Here is a question you probably did not expect to think about this week. Can an algorithm design better concrete than a materials engineer?
The answer, it turns out, is increasingly yes. And it matters far more than it sounds.
What Meta Built
Last year, Meta's engineering team published something genuinely interesting. Working with Amrize (one of the world's largest cement manufacturers) and the University of Illinois, they built an open-source AI tool that designs concrete mixes. Not theoretical formulas. Actual concrete that has been poured on an actual data centre slab in Rosemount, Minnesota.
The system uses Bayesian optimisation to balance multiple competing demands at once: compressive strength, curing speed, workability, surface finish, and carbon footprint. A human engineer optimising a concrete mix typically adjusts one variable at a time. The AI evaluates thousands of combinations simultaneously, learning from each test cycle to suggest the next promising formula.
The results are striking. Within two iterations, the AI-designed mixes exceeded standard low-carbon industry formulas across all key metrics. Faster curing. Higher strength. Lower embodied carbon. And the tool is open-source, meaning any concrete supplier can use it.
8%
of all global CO2 emissions come from concrete production
15%
reduction in pricing errors from AI-led valuation tools (PwC)
20%
embodied carbon saved by 3D-printed concrete vs block construction
Why Construction Should Care
Concrete accounts for roughly 8% of all global CO2 emissions, according to the World Economic Forum. In the UK, where the government's net-zero targets increasingly shape procurement decisions, the embodied carbon of structural materials is no longer a footnote. It is a specification requirement.
For quantity surveyors and project managers, this changes the conversation around material selection. If an AI-optimised concrete mix can deliver the same structural performance with measurably lower carbon, that has direct implications for lifecycle cost assessments, BREEAM scoring, and carbon reporting. It also raises questions about how we evaluate material specifications when the design process itself was algorithmic.
3D Printing Adds Another Layer
The concrete innovation story does not stop at mix design. In Ireland, three council homes in Dundalk have been completed using a 3D printer that fits on a single lorry. The 132-day project is the first in the world to meet internationally recognised technical standards. Each home cost roughly GBP 205,000 to build, and the architects, Manchester-based Harcourt, are already in talks with UK developers about similar projects.
The walls are load-bearing, double-cavity, and built in less than two-thirds of the time taken for standard construction. Compressive strength is about five times that of traditional methods. And the concrete uses locally sourced supplementary materials like GGBS, saving around 20% in embodied carbon compared to standard block construction.
The combination of AI-designed concrete and automated construction methods is not some distant future. It is happening now, on real projects, with real cost and carbon savings.
The Responsible AI Angle
There is a governance dimension here that is easy to miss. When an algorithm designs a concrete mix that gets poured into a structural slab, who validates that formula? The AI optimised for multiple variables simultaneously, but concrete performance depends on local conditions, aggregate sources, ambient temperature, and curing environment. The gap between lab-tested AI output and real-world performance is where professional judgement still matters.
This is not an argument against using AI in materials design. It is an argument for understanding what the AI did, how it made its recommendations, and where human verification fits in the chain. The same principle applies whether you are talking about concrete mix design, automated valuations, or AI-generated planning assessments.
The technology is genuinely impressive. The question is whether the governance is keeping pace.
