75% of Construction Project Professionals Are Now Using AI

The Association for Project Management (APM) and Censuswide published survey data this week that puts the sharpest number yet on the pace of AI adoption in UK construction. In 2024, 15% of construction project professionals were using AI. Today, that figure is 75%. Ninety-one per cent plan to increase AI investment in 2026.

But the headline adoption figure hides a more uncomfortable story. Only 36% of firms have both a clear AI strategy and tangible results to show for it. Seventy-six per cent of senior leaders say they expect AI to transform their business. Forty-two per cent identify legacy technology as the single biggest barrier to meaningful adoption. That gap - between aspiration and operational deployment - is where most firms are stuck.

The London Build Expo data adds texture. Sixty-eight per cent of firms believe construction is adopting AI faster than other sectors. But when you filter for firms with an actual strategy and documented outcomes, the picture narrows considerably. Using AI is not the same as getting value from it.

The FMB (Federation of Master Builders) published a practical guide this week on five ways SME builders are using AI right now - covering estimating, health and safety documentation, tender preparation, site communications, and customer queries. It's worth reading if you're a smaller firm and want a grounded starting point, not a case study from a tier-one contractor.

London-headquartered AVEVA announced a collaboration with NVIDIA in March 2026 to integrate its engineering and operations software into the NVIDIA Omniverse DSX Blueprint. The joint work is building physical and digital modules for large-scale data centres - what NVIDIA calls "AI Factories" - using lifecycle digital twin architecture.

The significance isn't just the technical partnership. It's the direction it points. The digital twin methodology that AVEVA and the broader UK construction industry have been developing for complex infrastructure projects is now being applied directly to the build-out of AI compute infrastructure. Data centres are some of the most demanding construction projects in the current pipeline: high-specification MEP works, complex power and cooling systems, tight tolerances on commissioning. Digital twin capability that's proven on infrastructure projects transfers directly to this market.

For UK contractors, the pipeline is real. Domestic data centre construction is growing - from the Kao Data campus in Harlow to QTS projects in Northumberland - and the international scale of what's being built (Microsoft's 2.1-gigawatt campus in Texas, covered here on 30 March) shows where that market is heading. Firms with AVEVA or equivalent digital twin capability are well-positioned for it.

Researchers have published a framework for automated conversion of paper construction drawings into 3D digital twins - directly addressing one of the biggest barriers to retrofit and net-zero compliance work in the UK. The EurekAlert-published research demonstrates that the process can be automated, rather than requiring manual re-drawing or expensive survey work.

The context matters. The UK has an estimated 5.6 million non-domestic buildings. A significant proportion were built before digital records were standard. Retrofit and net-zero programmes need asset data - floor layouts, structural information, M&E configuration - that often doesn't exist in a usable digital form. Traditional options are expensive: commission a new survey, or pay for manual digitisation. An AI-powered conversion pipeline from paper to digital twin changes that cost equation materially.

The framework is research-grade rather than a product you can deploy today. But the direction it points is clear. Scan-to-BIM tools are already in production (NavLive being the most prominent UK example, as covered on 31 March). The next step is making that pipeline accessible for the vast inventory of buildings that exist only as paper records.

An industry survey of nearly 800 architects and designers globally - conducted by Chaos and Architizer - finds that 64% are already experimenting with AI tools in their daily workflows. One in five firms has fully integrated AI across multiple processes. Between 73% and 93% of respondents plan to increase AI use in the coming year.

The survey is notable for what it doesn't find. There's no longer a significant "wait and see" cohort. The conversation in architectural practice has moved from whether to adopt AI to how to integrate it in a way that protects creative authorship and delivers measurable value. Key use cases: concept design acceleration, visualisation, and automating repetitive drawing tasks.

This matters for the construction industry beyond architecture. Architects are typically upstream of the construction process. When design AI is embedded in architectural practice, the models and data that arrive on site are different - richer, faster, more iterated. That has downstream implications for contractors, engineers, and project managers who rely on design-stage outputs.

The EU AI Act's general obligations take effect on 2 August 2026. For UK construction firms using AI for safety-critical applications, worker monitoring, or automated decision-making, that deadline is not a distant future concern. It's four months away.

Legal analysis from Browne Jacobson is direct on the implications: standard construction contracts have not yet adapted to AI deployment. That creates liability gaps and commercial risk for firms that haven't audited their AI tools. A firm using an AI system for risk assessment on a higher-risk building, for example, may carry obligations under the Act that aren't reflected in their contract or their supply chain agreements.

The anticipated UK AI Bill has been delayed to mid-2026 at the earliest. That leaves a regulatory gap between the EU Act applying (for firms with EU exposure) and any equivalent UK framework coming into force. Legal teams are navigating it on a case-by-case basis. Construction firms should be doing the same.

The compliance question isn't just about the EU Act. It's about understanding which AI tools in your stack - from site monitoring cameras to risk analysis software to automated scheduling - fall into categories the Act treats as higher-risk. If you don't know the answer, that's where to start.

KPMG's first quarterly Global AI Pulse Survey covers 20 countries and companies mostly with revenues above $1 billion. The headline finding: 74% of global leaders say AI remains a top investment priority even if a recession hits. Companies plan to invest a weighted global average of $186 million in AI over the next 12 months.

Sixty-four per cent say AI is already delivering meaningful business outcomes. Thirty-two per cent are already deploying and scaling AI agents. A further 27% are orchestrating multiple agents across their business.

The construction and real estate sector breakdown is due in the full report release on 15 April. That will be worth reading. But the macro signal is already clear: AI investment isn't cyclical in the way that construction spend is. Firms cutting AI budgets in a downturn are moving against the tide.

The top concern across all respondents: data security, privacy, and risk. Nearly three in four leaders flagged this. That's the same concern that construction firms deploying AI on safety-critical projects should be building into their governance frameworks now, not retrofitting later.

Anthropic's Claude Code CLI had its source code exposed via a source map file in their npm package registry this week - 500,000-plus lines of TypeScript made unexpectedly public. Anthropic moved to contain redistribution via DMCA takedowns. The model weights were not exposed.

The construction implication is indirect but worth noting. What the leak revealed - and what the developer community focused on - is that the performance gap between AI tools isn't primarily about the underlying model. It's about the orchestration layer: persistent memory, autonomous modes, planning and review loops, integration with external systems. The "harness" that wraps the model is where the real engineering leverage sits.

That has a direct parallel for construction firms building AI into their workflows. The value of AI on a project isn't mainly determined by which model you're using. It's determined by how well that model is integrated into your existing data, processes, and decision points. A well-integrated Claude Sonnet 4.6 will outperform a poorly-integrated GPT-5 every time.

The secondary risk from the leak is also instructive. Attackers quickly registered suspicious npm packages to target developers attempting to compile the leaked code. It's a reminder that AI supply chain security isn't just a concern for software developers - any firm integrating AI tools from external vendors needs to understand what dependencies those tools carry.