AI and Materials Science Forge a Greener Future for Concrete

Concrete underlies nearly every modern city’s infrastructure—from skyscrapers and highways to bridges and sidewalks. Yet its primary binding agent, Portland cement, accounts for an estimated 7–8 percent of global carbon dioxide emissions. As governments, companies, and researchers pursue decarbonization strategies, reimagining concrete’s composition emerges as a critical frontier. A recent collaboration between the Olivetti Group and MIT’s Concrete Sustainability Hub (CSHub) has turned to artificial intelligence to tackle the vast search for alternative cementitious materials. Their results, published in Nature Communications Materials on May 17, outline a data-driven framework that combs through scientific literature and geochemical databases to identify viable replacements for cement. By pairing AI-powered text mining with experimental insights, the team has charted a path toward more sustainable, circular concrete that could dramatically cut emissions and costs—without compromising durability or strength.

I. The Urgency of Cement Replacement
A. Concrete’s Carbon Footprint
Concrete ranks as the world’s most widely used engineered material, with annual production exceeding 10 billion tons. Each ton of Portland cement, however, generates roughly 0.8 tons of carbon dioxide, thanks to the calcination process that transforms limestone into clinker. As demand for housing and infrastructure continues to rise—especially in emerging economies—cement-related emissions are projected to grow unless alternatives scale rapidly. While electrification of kilns and carbon capture technologies offer partial mitigation, fundamental shifts in concrete’s formulation promise steeper reductions.

B. Traditional Supplementary Cementitious Materials and Their Limits
For decades, the construction industry has incorporated byproducts like fly ash (from coal combustion) and ground granulated blast-furnace slag (from steel production) into concrete mixtures. These supplementary cementitious materials (SCMs) partially replace Portland cement, cutting clinker content and lowering embodied carbon. In many regions, however, supply constraints have emerged: aging coal plants and steel mills produce declining quantities of fly ash and slag. Worse yet, fly ash availability has plummeted in countries phasing out coal-fired power. As steel demand booms in Asia and other regions, slag supplies face similar pressures. Hence small supply networks cannot support the global demand for high-volume SCM use, spurring an urgent search for alternative “cement extenders.”

C. The Proliferation of Candidate Materials
Researchers have long examined a diverse array of pozzolanic materials—substances that react with calcium hydroxide (a byproduct of cement hydration) to form additional calcium silicate hydrates, boosting strength and durability. Potential candidates range from volcanic ashes and natural pozzolans (e.g., calcined clays) to industrial wastes like rice husk ash, sugarcane bagasse ash, and mine tailings. The challenge: the sheer number of possibilities. Scattered across decades of scientific studies—often in disparate subfields of geology, materials science, and civil engineering—data on chemical composition, particle size, reactivity, and performance properties can span hundreds of thousands of pages. Manually reviewing each candidate and synthesizing viable mixtures becomes an overwhelming task.

II. Employing Artificial Intelligence for Material Discovery
A. The Rationale for Machine Learning
Recognizing that speed and scale are paramount, postdoctoral researcher Soroush Mahjoubi and his colleagues sought leverage from artificial intelligence—specifically large language models (LLMs) and machine-learning frameworks. “There’s so much data out there on potential materials—hundreds of thousands of pages of scientific literature,” Mahjoubi observes. “Sorting through them manually would take lifetimes, by which time new materials would appear.” AI offers a systematic approach: algorithms can rapidly ingest, classify, and rank candidate SCMs based on predetermined performance criteria.

B. Constructing the AI-Driven Framework
The research team began by compiling a vast corpus of scientific articles—peer-reviewed journals, conference proceedings, and technical reports—alongside geochemical data from over one million rock and soil samples worldwide. Using natural language processing (NLP) tools, LLMs parsed textual descriptions of candidate materials, extracting key features such as chemical oxide composition (SiO₂, Al₂O₃, Fe₂O₃, CaO, MgO, etc.), mineralogical phases (quartz, feldspar, clays, zeolites), and published measurements of pozzolanic reactivity or hydraulic performance. Simultaneously, a geospatial database detailed the global distribution of raw materials—clay deposits, ceramic waste sites, and mining tailing volumes—enabling assessment of logistical feasibility and regional availability.

