The concrete problem that brought Meta to a construction site in Minnesota
The United States pours roughly 400 million cubic yards of concrete every year. Nearly a quarter of the cement that goes into it is imported. When a producer wants to switch from an imported cement to a domestic alternative, the chemistry changes. A mix designed for one source will not necessarily perform the same way with another. Reformulating takes months of lab work, trial batches, and field testing.
Meta, which is building data centres at scale across the US, needed concrete that could meet the extreme load-bearing requirements of facilities housing thousands of GPUs.
As Nishant Garg, associate professor of civil and environmental engineering at the University of Illinois and the lead academic collaborator, explained: "These large-scale data centers are hosting much of the data that society is producing, which creates a sense of urgency in building timelines. Their concrete floor slabs must have the high performance and strength-bearing capacity needed to support very expensive GPUs."
Traditional mix design was too slow. So the company partnered with Amrize (the largest cement and concrete manufacturer in North America), Mortenson (the general contractor), and the University of Illinois to develop an AI model that could accelerate the process.
The result is BOxCrete: Bayesian Optimisation for Concrete. Released as open-source software on GitHub in late March 2026, the model uses Gaussian Process regression to predict concrete strength development and optimise mix designs against competing objectives, specifically compressive strength and embodied carbon.
Sign in to read the full story
To Keep Reading Join Project Flux Pro
Get weekly expert AMAs, exclusive AI tools, deep-dive podcasts, and join a community of project professionals mastering AI in project delivery.
Join ProWhat You'll Get::
- Weekly Live AMA & Expert Sessions
- Private Pro Community Access
- Exclusive Podcast & Deep Research
- AI Tools & Templates Library