Metal-organic frameworks (MOFs) are promising for CO2 capture, but finding optimal materials is challenging due to the vast chemical space. This paper introduces GHP-MOFassemble, a generative AI framework for designing MOFs with high CO2 adsorption capacity. GHP-MOFassemble generates novel linkers, assembles them with pre-selected metal nodes into MOFs, and screens for uniqueness, synthesizability, and structural validity. Molecular dynamics and Grand Canonical Monte Carlo simulations assess stability and CO2 adsorption capacity. The top six AI-generated MOFs show CO2 capacities exceeding 2 mmol g⁻¹, surpassing 96.9% of structures in a hypothetical MOF dataset.

Hyun Park, Xiaoli Yan, Ruijie Zhu, Eliu A. Huerta, Santanu Chaudhuri, Donny Cooper, Ian Foster, Emad Tajkhorshid