SandboxAQ has released a dataset containing 11 million high-fidelity quantum chemistry calculations designed to revolutionize how researchers develop catalysts and advanced materials.
Named AQCat25, the publicly available dataset addresses two critical barriers that have limited AI applications in computational heterogeneous catalysis, a field crucial for industries where catalysts play a vital role.
It provides extensive data on 40,000 intermediate-catalyst systems generated through accurate quantum chemistry calculations on GPUs, enabling machine learning models to deliver predictions up to 20,000 times faster than traditional physics-based methods.
AQCat25 also incorporates spin polarization measurements for materials beyond oxides, making it particularly valuable for applications involving Earth’s most abundant metals. This feature opens new possibilities for sustainable aviation fuel production, green hydrogen creation, fertilizer manufacturing and industrial waste conversion.
“AQCat25 enables scientists and engineers to design the next generation of chemicals, catalysts and advanced materials faster and more cost-effectively than traditional manufacturing processes or existing AI-accelerated approaches,” said SandboxAQ head of innovation Adam Lewis in the announcement.
The dataset was developed using Nvidia DGX Cloud, requiring more than 400,000 GPU-hours of computation on Nvidia DGX H100 cards. This infrastructure enabled SandboxAQ to create the comprehensive dataset in record time.
The technology has significant industrial implications, as more than 90% of commercially produced chemicals and over 80% of manufactured goods, including vehicles, medicines, gasoline and detergents, rely on catalysts during production.
SandboxAQ’s large quantitative models trained on AQCat25 aim to explore broader chemical possibilities, design entirely new compounds and identify optimal chemical formulations in days rather than months or years.
The AQCat25 dataset is now available for researchers and industry professionals worldwide on the Hugging Face platform.
SandboxAQ, a quantitative AI startup that draws on quantum computing techniques to develop quantitative artificial intelligence models for enterprises, spun out from Alphabet in 2022.
The announcement comes just months after the company closed its series E funding round with $450 million in April. The latest investment from major industry players, including Google, Nvidia and BNP, brought SandboxAQ’s total funding to $950 million.