Meta's AI Optimizes Concrete for U.S. Production and Sustainability

AI Rebuilds the Foundation of American Concrete

Meta is leveraging artificial intelligence to tackle a critical challenge in U.S. construction: reliance on imported cement. Announced at the 2026 American Concrete Institute Spring Convention, the company has open-sourced a new AI model called Bayesian Optimization for Concrete (BOxCrete) alongside a foundational dataset. This initiative aims to help concrete producers rapidly design high-performance mixes using domestically sourced materials.

The U.S. construction industry pours roughly 400 million cubic yards of concrete annually. While most ready-mix concrete is produced domestically, nearly a quarter of the cement required is imported. This dependence impacts U.S. manufacturing jobs and exposes supply chains to volatility. Meta’s AI tools are designed to accelerate the reshoring of this critical material.

The Challenge: A Slow, Costly Design Process

Concrete mix design is a complex balancing act. Engineers must meet competing requirements for strength, workability, cost, curing speed, and now, sustainability and material origin. Traditional methods rely heavily on lab-based trial-and-error and accumulated expertise, making adaptation slow and expensive.

Furthermore, cement chemistry varies. A mix perfected for one cement source may fail with another. As highlighted in source material, U.S.-made cement must comply with stringent domestic performance and environmental standards not consistent internationally. This creates a significant barrier for producers wanting to switch to American materials.

BOxCrete: An AI-Powered Solution

BOxCrete improves upon Meta’s previous open-source concrete AI models. It is built on the company’s Adaptive Experimentation (Ax) platform, which uses Bayesian optimization. The AI doesn't replace lab validation but dramatically accelerates discovery.

The process begins by learning from historical mix designs and performance data. The model then intelligently proposes new formulations most likely to meet target specs while adhering to constraints like using specific U.S.-made ingredients. Each lab test result refines its predictions, creating a continuous improvement loop.

New features in BOxCrete include enhanced robustness to noisy real-world data and the ability to predict concrete slump, a key indicator of workability. Meta has also released what it calls the "best systematic foundational data" for concrete mix performance, based on the development of an award-winning mix used in its Rosemount, Minnesota data center.

Real-World Impact and Industry Partnerships

Meta’s approach is already yielding tangible results through strategic partnerships. A major collaborator is Amrize, the largest cement and concrete manufacturer in North America. Amrize has launched a "Made in America" cement label and announced nearly $1 billion in capital investments for 2026 to boost domestic production.

In Minnesota, an AI-optimized mix using U.S. materials was deployed at scale in a Meta data center slab. The mix achieved full structural strength 43% faster and reduced cracking risk by nearly 10% compared to the original formula. This successful validation opens the door for its use in other critical structural elements.

The open-source model, released under the MIT license, is also being commercialized. Pennsylvania-based Quadrel, an enterprise SaaS platform for the ready-mix industry, has integrated Meta’s AI framework into its software. It now informs daily quality control and mix design decisions for its customers.

A Broader Shift Toward Data-Driven Construction

Meta’s work aligns with a wider industry trend of using AI and machine learning to optimize construction materials. A separate study highlighted in the sources examined AI-driven design of high-performance fiber-reinforced concrete (FRC) using metakaolin, a sustainable cement alternative. This research underscores the dual focus on performance and environmental impact.

Furthermore, concepts like Agentic AI in manufacturing, as discussed by Atos, show how intelligent systems can monitor processes, guide operators, and feed learnings back into product lifecycle management (PLM) systems. This creates a similar closed-loop improvement cycle for quality and efficiency.

The Economic and Environmental Imperative

The push for domestic production isn't just about supply chain security. The cement and concrete sector contributes over $130 billion annually and supports roughly 600,000 U.S. jobs. Reshoring efforts have brought over 1.1 million jobs back to the U.S. since 2020, and manufacturing has a high economic multiplier effect.

Simultaneously, sustainability pressure is mounting. Cement production is a major source of global CO2 emissions. AI that enables efficient use of supplementary cementitious materials like slag or metakaolin, or that optimizes mixes to use less cement overall, directly addresses this challenge.

The Path Forward for AI in Construction

Meta’s near-term roadmap includes deeper collaboration with academia, such as the University of Illinois at Urbana-Champaign, to explore broader challenges in concrete sustainability. The goal is an industry-wide shift where AI-optimized mix design becomes a standard, accessible tool.

By reducing the time, cost, and risk associated with reformulating concrete around domestic and sustainable materials, AI has the potential to reshape American construction. It empowers producers to compete more effectively, reduce the carbon footprint of the built environment, and build greater resilience into national infrastructure—one optimized mix at a time.

Developers and researchers can explore Meta’s open-source BOxCrete model and the accompanying pre-print paper on GitHub and arXiv to understand and build upon this technology.