Meta’s AI cuts concrete carbon by 35% while boosting strength 43% faster

Meta and ready-mix supplier Amrize have deployed an AI-optimized concrete recipe at Meta’s Minnesota data center that reaches 4,000-psi strength 43% faster while cutting embodied carbon by 35%. The breakthrough demonstrates how Bayesian optimization can solve concrete’s carbon paradox—delivering both speed and sustainability for the trillion-dollar data center construction boom.

How it works: The collaboration used Meta’s open-source Ax and BoTorch frameworks combined with University of Illinois research to create a “bespoke” mixture design.

• Researchers fed hundreds of thousands of historic cylinder tests, aggregate gradations, and supplementary material chemistries into a Bayesian optimization engine that treated strength, shrinkage resistance, and carbon impact as a single problem.
• The algorithm converged after just 18 physical trials—about one-tenth the number required by conventional design methods.
• The winning recipe replaced 40% of portland cement with ground-granulated blast-furnace slag and Class F fly ash, using polycarboxylate admixtures to trigger rapid hydration.

In plain English: Think of this like teaching a computer to perfect a recipe by learning from thousands of previous attempts. Instead of trial-and-error testing that might require 200 concrete samples, the AI system analyzed historical data to predict the best mix after just 18 tries. The final recipe swaps out nearly half the traditional cement (the most carbon-intensive ingredient) with industrial byproducts like slag from steel production and fly ash from coal plants, while adding chemical accelerators to make it set faster.

The big picture: Cement production accounts for roughly 8% of global carbon dioxide emissions, creating a sustainability challenge as data center construction accelerates toward a projected $1 trillion in global capital expenditure by 2029.

Key performance metrics: Full-scale testing at the 715,000-square-foot Rosemount campus validated the lab results.

• The EcoPact mix reached target strength 43% faster than baseline concrete, enabling follow-on trades to advance rack installation schedules.
• Drying shrinkage decreased 9.5% compared to standard mixes.
• Cost increases were “not substantial,” according to Amrize, aligning with Rocky Mountain Institute research showing 19%-46% carbon reductions can be achieved with less than 1% cost increases.

Who else is involved: The partnership spans Silicon Valley tech, traditional materials manufacturing, and academic research.

• Mortenson Data Center Solutions serves as the contractor, with campus director Blair McNeil noting the project brings “the use of AI into concrete mix design” to optimize sustainability and constructability.
• University of Illinois Urbana-Champaign professor Nishant Garg supervised testing and emphasized that “AI-driven mix design lets us optimize concrete for performance, cost and carbon in a single step.”

Why this matters: The success comes as construction’s slow digital adoption threatens data center delivery timelines, according to a June white paper by Revizto, a cloud-based building information modeling collaborative.

• Meta’s procurement strategy now requires suppliers to publish third-party-verified Environmental Product Declarations and favors performance-based specifications over prescriptive cement limits.
• The company has open-sourced its “SustainableConcrete” code on GitHub to accelerate industry adoption.

Scaling challenges: The approach faces material supply constraints that could limit widespread adoption.

• The American Cement Association warns of a potential 50-million-ton shortfall of supplementary cementitious materials by 2050—roughly half of today’s consumption.
• Amrize is piloting calcined-clay and harvested-ash blends to diversify supply, with the Bayesian optimizer remaining “agnostic to the binder, so long as we have the right input data,” according to senior VP Rodrigo Gallardo.

What’s next: Standards bodies are beginning to accommodate AI-optimized concrete designs.

• ACI Committee 318 and ASTM Committee C09 have discussed draft language that may allow performance-verified mixes to qualify under structural codes, with the earliest formal ballot expected in 2026.
• Long-term durability testing is underway at UIUC, with 56- and 90-day sulfate-resistance results due this fall.

What they’re saying: Industry leaders see the pilot as proof that multiple innovation levers can work simultaneously.

• “Owners are asking for carbon disclosures, tech vendors are opening their code and the codes themselves are moving toward performance,” said Gallardo. “You need all three levers to pull at once; Rosemount proves that can happen.”
• Jaime Hill, president of Amrize Building Materials, added that “Rosemount is just the beginning,” noting digital design allows bespoke concretes to target applications from thermal-mass management to radiation shielding.