AI Briefing
KO

AI for American Cement and Concrete

·2026.03.31 01:00

Key point

Meta released BOxCrete, a model that optimizes concrete mix design, along with the source data behind it.

1 / 2

Details

Meta has released a new AI model called BOxCrete for concrete mix design, along with the source data used to build it. The goal is to make it easier to use American-made raw materials while maintaining performance and sustainability.

The US pours roughly 400 million cubic yards of concrete every year, but 20-25% of cement is imported. Traditional mix design has relied heavily on lab trial-and-error and engineers' experience, making it slow and costly, and because cement chemistry varies, a mix that works well with one material can fail with another.

Meta addresses this with adaptive experimentation.

  • Bayesian optimization efficiently explores possible mix combinations.
  • It learns from historical mix data, experimental results, and performance metrics to suggest promising candidates.
  • Users can specify constraints upfront, such as strength, workability, cost, and sustainability.
  • It creates an iterative improvement loop where the model gets more accurate as experimental results accumulate.

This release includes BOxCrete, which is more robust to noisy data than previous models, along with the foundational data behind the new concrete mixes used in Meta's data centers. Meta describes this data as the most systematic collection of publicly available concrete mix performance data, and has also submitted a related paper.

Field results are already in. Through collaboration with the University of Illinois Urbana-Champaign, Amrize, and Mortenson, an AI-optimized mix applied to part of a structural slab at a Rosemount, MN data center reached structural strength 43% faster than the conventional formula, and reduced cracking risk by nearly 10%. This mix is now ready to be applied to other areas as well.

Quadrel in Pennsylvania has integrated Meta's open-source framework into its SaaS, incorporating batch normalization, feature engineering, and customer-specific model training into its operations. Meta plans to continue collaborating with industry partners and academia to spread the adoption of concrete mix designs with lower emissions and stronger supply chain resilience.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.