C. Key Selection Criteria: Hydraulic Reactivity and Pozzolanicity
Two fundamental properties govern a substitute’s viability: hydraulic reactivity and pozzolanic reactivity.

1. Hydraulic Reactivity: Portland cement hardens when mixed with water—a process governed by the hydration of tricalcium silicate (C₃S) and dicalcium silicate (C₂S). An effective replacement must replicate this mechanism or supply reactive silica and alumina that coalesces into calcium silicate hydrate (C–S–H) gel. The AI framework quantified hydraulic potential by evaluating materials’ crystalline phases (e.g., belite, ye’elimite, or rankinite) and particle fineness—parameters associated with early-age strength development.
2. Pozzolanicity: Some materials—historically used in Roman and Byzantine structures—react with calcium hydroxide (CH) liberated during cement hydration. This secondary reaction produces additional C–S–H, promoting long-term strength and durability. Pozzolanicity often correlates with amorphous silica content; amorphous phases dissolve more readily than stable crystalline forms. The model flagged materials with high reactive amorphous content—such as certain calcined clays, volcanic ash analogues, or ceramic fragments—for further evaluation.

III. From Data to 19 Material Categories
A. Clustering and Categorization
With performance metrics quantified, the AI pipeline grouped thousands of candidates into 19 material categories. Clustering algorithms considered geochemical fingerprint (major and minor oxide composition), mineralogy, and historical performance data where available. Categories spanned a wide spectrum:

• Calcined Clays (metakaolin, kaolinite-rich mudstones)
• Rice Husk Ash and Agricultural Wastes (bagasse, coconut shell ash)
• Ceramics and Demolished Construction Debris (bricks, roof tiles, pottery shards)
• Mining and Quarrying Byproducts (bauxite residue, nickel laterite tailings, copper slag, iron blast-furnace slag)
• Volcanic Tuffs and Natural Pozzolans (tuff, pumice, perlite)
• Engineered Glass Precursors (waste glass cullet, cathode ray tube fragments)
• Slag Variants Beyond Blast-Furnace Slag (steel ladle slag, tin smelting slag)
  Such a taxonomy not only guides laboratory testing but also highlights regional specialties: for example, bauxite residue (red mud) is abundant in Australia and China, while ceramic tile waste may predominate in Europe due to older building stock.

B. Global Availability Mapping
Geospatial overlays linked categories to regional extraction and production hotspots. For instance, the database identifies that:

• Agricultural Wastes: Southeast Asia (rice belt) produces millions of tons of rice husk ash annually.
• Ceramic Waste: European Union building demolition yields significant volumes of tile and brick fragments.
• Mining Tailings: Latin America and Australia host major copper and nickel operations, generating tailings with moderate silica-alumina content.
  By assigning tonnage estimates to each category’s global deposits, the team could forecast realistic substitution rates. In many countries, ceramics alone could replace up to 10–15 percent of cement content without extensive processing—simply requiring crushing and grinding to achieve targeted fineness.

IV. Promising Candidates and Preliminary Findings
A. Ceramics: Ancient Wisdom, Modern Application
One of the most intriguing insights involves ceramics. Ancient Roman engineers used crushed pottery and volcanic ash to produce hydraulic mortars that remain intact after two millennia. Modern analyses confirm that brick dust and tile fragments exhibit pozzolanic reactivity when ground below 20 µm median particle size. The AI framework identified specific ceramic compositions—rich in metakaolin and amorphous silica—that rival commercial metakaolin in reactivity. Initial laboratory trials demonstrate that replacing 10–20 percent of cement with ceramic waste can achieve 28-day compressive strengths within 5 percent of control mixes, while steadily reducing embodied carbon by up to 40 percent per cubic meter of concrete.

B. Calcined Clays and Natural Pozzolans
Calcined clays (e.g., metakaolin and processed illite) scored highly in the AI screening. Kaolinite-rich formations—abundant in parts of India, China, and the southeastern United States—can be thermally activated at relatively low temperatures (650–800 °C), unlike clinker production (1,400–1,500 °C). The result: dramatic CO₂ savings. The team’s geospatial analysis confirms that nearly half of the world’s cement-producing regions sit within 200 km of viable calcined clay deposits, suggesting quick scalability. Additionally, natural pozzolans—such as pumice, volcanic ash, and tuff—found in Italy, Japan, and Central America, were flagged as high-potential supplementary materials, especially when processed to sub-10 µm particle sizes.

