How a new AI model could help crushed rock fight climate change
Across vast stretches of farmland, something simple is being put to work against climate change: powdered rock.
Spread like fertilizer over agricultural fields, the rock reacts with rainwater and carbon dioxide in a technique known as enhanced rock weathering, which can speed up a natural process that removes CO₂ from the atmosphere to help limit further global warming. But there’s a catch: Figuring out which fields hold the most potential and how much rock reacts in the soil, one key input for estimating how much carbon could be removed, requires extensive and expensive measurements across enormous areas.
Carbon-removal company Terradot and Microsoft are using AI to tackle both challenges. They’ve combined a Microsoft weather model with Terradot’s scientific data to create TerraNova, a new AI model demonstrating an approach that could make it easier and less costly to remove carbon through enhanced rock weathering and eventually help expand the solution around the world.
“We’re going to need to remove nearly 10 billion tons of carbon dioxide from the atmosphere every single year” to avoid the worst effects of climate change, says Terradot co-founder and CEO James Kanoff. “The big problem is we don’t have many solutions that can operate at those scales today, let alone at a price that the world could really afford to pay.”
TerraNova builds on Aurora, Microsoft’s foundation AI model for predicting global weather and other environmental conditions. Terradot co-founder Sasankh Munukutla collaborated with experts from the Microsoft Climate Innovation Fund, AI for Good Lab and Microsoft Research to adapt Aurora to predict chemical changes in soil using data from fields where Terradot works with farmers — a significant new use for the technology.
The findings so far suggest that AI could help overcome some of the biggest practical obstacles to expanding enhanced rock weathering. The research will be described in a forthcoming preprint ahead of peer review. Earlier versions were accepted for a climate-focused workshop at NeurIPS 2026 and presented at a workshop on AI for science at ICLR 2026, two major AI research conferences.
“TerraNova is an exciting example of how AI can accelerate climate progress by addressing a real barrier to scaling carbon removal,” says Melanie Nakagawa, Microsoft’s chief sustainability officer. “It shows what’s possible when Microsoft brings together its technology, investment and partnerships to help promising solutions mature and, ultimately, build the markets needed for a more sustainable future.”
The stakes and the science
Reducing greenhouse gas emissions remains essential to slowing climate change. But scientists say that’s no longer enough. There’s already so much carbon dioxide in the atmosphere that the world also needs ways to remove it, and doing so in any region can benefit the planet as a whole.
Enhanced rock weathering is one potential solution that caught the attention of Munukutla and Kanoff, who met at Stanford University and founded Terradot in 2022.
Here’s how it works:
Scientists have long known that rainwater gradually weathers, or breaks down, certain rocks through chemical reactions that consume and trap carbon dioxide. That’s one of the major ways Earth’s climate regulates itself.
The problem is that the process ordinarily takes thousands of years.
That’s where enhanced rock weathering comes in. It speeds things up by crushing rocks into powder, exposing more surface area to humidity and allowing those natural chemical reactions to happen faster.
Terradot works with farmers to spread that powder over agricultural fields, where it removes carbon while crops continue to grow. An added benefit: Certain types of rock can also reduce soil acidity, akin to how farmers may apply limestone to control the pH of their fields. That can make nutrients more available to plants, potentially improving crop yields.
The challenge
How much carbon is removed depends on the type of rock, the soil’s chemistry and the rainfall, humidity and temperatures in the area. Conditions can vary even from one field to another, requiring scientists to collect large numbers of samples before and after spreading the rock, then analyze changes in the soil’s chemical composition. That takes specialized laboratory equipment, making it expensive to repeat across large areas.
Terradot’s team of about two dozen PhD-level scientists has been building robust data sets over the past few years by collecting tens of thousands of soil samples from fields in Brazil and the U.S. They’ve measured rock weathering rates, drilled wells to see where the water goes, taken geochemical measurements and more to analyze where enhanced rock weathering works best.
They’re “very complex measurements” that account for a substantial share of the cost, says Felipe Oviedo, a principal research scientist at Microsoft’s AI for Good Lab.
That’s a significant obstacle for a solution that needs to expand dramatically to make a difference in addressing climate change. But it’s of crucial importance to the companies and governments paying for carbon removal, who need credible evidence of the results.
Site selection is another challenge that requires complicated assessments of a host of factors above and below ground — and trial and error is expensive.
“To build confidence in enhanced rock weathering around the world, we will need to run intensive studies, from taking many samples in the soil to drilling wells down to the bedrock to track where the water goes,” Kanoff says. “We have to do that a couple times to learn from the data and then work with the academic community to apply what we learn more broadly.”
The innovation
Munukutla had worked on AI research at Stanford before co-founding Terradot and had always seen the technology as key to the company’s mission of expanding carbon removal.
While visiting a chocolate maker on Hawaii’s Big Island in 2024, he came across a display explaining how tropical soils are “fast-forwards” of soils elsewhere because the warm, wet climate accelerates natural weathering. That brought to mind Microsoft’s Aurora, an AI model introduced that year and trained on more than a million hours of atmospheric data to predict weather and other environmental conditions. Could it also be used to understand complex geochemical processes?
“We’ve seen AI models develop new abilities as they learn from more data,” Munukutla says. “My intuition was that Aurora could do the same.”
An early investor in Terradot, the Microsoft Climate Innovation Fund recognized the potential for enhanced rock weathering to remove large amounts of carbon. Looking for ways to speed up progress, the Fund worked with Terradot and experts across Microsoft to pursue the idea.
“We went for a moonshot together,” says Erika Basham, director of natural capital investments with the Microsoft Climate Innovation Fund. “We wanted to do more than invest in and purchase from Terradot. We wanted to help solve a problem holding back its growth, and by extension, the industry’s growth.”
The TerraNova model now analyzes factors including the soil composition of a field, minerals in crushed rock from nearby quarries, regional humidity, rainfall, temperatures and global weather patterns such as La Niña to predict carbon removal under different conditions.
That helps it identify locations where the process is likely to be most effective. For example, training TerraNova on data from fields in Brazil enabled it to make predictions for fields in Mississippi, says Oviedo.
The potential
Rather than replacing physical measurements, TerraNova’s approach combines predictions with a smaller number of soil samples to estimate results. In simulations using data from agricultural fields, researchers found that the model could reduce the number of samples needed at the largest sites by 75% while maintaining the same level of accuracy.
If further research confirms those results, that could be significant because measuring and verifying carbon removal can account for a third to more than half the cost of enhanced rock weathering.
Terradot says rock powder it has spread over fields in Brazil and the U.S. is expected to remove up to 100,000 tons of carbon dioxide over time. The company is aiming to remove one million tons annually by 2035.
While research has shown TerraNova’s potential, Munukutla says continued development and extensive validation are needed to use AI models like it responsibly. Terradot will continue building on the approach while working with the broader scientific community, he says.
“Every sample we take helps us improve the model, which means we can use it in more fields and eventually make carbon removal more affordable as we expand,” he says. “But what really lights me up is to see these novel capabilities emerge from models like Aurora. There are so many problems to solve at a planetary scale, and I think we’re just beginning to see what AI could help us do.”
Susanna Ray writes about AI and technology, with stories that show its real world impact and examine how innovation is reshaping work, business and society. She previously reported for Bloomberg News and other major international news organizations in the U.S. and abroad, covering beats ranging from politics and government to business and aviation. Follow her work on Microsoft Source.
