matt's episode: https://www.youtube.com/@bricksbytes
One company stood out across all fifty interviews. They run what Matt calls the "Rule of Five": if a task takes more than five minutes or you'll do it more than five times, use an AI tool. Every employee gets a monthly AI budget. No mandated tools. Open sharing across the organisation. That was the outlier. One out of fifty.
Matt also raised something worth sitting with. Katerra tried to vertically integrate the entire construction process with capital and failed. His question: does AI now make that kind of integration possible digitally? Virtual vertical integration. Not owning the supply chain, but aligning it through technology so everyone can see what's happening in real time from concept to delivery. The ambition was right. The timing was wrong. Maybe it isn't anymore.
Construction doesn't have a moonshot
Healthcare's AI moonshot is cure all known disease. That purpose rallies collaboration, unlocks data sharing, and attracts capital at a different scale. Construction's equivalent is productivity. And productivity doesn't inspire anyone to break silos.
Matt's point is that without a unifying purpose bigger than efficiency, AI adoption will stay shallow. Every firm optimises in isolation. Risk transfer stays adversarial. The transformational potential stays locked behind organisational walls.
Construction does incredible things. The water flows when you turn the tap. The lights work. You have somewhere to live. But the industry is an enabler of outcomes, not an identity in itself. And that matters commercially, not just philosophically. When healthcare professionals share clinical trial data with competitors to cure disease, that's unprecedented collaboration driven by purpose. Construction is nowhere near that. Until it is, every firm's AI strategy will be limited to what they can do inside their own four walls with their own data.
Maybe the moonshot exists at programme level. Matt pointed to the National Hospitals Programme in the UK as a possible example, where saving lives through better healthcare infrastructure creates something tangible enough to change behaviour. Maybe productivity itself just needs better framing: build the same project in half the time with half the waste and double the wages.
Either way, the question stands. If you can't articulate what your firm is building towards beyond "be more efficient," don't be surprised when your AI strategy stays tactical.
The market is splitting, and now it's transatlantic
Two weeks ago on this show, we covered bifurcation in the US market. This week, the same pattern showed up in the UK, and the data is hard to argue with.
Recommended by LinkedIn
In the US, data center construction starts hit $25.2 billion in January alone, with $103.7 billion in trailing twelve-month spend. Contractors with data center work carry 11.2 months of backlog versus 7.6 for those without. The top five hyperscalers have roughly $969 billion in future data center lease commitments, $662 billion of which hasn't broken ground. That's pre-committed demand sitting in the system regardless of what the broader market does.
In the UK, PwC's Construction and Housebuilding Outlook published this week showed industrial construction output up 19% to roughly £11 billion, powered by energy, defence, and advanced manufacturing. Public non-residential was up 18%. Meanwhile, residential merely stabilised, commercial remains subdued, and the Glenigan March index showed overall starts down 10% in the three months to February. Data centre and office starts, however, were up 40%.
Read the report: https://www.pwc.co.uk/industries/documents/construction-and-housebuilding-outlook-h1-2026.pdf
The same week, North Lincolnshire approved the UK's largest AI data centre campus. Elsham Tech Park: 176 hectares, up to 15 hyperscale buildings, 1GW capacity, up to £10 billion in private investment, and over 100,000 tonnes of structural steel across a ten-year build. The government is now consulting on fast-tracking data centres as Nationally Significant Infrastructure Projects and clearing the grid connection queue (which grew 460% in six months) to make room.
Balfour Beatty posted a 3.5% UK construction margin, beating their own target a year early, with a record £22.7 billion order book up 23% year on year. Nuclear, energy, and defence are powering that. Not residential. Not commercial.
The labour data reinforces the split. RSMeans showed US construction wages up 4.6%, with 98% of trade categories above 3%. The BlackRock infrastructure summit identified workforce as the single biggest delivery constraint. The US needs 349,000 workers in 2026. The UK needs 706,000 over five years. The boom lanes are absorbing capacity, and everyone else will feel it in cost and availability.
One signal worth flagging: a private equity firm called E3 Tech is acquiring specialty MEP contractors and embedding AI-native founders from Silicon Valley directly inside them. Their acquisition candidates, even ones they haven't done deals with, are proactively reaching out to access the AI capability. Smart capital has read this split and is already positioning.
World models: the next frontier
AMI Labs raised $1.03 billion this week at a $3.5 billion valuation. Co-founded by Yann LeCun, a Turing Prize winner and formerly head of AI at Meta, who left because he believes large language models have hit a ceiling.
