Amid global pressure on cement‑industry carbon emissions, a joint team from the Swiss Federal Institute of Technology Lausanne (EPFL) and the Paul Scherrer Institute (PSI) on 18 June 2025 (CET) announced a breakthrough: using AI algorithms to redesign cement formulations in milliseconds, reducing carbon emissions by up to 80% while maintaining material strength and industrial applicability. The study simulated thousands of ingredient combinations and optimized the cement's core component—the clinker—replacing some high‑carbon materials with low‑carbon substitutes such as volcanic ash and limestone. ScienceDaily reported that this innovation could save millions of tonnes of CO2 annually and inject sustainability into the construction sector, with estimates suggesting a 30% reduction in the global cement carbon footprint by 2030. Ahead of COP30, the result has been hailed as an exemplar of an "AI green revolution" and has drawn attention from major building‑materials players such as LafargeHolcim.
Cement production is the world's second‑largest industrial source of carbon, accounting for about 8% of total emissions and releasing roughly 4 billion tonnes of CO2 each year, mainly from the high‑temperature calcination of clinker (around 1450°C). Traditional formula optimization relies on laboratory testing that can take months and incur high costs. The Swiss team's AI approach overturns this paradigm: they developed a machine‑learning simulation platform capable of testing tens of thousands of recipe variants in seconds, evaluating strength, durability and carbon footprint. Research lead Emma Brun, a professor of materials science at EPFL, emphasized, "AI is not a simple replacement for humans but an accelerator of innovation—we found low‑carbon formulations that would have taken years to discover with traditional methods." For example, a new formulation reduces clinker content from 90% to 20%, substitutes industrial byproducts such as fly ash, and lowers emissions from about 900 kg CO2 per tonne to 180 kg, while maintaining a compressive strength of 50 MPa that meets international standards.
The breakthrough stems from combining Generative Adversarial Networks (GANs) with Reinforcement Learning (RL). The team trained models on terabytes of materials data, including historical recipes and environmental datasets; the GAN generates candidate formulations and the RL agent iteratively optimizes them using a reward function that balances strength and low carbon. Cosmos reported that the model is roughly 1,000 times more computationally efficient than traditional simulations and can run on standard GPUs, making it accessible to small and medium enterprises. In laboratory validation, the new formulations passed the EU EN 197 standard tests, showing durability comparable to conventional cement. Forbes notes that the technology can be integrated directly into production lines to enable real‑time AI adjustments, potentially delivering a further 20% carbon reduction.
The Swiss research has far‑reaching global implications. Cement emissions represent about 90% of the construction sector's carbon output, while construction contributes some 13% of global GDP. Earth.Org observes that if cement plants worldwide adopt AI optimization, cumulative reductions of around 10 billion tonnes of CO2 could be achieved by 2050—equivalent to the EU's annual emissions. The team has partnered with Swiss cement giant Holcim and plans a pilot production line in 2026. Rinnovabili reports the approach can be extended to concrete and mortar, potentially cutting the overall building‑materials carbon footprint by 50%. In developing countries such as India and Brazil, the AI tools could reduce costs (formula optimization can save about 10% in raw materials) and accelerate the transition to net‑zero construction.
However, challenges remain. Interesting Engineering stresses that AI models depend on high‑quality data; if training datasets are biased toward European materials, the models may not generalize to Asian geology. Industrial deployment also requires supply‑chain adjustments, with initial payback periods expected to be around 2–3 years. The EPFL team is developing an open‑source version of the platform and inviting global researchers to contribute data to promote equitable application.
This research is not only a technical victory but also an example of AI‑enabled climate action. At a press conference Brun said, "Cement used to be a 'grey giant'; AI is turning it green." With progress toward the UN Sustainable Development Goals (SDG 9 and 13), Switzerland's "AI cement revolution" could become a global template to help achieve carbon neutrality by 2050. The building‑materials industry may be entering a period of transformation in which AI shifts from an auxiliary tool to the engine reshaping the future.