SandboxAQ (Palo Alto, Calif.; www.sandboxaq.com) www.today announced the launch of AQCat25, a breakthrough large-scale AI dataset offering deep insights into the atomic-scale dynamics that drive state-of-the-art industrial chemistry. The AQCat25 dataset enables researchers to make highly accurate predictions of material properties in catalytic reactions from atomic structures.
Today, more than 90% of all commercially produced chemicals and over 80% of all manufactured goods rely on catalysts in their production. Mass-produced goods such as autos, medicines, gasoline and detergents all need catalysts for manufacture. AQCat25 delivers material value to the chemical and catalyst industries by overcoming two critical barriers that have hindered the use of AI for computational heterogeneous catalysis.
First, the dataset includes 11 million data points on 40,000 intermediate-catalyst systems generated using highly accurate quantum chemistry calculations on GPUs to ensure more reliable modelling predictions. AQCat25 extends the capability of existing fast-computing machine learning models to new, industrially-relevant problems and enables training frontier models to deliver up to 20,000x faster performance over physics-based methods for catalyst design.
Second, AQCat25 is the only large-scale catalytic AI dataset to include spin polarization, measuring magnetic effects, for materials beyond oxides. Since many of earth’s most abundant metals are spin polarized, AQCat25 is highly relevant for a broad range of applications such as producing sustainable aviation fuel and fertilizer, creating stable green hydrogen, converting industrial waste streams into useful materials, and other applications.
“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 Dr. Adam Lewis, Head of Innovation SandboxAQ. “By publicly releasing AQCat25, SandboxAQ enables the world’s leading industrial companies and academic institutions to significantly accelerate their existing R&D and advance their efforts to bring innovative new products and solutions to market faster.”
AQCat25 was generated on NVIDIA DGX™ Cloud, leveraging more than 400,000 GPU-hours of computation using NVIDIA DGX H100 cards. The unified AI platform provided SandboxAQ with the optimized computing infrastructure needed to develop AQCat25 in record time.
“Catalysts are essential for advancing industrial processes and converting raw materials into value for the global economy,” said xxx, DGX. “With the combined power of NVIDIA DGX Cloud platform and SandboxAQ’s Large Quantitative Models, AQCat25 will transform catalytic discovery and optimization processes, decrease R&D time, cost, and risk, and elevate the field of AI-powered materials science to its highest potential.”
Large Quantitative Models (LQMs) trained on datasets like AQCat25 can explore a broader chemical space, design novel compounds not currently found in literature, and identify optimal chemical compounds in days instead of months or years.
The AQCat25 dataset is publicly available today on the Hugging Face platform. To learn more, visit https://sandboxaq.com/aqcat25.
SandboxAQ is a B2B company delivering solutions at the intersection of AI and quantum techniques. The company’s Large Quantitative Models (LQMs) deliver critical advances in life sciences, financial services, navigation, and other sectors. The company emerged from Alphabet Inc. as an independent, growth-backed company funded by leading investors including funds and accounts advised by T. Rowe Price Associates, Inc., IQT, US Innovative Technology Fund, S32, Hillspire Capital, Breyer Capital, Marc Benioff, Thomas Tull, Paladin Capital Group, and others.