The argument is straightforward. LLMs read and generate text brilliantly but don't understand the physical world. A three-year-old knows that pushing a glass off a table means it falls. No AI model has that intuition. World models aim to teach AI to understand space, geometry, and physics, learning from reality rather than language.
This is where the frontier money is flowing. AMI Labs is backed by NVIDIA, Samsung, Toyota Ventures, and Bezos Expeditions. Fei-Fei Li's World Labs raised a separate billion, including $200 million from Autodesk specifically for 3D design workflows.
For construction, the implication is significant. When AI can reason about how structures behave, how materials interact, how a design decision creates a clash downstream, that's not document review. That's a reasoning engine for the built environment. The CEO of AMI Labs is honest about the timeline: this is fundamental research, not a product you can buy next quarter. But the firms building their data foundations and capturing institutional knowledge now will be ready to plug in. The ones that aren't will be starting from scratch.
The bottom line
The market is rewarding discipline, positioning, and ambition. Balfour Beatty proves you can hit margins the industry said were impossible if you pick the right work. The data centre boom is creating structural winners on both sides of the Atlantic. And the firms that are only using AI to speed up what they already do are leaving the real value on the table.
The question that sits underneath all of it: what is construction actually building towards? Until the industry has a better answer than "be more productive," the transformation will stay shallow.
This article is the companion piece to the Bricks and Bytes Executive Weekly Briefing podcast. Listen to the full episode wherever you get your podcasts. Join the conversation.
Interesting article 👌 The uncomfortable truth is that AI isn’t exposing a technology gap - it’s exposing an operating model problem. We’re applying intelligence to optimise tasks, but the real inefficiency sits between them - in how decisions, data and risk move across phases and actors. Until we address that, we’ll keep improving local performance while the system underperforms globally. The real opportunity is not better tools, but shared decision frameworks and continuous data flow from concept to construction. That’s also why the idea of “virtual vertical integration” is interesting - not owning the supply chain, but aligning it through a common data and decision layer.
I agree that much of the investment in construction tech today focuses on automating existing workflows rather than rethinking the operating model of projects. In many cases we are simply making fragmented processes faster. The real opportunity for AI may be in acting as a coordination layer across the entire project ecosystem, connecting documents, communications, schedules, and decisions so teams can reason over the project as a whole rather than navigating disconnected tools. This is exactly the direction we are exploring with WisyPlan where we are experimenting with agent-based AI that can understand and connect multiple sources of project information rather than just automate individual tasks.
Spot on, Owen Drury. We've poured a ton of money into digitising old processes rather than actually preparing for the future of production. Our industry is so fragmented and bogged down by regulations that we've fallen into a "compliance-first" tech mindset. We've been treating BIM more as a way to defend against claims and manage documents, rather than as a way to boost site production. In my opinion, we should aim for a true Digital Twin and reduce data fragmentation in unique surfable waves, but we can't just jump straight there. The industry can't handle a complete rethink all at once. First, we need to stabilise the technology by bringing in AI as a "Universal Translator." If we use tools for different contexts, we can tackle immediate challenges and build up from there. Once we get the "work face" sorted and make our current data easy to access and machine-readable, the whole Digital Twin thing will just naturally follow. Instead of a risky, decade-long leap, we'll see a "Tech Push" transform into a "Production Pull," and the Digital Twin will become a straightforward next step.
My recent post on Industrialised Homes vs Industrialised Data Centres: Where Will Innovation Converge? https://www.linkedin.com/posts/audley-english-mbe-frsa-aa-dipl-riba-10981191_data-digital-industrialisedconstruction-activity-7436000869499691008-5Iau?utm_source=share&utm_medium=member_desktop&rcm=ACoAABOFCFkBF-orYGssEXH2z14jRN6jqkIH1FA
Katerra right idea to vertically integrate the entire construction process, but despite strong capital, it failed to create a construction assembly system that worked across all sectors. World Models are the future, and their influence will grow with quantum computing, enabling a unified platform like how Apple’s App ecosystem reshaped mobile technology. Data centres will accelerate “moonshot” innovation, yet most housing solutions remain tied to traditional thinking, focused on automating old methods rather than starting from a blank sheet. In the future, all products will be digitally designed, simulated, tested, and costed before physical application already standard in the automobile and aerospace industries, which began in 1886 and 1903. The construction industry needs true visionaries. I once met Alan Bean at NASA, the Apollo 12 Lunar Module Pilot and fourth person to walk on the Moon. That meeting introduced me to the use of mirroring during Space Station development, which I saw under construction. NASA later introduced the term “Digital Twin” in 2010, now widely used alongside “moonshot,” both originating from NASA.