AI for American-Produced Cement and Concrete

Meta is releasing BOxCrete, a Bayesian Optimization AI model for designing concrete mixes, alongside foundational dataset from its Rosemount, MN data center. The model uses Meta's Adaptive Experimentation (Ax) platform to intelligently navigate concrete formulations, learning from historical data and refining predictions with each lab test. In a real-world deployment, the AI-optimized mix achieved full structural strength 43% faster and reduced cracking risk by ~10% using domestically sourced U.S. materials. The open-source framework (MIT license) has already been adopted by Pennsylvania-based SaaS platform Quadrel for daily mix design workflows. Meta is partnering with Amrize and the University of Illinois Urbana-Champaign to expand adoption across the U.S. concrete industry.