Meta, Georgia Tech, Cusp AI Unveil Carbon Dataset

Meta’s Fundamental AI Research (FAIR) team has partnered with the Georgia Institute of Technology and AI startup Cusp AI to launch the Open Direct Air Capture (DAC) 2025 Dataset. This dataset is the largest of its kind and aims to transform carbon removal research and development by using artificial intelligence to speed up the discovery of sorbent materials used in Direct Air Capture.

The dataset includes over 100 million data points. It provides detailed simulation data on many potential DAC sorbents, materials that can chemically bind and remove carbon dioxide from the atmosphere. In the past, finding and testing such materials has been expensive and slow, needing extensive physical experimentation in labs. This new dataset changes that by letting researchers avoid these constraints and use AI models to simulate and predict sorbent properties. As a result, the need for extensive lab testing is greatly reduced, cutting research and development costs and speeding up innovation in carbon capture.

Central to this initiative is the use of machine learning techniques like graph neural networks and density functional theory (DFT) simulations. These methods allow AI models to understand complex molecular structures and predict how sorbent materials will perform in different conditions. The outcome is a tool that gives useful insights to scientists, engineers, and companies working on solutions to lower the energy demand and costs of DAC technology.

Meta highlighted how important is dataset for both academia and industry in its official statement. The new tool enables users to explore new combinations, improve existing ones, and speed up the commercialisation of carbon removal technologies. It works well with AI models, making it easy to incorporate into research processes and development workflows. It serves as a valuable resource for various stakeholders, from startups to large industrial companies.

It is mentioned that companies can use the data to find promising material candidates, optimise them for specific operational conditions, and develop in-house solutions for their carbon capture needs. These proprietary materials can also be licensed to others, creating new revenue streams and collaboration opportunities in the carbon removal market.

The International Energy Agency (IEA), in its 2023 World Energy Outlook, pointed out the key role of AI in lowering the cost of DAC technologies. The report stated that AI-driven innovations could help bring down the cost of carbon capture to under $100 per tonne by 2030. This milestone would make DAC economically feasible at scale. The Open DAC 2025 Dataset is seen as a foundational step toward this goal, especially given the urgent need to decarbonise hard-to-abate industries like cement, steel, and aviation.

The dataset includes data from simulations conducted in 2024, ensuring users have access to the latest and most relevant information. Its open-access nature supports broader scientific goals of transparency, reproducibility, and collaboration, allowing researchers worldwide to share knowledge and benefit from it.

Alongside its work on carbon capture, Meta is also investing in other sustainability efforts, such as using mass timber instead of steel and concrete for its data centers. This comprehensive approach to environmental responsibility reflects a growing awareness among tech companies of their role in fighting climate change, not only through reducing corporate emissions but also by creating technologies that can benefit society as a whole.

The collaboration between Meta, Georgia Tech, and Cusp AI demonstrates how partnerships across different sectors can leverage artificial intelligence to address some of the planet’s most pressing challenges. As the world strives to meet climate targets and expand carbon removal technologies, tools like the Open DAC 2025 Dataset could be crucial in connecting scientific research with real-world application.