Brainstorming Materials, Simulating Tests, Optimizing Supply Chains and More: How AI is Revolutionizing Carbon Capture

It has been widely recognized that achieving ambitious climate targets will require deep decarbonization, going beyond known pathways such as renewable energy and energy efficiency. Although by no means an alternative to cutting back emissions, carbon capture utilization and storage (CCUS)–particularly important for tackling emissions from hard-to-abate sectors–will necessarily form part of a comprehensive strategy for reaching net zero. 

The IEA’s estimates place the amount of CO2 that will need to be captured annually by 2050 according to the Net Zero by 2050 Scenario at 1 gigaton. There are currently only around 45 commercial facilities operating globally, with a total capture capacity of roughly 50 megatons. Scaling carbon capture to the necessary level faces significant obstacles, namely, immense energy requirements and high costs, highlighting a dire need for innovation. Today, artificial intelligence has stepped in to fill that need, unlocking vast capabilities in the generation and testing of new materials to develop cost-effective CCUS technologies, as well as gaining insights into the implementation and deployment of CCUS within energy systems.

AI’s Role in the Search for New Materials

Carbon capture can be grouped into four principle methods: post-combustion (the most traditional), pre-combustion, oxy-fuel combustion, and direct air capture (DAC). The first three each involve capturing CO2 from the exhaust gas of fossil fuel combustion, whereas the latter captures CO2 directly from the atmosphere. A plethora of technologies have been developed for combustion-based capture, the predominant ones being absorbent (aka solvent-based), adsorbent, in which CO2 molecules bind to a solid sorbent material such as zeolites or metal organic frameworks (MOFs), and membrane separation, in which CO2 molecules selectively permeate through a material barrier.

Many of these materials–whether liquid solvents, solid sorbents, membranes, or otherwise–have millions of possible variations, including ones we still aren’t aware of. Physically testing every single potential candidate is unfeasible and wastefully time-consuming, which is why researchers are employing AI techniques to generate new potential materials and filter viable options.

For example, MOFs consist of inorganic nodes, organic nodes, and organic linkers, which self-assemble into different configurations, yielding distinct chemical and physical properties. Researchers at the University of Illinois developed an AI model that assembled 120,000 possible structures, then conducted further testing to filter out structures that were physically improbable or too expensive to synthesize. After narrowing down to 364 candidates, they ran 3D molecular simulations, which led to the identification of six highest-performing structures that can be synthesized for physical testing.

AI’s Role in Optimizing CCUS Implementation

Beyond accelerating the discovery and testing of new materials, artificial intelligence can play several important roles in the deployment of CCUS. Once carbon is captured, there are two potential pathways: utilization, in which CO2 is repurposed for industrial applications, such as the production of synthetic fuels, and sequestration, in which CO2 is stored in underground geological formations, like saline aquifers.

For example, AI can be leveraged to determine how much carbon should be captured in energy systems, and how much of this captured carbon should be sequestered versus utilized. Researchers in the UAE proposed a superstructure network that encompasses all components of CCUS supply chains and used it to develop a supply chain optimization model. Their results showed that while storage yields better economic results, utilization is a more sustainable pathway.

Another promising application of AI is in the scheduling of carbon capture within energy systems. Due to its high energy demand, carbon capture can incur steep costs if inefficiently managed. One study employed a deep reinforcement learning (DRL) agent for the real-time scheduling of a multi-energy system including carbon capture. The DRL agent successfully met energy demands within the specified constraints and outperformed rule-based scheduling by 23.65%.

Policy Implications

These examples are just a fraction of AI’s potential applications across the spectrum of carbon capture, from streamlining research and testing to determining optimal CCUS implementations. For CCUS to reach the scale required in net-zero scenarios, policymakers should both support the leveraging of these various applications and integrate the meaningful insights they can provide into decision-making. Results of optimization models can inform governments on where to direct subsidies and how to structure energy policies, by indicating which carbon capture pathways and configurations are best suited for their goals. Furthermore, governments can help by improving access to databases to be leveraged by AI-based exploration, while private actors can funnel investments toward accelerating the commercialization of the most promising technologies.

Anya Draves

Undergraduate Seminar Fellow

Anya Draves is an undergraduate student majoring in physics. She has been involved in sustainability initiatives including Eco-Reps, Penn Farm and the Student Advisory Group for the Environment, and is a 2025 Kleinman Center Undergraduate Fellow.