SandboxAQ Unveils Spin-Aware AI Model for Catalyst Discovery

The new dataset includes 13.5 million density functional theory calculations covering 47,000 catalyst systems. While catalysts are essential to the production of over 80 percent of manufactured goods—ranging from fertilizers to fuel—previous computational models frequently omitted the magnetic properties of metals like iron, cobalt, and nickel due to high processing costs. This omission significantly hampered the accuracy of virtual material screening.

Developed using 400,000 GPU-hours on NVIDIA DGX Cloud, the model accounts for spin polarization across 18 elements, including new additions such as lithium and magnesium. According to the company, this approach is up to 20,000 times faster than traditional first-principles simulations. CEO Jack Hidary stated that the release aims to unlock sustainability and performance breakthroughs by making high-throughput screening practical for industrial applications. The data is now available to the public on Hugging Face under a Creative Commons license, providing researchers with a tool to bypass previous limitations in computational chemistry.