READ MORE: Feeding Robots Could Solve Energy Shortage Issue

C. Mining Tailings: Transforming Waste to Resource
Iron blast-furnace slag and fly ash from steel mills have proven effective SCMs for decades. But lesser-known mining tailings—residuals from gold, copper, and nickel extraction—often comprise fine-grained silicates ideal for pozzolanic reactions. The AI model pinpointed copper tailings in Chile and Peru, as well as nickel laterite residue in Indonesia and the Philippines, as sources containing 30–50 percent reactive silica-alumina. By repurposing these tailings—typically stockpiled in environmental liabilities—concrete producers can reduce both waste and cement demand. Pilot tests show that replacing up to 15 percent of clinker with ground copper tailings yields comparable early strength gains to slag-blended mixes, while improving sulfate and chloride resistance.

D. Waste Glass and Engineered Glass Compositions
Although glass cullet recycling for container manufacturing has matured, post-consumer glass still sources suboptimal streams (e.g., CRTs, tempered safety glass, photovoltaic panels). The AI framework evaluated glass compositions high in Na₂O and SiO₂—attributes enabling them to react pozzolanically when finely milled. Laboratory results reveal that ground architectural glass (beyond conventional beverage bottles) can achieve significant reactivity when activated by alkaline solutions. This “alkali activation” route parallels geopolymer chemistry, raising the prospect of partially replacing cement with waste glass in geopolymer or hybrid cement formulations.

V. From AI Predictions to Experimental Validation
A. Prioritizing Top Candidates
With the 19 categories mapped and ranked, the team began narrowing down to a manageable shortlist for experimental testing. Selection criteria included:

1. High Pozzolanicity Index (amorphous content > 40 percent).
2. Ease of Processing (low-energy mechanical grinding).
3. Regional Abundance (logistical feasibility within 200-km radius of major concrete plants).
4. Cost Competitiveness (waste or low-value streams).

Pilot experiments at the CSHub laboratories focused on:

• Crushed Brick and Tile: Sourced from local demolition debris. Ground to median particle size of 8 µm, blended at 10 percent replacement levels.
• Calcined Illite Clay: Excavated from geologically mapped deposits in North Carolina, thermally activated at 700 °C, ground to sub-5 µm.
• Copper Tailings: Obtained from a Chilean mine tailing repository. Milled and assessed for mineralogical phases.
• Sugarcane Bagasse Ash: Targeted as a representative agricultural waste, incinerated at 600 °C and milled.

B. Performance Metrics
Key performance tests included:

1. Compressive Strength: Standard 28-day cylinder tests (ASTM C39).
2. Chloride Penetration Resistance: Rapid chloride permeability tests (ASTM C1202).
3. Sulphate Attack Resistance: Immersion in sodium sulfate solution per ASTM C1012.
4. Heat of Hydration: Calorimetric analysis to ensure compatibility with mass concreting applications.
5. Carbonation Rate: Assessment of CO₂ ingress over 90 days in accelerated carbonation chambers.

C. Experimental Findings to Date
Initial results show encouraging trends:

• Brick and Tile Blends: Concrete with 15 percent ceramic waste replacement achieved 28-day compressive strengths within 94 percent of control samples. Chloride migration reduced by 20 percent, attributed to refined pore structure. Sulfate resistance also improved due to lower CH content.
• Calcined Illite Clay: At 20 percent replacement, early-age strength (7 days) matched control specimens, while 28-day strength exceeded controls by 5 percent. The fine calcined clay enhanced particle packing—boosting microstructural refinement. Carbonation depth reduced by 30 percent.
• Copper Tailings: Blends with 10 percent tailings exhibited 28-day strength at 90 percent of baseline, but with notable gains in chloride resistance, likely due to iron-oxide phases refining porosity. Sulfate resistance remained comparable to controls.
• Sugarcane Bagasse Ash: At 15 percent replacement, samples gained strength more slowly in the first week (70 percent of control at 7 days) but reached 95 percent of control by 28 days. High silica content contributed to lasting pozzolanic reactions, though thermal processing costs present logistical hurdles.

