Modern concrete lasts about 100 years. Roman concrete has lasted over 2,000. And the material we’ve built the modern world with is responsible for roughly 8% of global CO2 emissions — a problem that only compounds as cities rebuild after disasters like the January 2025 Los Angeles wildfires.
Watching those fires, researchers at USC’s Viterbi School of Engineering began asking a harder question: what if materials science could do radically better — not just stronger or greener, but both at once?
A material that builds problems as fast as buildings
Concrete is everywhere — in bridges, homes, roads, and the foundations of nearly every modern city. But ubiquity comes at a cost. Concrete production accounts for roughly 8% of global CO2 emissions, placing it among the most environmentally damaging industries on the planet. And despite that footprint, the material doesn’t last. Modern concrete has a lifespan of about 100 years, while Roman concrete has held together for over 2,000.
That gap has long troubled materials scientists. When the January 2025 Los Angeles wildfires swept through the region, USC professors Aiichiro Nakano and Ken-Ichi Nomura — collaborators for more than two decades — found themselves returning to that question with new urgency: could materials science produce something both more durable and less damaging to the climate? That conversation became Allegro-FM.
What Allegro-FM actually does
Allegro-FM is an AI-driven molecular simulation model developed at USC’s Viterbi School of Engineering, and its defining feature is scale. Conventional molecular simulation methods work with thousands or millions of atoms. Allegro-FM can simulate the behavior of over four billion atoms simultaneously — roughly 1,000 times larger than existing approaches. The model ran on the Aurora supercomputer at Argonne National Laboratory, achieving 97.5% efficiency at that scale.
It also covers 89 chemical elements, allowing it to predict atomic interactions across nearly the entire periodic table without requiring separate formulas for each element. That breadth makes it applicable well beyond concrete, spanning cement chemistry, carbon storage, and drug molecule research.
How AI replaced decades of quantum mechanics equations
Traditionally, simulating atomic behavior required researchers to derive precise quantum mechanics formulas from scratch. As Nomura described it, these were “profound, deep quantum mechanics phenomena” — painstaking, slow, and computationally expensive to produce for each new material or element combination.
Machine learning has substantially changed that process. Instead of deriving equations by hand, researchers now generate a training set and let the model learn how atoms interact on its own — faster, more flexible, and far more scalable. Allegro-FM can achieve what Nakano calls “quantum mechanical accuracy with much, much smaller computing resources.” That efficiency frees up supercomputer capacity for more complex downstream research, rather than consuming that power on calculations AI can now handle reliably.
Carbon-neutral concrete: from simulation to possibility
The most striking finding from Allegro-FM is also the most counterintuitive. Concrete production releases large amounts of CO2 — but the simulations suggest that CO2 doesn’t have to stay in the atmosphere. It can theoretically be recaptured and stored back inside the concrete itself.
“You can just put the CO2 inside the concrete, and then that makes a carbon-neutral concrete,” Nakano said.
The process does more than cut emissions. When CO2 is incorporated into concrete, it forms what researchers call a “carbonate layer,” which Nakano says makes the material more structurally robust — potentially pushing its lifespan well beyond the current 100-year average. The same intervention that addresses the carbon problem may also resolve the durability problem. Concrete is already fire-resistant, making it a practical rebuilding material for wildfire-prone cities like Los Angeles, and a carbon-neutral, longer-lasting version would be a meaningful step beyond what currently exists. Allegro-FM lets researchers test different concrete chemistries at the atomic scale before committing to costly real-world experiments.
What comes next for the research team
Nomura is direct about where the work goes from here. “We will certainly continue this concrete study research, making more complex geometries and surfaces,” he said. The team plans to model more intricate structural scenarios, pushing simulations closer to real-world application.
The broader ambition reaches well beyond concrete. Allegro-FM is designed as a foundation model — a platform for materials discovery across multiple domains, including drug molecules and carbon storage. The research was published in The Journal of Physical Chemistry Letters and selected as the journal’s cover image, a recognition of its significance to the field.
Proving something works at the molecular level is not the same as building it at city scale, and that gap remains the next major challenge. But for a material as old and as flawed as modern concrete, a credible simulation of something better is a meaningful place to start.
Carlos is an engineer with strong expertise in technical and industrial topics. He previously worked at international companies such as Siemens and speaks Spanish, German, English, and Italian.
