Meta Releases BOxCrete AI Model for Concrete Mix Design

Meta's AI for Domestic Concrete Production
Meta has released Bayesian Optimization for Concrete (BOxCrete), a new AI model for designing concrete mixes that prioritize U.S.-produced materials and sustainability. The model is open source and available on GitHub, released alongside foundational data used to develop award-winning concrete mixes.
Problem Context
The U.S. pours approximately 400 million cubic yards of concrete annually, but imports about 20-25% of the cement required for it. Traditional concrete mix design relies on trial-and-error lab work and engineer intuition, which is slow and expensive to adapt. Different cements have different chemistries, requiring producers to rapidly explore and validate new formulations without months of lab work.
BOxCrete Features
- Improves over Meta's previous models with more robustness to noisy data
- Includes new features for predicting concrete slump (an important indicator of workability)
- Designed specifically for sustainable and domestically-produced concrete
Real-World Implementation
Meta has partnered with Amrize, the largest cement and concrete manufacturer in North America, which operates 18 cement plants, 141 cement terminals, and 269 ready-mix concrete sites. Amrize recently launched a Made in America cement label guaranteeing U.S. standards compliance and domestic manufacturing, and announced close to $1 billion in capital investments for 2026 to increase domestic cement production.
The partnership has already received recognition including a 2025 Building Innovation Award for Best Partnership and a 2025 Slag Cement Award for Sustainable Concrete Project of the Year, shared with the University of Illinois at Urbana-Champaign.
Economic Context
The cement and concrete sector contributes more than $130 billion annually to the U.S. economy and supports roughly 600,000 jobs. Manufacturing has a high economic multiplier, with every $1.00 spent adding $2.69 to the U.S. economy. Reshoring and related foreign direct investment have brought over 1.1 million jobs back to the U.S. since 2020.
📖 Read the full source: HN AI Agents
👀 See Also

Analysis of TB2 Benchmarking Issues in db-wal-recovery Task
A Reddit analysis reveals problems with Terminal Bench 2.0's db-wal-recovery task, where agents can accidentally destroy evidence by opening SQLite databases, and shows how prompt injection affects leaderboard results.

US Power Demand to Hit Record Highs in 2026–2027 Driven by AI and Data Centers
The U.S. Energy Information Administration (EIA) forecasts record-high power consumption in 2026–2027, primarily driven by surging AI workloads and data center expansion.

Bram Cohen critiques 'vibe coding' and AI-assisted development practices
Bram Cohen argues that 'vibe coding'—where developers avoid looking at code while using AI assistants—leads to poor software quality, using Claude's source code leak as an example of the problems with excessive dogfooding.

Grokipedia Stalled: No Accepted Edits Since April, Musk's AI Wikipedia Quietly Dies
xAI's Grokipedia—an AI-generated Wikipedia rival—hasn't accepted or rejected any of 225,496 suggested edits in over three months. Is it dead?