VI. Scaling Considerations and Industry Adoption
A. Processing and Quality Control
Although many candidates require only grinding, achieving narrow particle size distributions (e.g., median diameter < 10 µm) demands energy and specialized milling equipment. For ceramic waste—often heterogenous—magnetic and density separation may be needed to remove contaminants (metals, plastics). Strict quality control protocols must be established to ensure consistent reactivity: periodic chemical assays (XRF), mineralogical scans (XRD), and pozzolanic reactivity tests (Frattini test or Chapelle test).

B. Regulatory Hurdles and Standards
In most jurisdictions, SCMs fall under existing concrete standards (ASTM C618 in the U.S., EN 197 in Europe), which specify allowable chemical and physical properties. Introducing novel materials—such as mining tailings—will require extended validation protocols, including long-term durability studies, freeze–thaw tests, and alkali–silica reaction (ASR) evaluations. Standards committees must adapt quickly to incorporate AI-identified candidates, balancing safety with the urgency for low-carbon solutions.

C. Supply Chain and Logistics
Deploying regionally abundant materials demands coordination among concrete producers, local governments, and waste generators. For example, aggregate quarries could partner with ceramic recyclers to establish centralized grinding facilities. Former brickworks or tile manufacturers might diversify into SCM production, creating circular loops that reduce landfill usage. Mining companies could valorize tailings by investing in on-site milling plants; in return, concrete producers secure low-cost, local raw materials. The MIT study’s geospatial mapping highlights where such synergies are most feasible.

D. Economic Impacts and Cost Analysis
Substituting 10–20 percent of Portland cement with lower-cost waste or natural materials can reduce material costs by 5–15 percent per cubic meter of concrete. Moreover, many SCMs are essentially free at the gate—waste ceramics, mine tailings, and agricultural ashes often carry negative disposal costs. Lower clinker demand reduces fuel and limestone consumption, saving 200–300 kg CO₂ per ton of cement replaced. Lifecycle assessments (LCAs) project that a 20 percent cement replacement can cut overall concrete emissions by up to 25 percent, a substantial gain at scale.

VII. Advancing toward Circular Concrete
A. The Circular Economy Imperative
In linear economic models, construction and demolition waste typically flows to landfills, squandering embodied energy and materials. By identifying high-performance SCMs from demolished structures—bricks, tiles, ceramics—civil engineers can close material loops. The MIT team underscores that repurposing ceramics not only cuts raw material demand but also alleviates landfill pressure. Similarly, mining tailings—often stored in vast impoundments—can be converted from environmental liabilities to assets. This aligns with circular economy principles: maintain products and materials in use as long as possible, regenerate natural systems, and decouple growth from resource consumption.

B. Opportunities for Rebuilding War-Torn Infrastructure
Beyond climate considerations, countries facing reconstruction—such as regions of Ukraine and parts of the Global South—stand to benefit from low-cost SCMs. War-damaged areas generate enormous construction debris: rubble from collapsed buildings, broken bricks, and ceramic floor tiles. By implementing on-site recycling programs—crushing, grading, and testing for reactivity—reconstruction efforts can reduce reliance on imported cement, expedite rebuilding, and lower greenhouse gas footprints.

C. Synergies with Geopolymer Concrete and Low-Carbon Cements
While the MIT study focuses on Portland-cement-based mixtures enhanced with SCMs, broader research into geopolymer and blended cements offers complementary pathways. Geopolymers—alkali-activated aluminosilicates—use fly ash or slag with alkaline activators, replacing cement entirely. Combining AI-identified SCMs with geopolymer chemistries may yield hybrid binders that meet demanding performance requirements while slashing emissions by up to 80 percent. Such synergies suggest a future in which multiple alternative approaches converge, giving concrete producers a toolkit of low-carbon solutions tailored to local resources.

VIII. Next Steps: Experimental Validation and Model Refinement
A. Scaling Laboratory to Field Trials
Having identified and tentatively tested numerous candidates, the research team plans larger-scale pilot projects with industry partners. These trials will produce cubic meters of concrete in realistic production settings—ready-mix plants and precast factories—assessing workability, pumping characteristics, set times, and finishing properties. Durability assessments, such as freeze–thaw cycling, sulfate immersion, and long-term creep tests, will be conducted over two to five years to ensure that AI predictions translate into field performance.

