AI Finds Low-Carbon Cement Recipes in Seconds, Slashing Emissions Without Sacrificing Strength

The AI that writes climate-friendly cement recipes in seconds

Date: June 19, 2025
Source: Paul Scherrer Institute

Summary

Researchers in Switzerland have developed an AI system that can rapidly design new cement formulations. This technology evaluates thousands of ingredient combinations to find mixes that maintain cement strength while drastically reducing CO2 emissions — all within seconds.

Reducing Cement’s Carbon Footprint with AI

Cement, when mixed with water, sand, and gravel, forms concrete—the most widely used building material globally. Yet, cement production is a major CO2 emitter, accounting for roughly 8% of global emissions, surpassing the entire aviation sector.

The Paul Scherrer Institute (PSI) has created an AI-driven model that accelerates the search for sustainable cement recipes. These recipes aim to preserve mechanical performance while lowering carbon emissions significantly.

Producing clinker, the main raw material for cement, involves heating rotary kilns to about 1,400°C, a process that requires energy-intensive combustion. Surprisingly, the main source of CO2 is not just the combustion but also chemical CO2 released from limestone during clinker production.

Optimizing Cement Recipes with Machine Learning

One effective strategy to reduce emissions is altering the cement recipe by partially replacing clinker with alternative cementitious materials. The PSI team’s approach leverages machine learning to simulate and optimize thousands of formulations quickly, eliminating the need for slow, costly experiments.

"Instead of testing thousands of variations in the lab, our model generates practical recipe suggestions within seconds—like a digital cookbook for climate-friendly cement," explains mathematician Romana Boiger.

By filtering through vast composition possibilities, the AI identifies candidate formulations that meet criteria for strength and carbon footprint. This method drastically speeds up development cycles, focusing lab tests on the most promising options. The study is published in Materials and Structures.

Balancing Material Availability and Quality

Currently, industrial by-products like slag and fly ash partially replace clinker to reduce emissions. However, the enormous global demand for cement means these substitutes alone cannot meet needs.

"We need the right combination of widely available materials that can produce high-quality cement," says John Provis, head of the Cement Systems Research Group at PSI.

Cement acts as a mineral binder in concrete, creating artificial minerals that hold the material together. Modeling this process accurately is complex and computationally demanding, which is why the team turned to AI.

AI as a Computational Accelerator

The team trained artificial neural networks using data generated from the open-source thermodynamic software GEMS, developed at PSI. This software models mineral formation and geochemical reactions during cement hardening.

By combining these results with experimental and mechanical data, the AI learned to predict mechanical properties and CO2 emissions for any given recipe. While traditional modeling takes minutes or hours, the trained neural network delivers results in milliseconds—about 1,000 times faster.

Finding the Optimal Cement Formulation

Instead of randomly testing recipes, the researchers reversed the problem: determining which recipe achieves specified CO2 and mechanical property targets.

This mathematical optimization simultaneously maximizes material strength and minimizes emissions. To solve this, they incorporated genetic algorithms—methods inspired by natural selection—to efficiently identify ideal formulations.

This reverse approach avoids blind trial and error, directly targeting formulations that meet set requirements.

Interdisciplinary Collaboration Yields Promising Candidates

Several AI-identified cement recipes show strong potential for reducing emissions while maintaining quality and feasibility.

"These formulations still require laboratory validation before industrial use," notes Nikolaos Prasianakis, head of the Transport Mechanisms Research Group at PSI, emphasizing caution.

The project serves as a proof of concept that AI and mathematical modeling can guide cement development. Researchers plan to extend the model to consider factors like raw material supply and environmental conditions, such as marine or desert applications.

This work was made possible by collaboration between cement chemists, thermodynamics experts, and AI specialists, as well as partnerships with institutions like EMPA within the SCENE project, which focuses on solutions for net zero emissions in industry.

Conclusion

This AI-driven approach dramatically reduces the time required to create sustainable cement recipes, offering a scalable path for lowering the construction sector’s carbon footprint. It represents a practical example of how AI can accelerate materials science breakthroughs with real-world environmental impact.