Why Concrete Needs an AI Upgrade

Every year, the U.S. pours about 400 million cubic yards of concrete—enough to build a two-lane highway circling the Earth multiple times. Yet nearly a quarter of the cement used in that concrete is imported. That dependency creates supply chain risk, stifles domestic manufacturing, and introduces environmental inconsistencies because imported cement doesn't always meet U.S. standards.

Traditional concrete mix design is a slow, trial-and-error process that relies on engineer intuition and decades of accumulated lab data. When a supplier wants to switch to a domestic cement source—which has different chemistry—they often need months of lab validation before they can be confident the new mix will hold up. Meta saw this bottleneck and decided to apply Bayesian optimization to solve it.

Source: Meta Engineering Blog

AI model Bayesian optimization interface for concrete mix design showing strength prediction graph Coding Session Visual

How BOxCrete Works (Code-Free Explanation)

Meta’s Bayesian Optimization for Concrete (BOxCrete) model is built on their existing Adaptive Experimentation (Ax) platform. Instead of testing random combinations of cement, aggregates, water, and admixtures, the AI:

  1. Learns from historical data – past mix designs, lab results, and performance metrics train a Gaussian Process model.
  2. Proposes high-potential candidates – suggests new mixes most likely to meet target strength, workability, and cost constraints.
  3. Incorporates constraints upfront – users specify exactly which domestic materials must be used and what performance thresholds are required.
  4. Refines with each test – every lab result improves the model’s predictions, creating an automatic improvement loop.

The key insight is that Bayesian optimization is sample-efficient. It doesn't need thousands of data points to start being useful—just a few dozen well-chosen experiments can dramatically narrow the search space.

Real-World Impact: Rosemount, MN

In partnership with Amrize (North America’s largest cement manufacturer), Mortenson, and the University of Illinois Urbana-Champaign, BOxCrete generated a concrete mix for the foundation slab of Meta’s Rosemount data center. The AI-optimized mix:

  • Reached full structural strength 43% faster than the original formula
  • Reduced cracking risk by nearly 10%
  • Used only domestically sourced materials

The mix is now qualified for use in additional areas of the data center.

Pennsylvania: Quadrel Integrates BOxCrete

Quadrel, a leading enterprise SaaS platform for the ready-mix industry, has adapted Meta’s AI framework into its daily quality control workflows. They use it for data preprocessing, batch normalization, feature engineering, and customer-specific model training. The models improve continuously as field test results are fed back in.

Data center construction site with concrete foundation and server racks in background Software Concept Art

Limitations and Cautions

While BOxCrete is a powerful accelerator, it doesn't replace physical lab validation, field trials, engineering sign-off, or building code compliance. The model's predictions reflect controlled lab conditions; actual field performance can differ due to batching variability, curing environment, and aggregate properties. Always treat AI suggestions as starting points, not final answers.

What's Next

Meta is releasing the foundational dataset used to develop the Rosemount mix—the best systematic open-source dataset for concrete mix performance to date. They’ve also submitted a paper on BOxCrete for publication. Over the next few years, they plan to collaborate further with the construction industry to develop new AI tools addressing broader sustainability challenges.

If you're a concrete producer or a software vendor serving the construction industry, this is a great time to experiment with BOxCrete. The model is available on GitHub under an MIT license.

Related Content

Data analysis dashboard comparing domestic and imported cement performance metrics Programming Illustration

Conclusion

Meta’s BOxCrete model is a textbook example of applied AI: narrow in scope, deeply grounded in domain expertise, and delivering measurable real-world outcomes. By reducing the time and cost of reformulating concrete with domestic materials, it helps American producers compete on cost, reduce emissions, and build supply chain resilience. The open-source release means this technology is no longer locked inside Meta—any producer or software vendor can adopt it.

Next step: Clone the BOxCrete GitHub repo and try running the interactive strength explorer on your own historical mix data.

This content was drafted using AI tools based on reliable sources, and has been reviewed by our editorial team before publication. It is not intended to replace professional advice.