B. Enhancing the AI Framework
The current model evaluates materials on two axes—hydraulic reactivity and pozzolanicity—but future iterations aim to incorporate additional parameters:

• Life-Cycle Embodied Energy: Incorporating cradle-to-gate energy data for each candidate.
• Leachate and Toxicity Assessments: Screening for heavy metals or harmful leachates in mining waste or industrial byproducts.
• Cost-Benefit Optimization: Integrating real-time market data on waste disposal fees, transportation costs, and energy prices.
• Climate Risk Adaptation: Evaluating materials’ performance under extreme weather projections—e.g., high salinity, elevated temperatures, or flash flooding.

By layering these criteria, the framework will narrow its focus to candidates that not only perform technically but also align with economic, environmental, and regulatory requirements worldwide.

C. Collaboration with Global Stakeholders
To mobilize findings at scale, the CSHub and Olivetti Group plan to engage material suppliers, concrete producers, policymakers, and standardization bodies. Workshops and collaborative consortia will convene stakeholders from regions such as India, Brazil, and sub-Saharan Africa—areas experiencing rapid urban growth and cement demand. The goal: create open-access repositories of AI-validated material recipes, region-specific guidelines, and case studies to accelerate adoption.

IX. Broader Implications and Future Outlook
A. Revolutionizing Sustainable Construction
The marriage of AI and materials science demonstrates that solving grand challenges—such as cement decarbonization—requires interdisciplinary approaches. By automating tedious literature reviews and data sifting, the AI-driven framework frees researchers to focus on experimental validation and scale-up. As more low-carbon SCMs move from lab to production line, the construction industry stands on the cusp of a revolution: every ton of alternative binder deployed translates to significant emissions avoided. Over the next decade, mainstreaming these practices could shave up to 20 percent off global concrete emissions—equivalent to taking hundreds of millions of cars off the road.

B. Democratizing Material Innovation
Historically, cement R&D has been dominated by large multinational corporations with extensive laboratory infrastructure and deep pockets. The AI-based approach democratizes discovery: smaller research labs and universities can leverage open-source code and public geochemical datasets to explore regional materials. Emerging innovators—startups, social enterprises, and community-driven groups—can tailor sustainable concrete solutions to local resource streams, spurring decentralized manufacturing and job creation.

C. Toward a Circular Built Environment
Beyond emissions, repurposing waste materials in concrete fosters a more circular built environment. Every demolished brick or discarded roof tile becomes feedstock for the next century’s infrastructure, closing resource loops and mitigating landfill pressures. In parallel, modular construction techniques—combining prefabricated low-carbon concrete panels—can accelerate building projects while ensuring high performance and recyclability at end-of-life. In this paradigm, urban landscapes evolve from resource drains into resource reservoirs, harmonizing development with environmental stewardship.

D. Interdisciplinary Lessons for Sustainability
Finally, the MIT–Olivetti study exemplifies how AI can accelerate sustainability across industries. Similar frameworks may emerge to optimize steel alloys, plastics recycling, battery chemistries, or agricultural inputs—areas where abundant data exists but human analysis alone cannot keep pace with evolving research. By integrating LLMs, machine vision, and big data analytics, scientists and engineers can identify levers for near-term impact, bridging the gap between theoretical promise and practical application.

Conclusion: A New Era for Concrete
Concrete’s ubiquity—and its hidden carbon burden—have long posed a formidable barrier to decarbonizing the built environment. Yet the collaboration between MIT’s Concrete Sustainability Hub and the Olivetti Group has shown that artificial intelligence can transform a once intractable search for alternative cementitious materials into a tractable engineering problem. By screening millions of data points, categorizing candidates into 19 material groups, and validating promising contenders in the lab, the team has charted a roadmap toward greener, circular concrete.

From pulverized ceramics and calcined clays to agricultural ashes and mining tailings, region-specific feedstocks now stand ready for further testing and scaled deployment. As governments tighten emissions regulations and investors demand sustainable building practices, AI-driven material discovery positions the construction industry to meet climate targets. In an era defined by technological innovation and environmental urgency, harnessing AI to reinvent concrete’s recipe may prove a decisive stride toward net-zero infrastructure—reshaping not only our cities, but the relationship between human ingenuity and the planet’s finite resources.