Dr. Adam Lewis from SandboxAQ announced the launch of AQCat25, a dataset comprising 11 million high‑fidelity quantum‑chemistry calculations on 40 000 intermediate‑catalyst systems, generated with more than 400 000 GPU‑hours on NVIDIA DGX H100 cards. The data, which includes spin‑polarised calculations for non‑oxide materials, enables machine‑learning models to predict catalytic properties up to 20 000 times faster than physics‑based methods, thereby addressing two long‑standing barriers to AI‑driven heterogeneous catalysis. With more than 90 % of commercially produced chemicals and over 80 % of manufactured goods relying on catalysts, AQCat25 is poised to accelerate the design of catalysts for sustainable aviation fuel, green hydrogen and waste‑to‑material conversion, and is now publicly available on the Hugging Face platform. Jeff Graf, Global Head of Business Development at SandboxAQ, added that the dataset will reduce R&D time, cost and risk for industrial and academic users alike.
On 10 September 2025 SandboxAQ announced AQCat25, a public dataset of 11 million high‑fidelity quantum‑chemical calculations covering 40 000 intermediate‑catalyst systems. The collection, released from Palo Alto, California, is hosted on the Hugging Face platform and can be integrated into SandboxAQ’s Large Quantitative Models (LQMs), which traverse chemical space and propose optimal catalysts in days rather than months or years.
The calculations were performed on NVIDIA DGX Cloud using H100 GPUs, consuming more than 400 000 GPU‑hours. The H100 cards, built on the Ampere architecture, provide peak double‑precision performance essential for density‑functional theory simulations that resolve spin‑polarised electronic structures. A unified AI stack pre‑configures GPU kernels, memory bandwidth and inter‑node communication, enabling thousands of parallel jobs with minimal overhead and keeping the 400 000‑hour budget on track.
Spin‑Polarised Data Enables New Catalytic Applications
Each entry in AQCat25 incorporates a highly accurate, spin‑resolved quantum‑chemical method that explicitly accounts for magnetic interactions in non‑oxide materials. This feature, absent from other large‑scale catalytic AI datasets, is particularly relevant for earth‑abundant metals such as iron, cobalt, nickel and copper, and allows the exploration of magnetic effects that influence catalytic performance. The dataset’s spin‑polarised component opens new avenues for catalyst design in sustainable aviation fuel synthesis, green hydrogen production, fertilizer manufacture, industrial waste conversion, and waste‑to‑materials conversion.
Training state‑of‑the‑art machine‑learning models on AQCat25 enables predictions of catalytic performance in a single forward pass, a process that would normally require months of high‑level theory calculations. The result is a 20 000‑fold speed‑up over conventional physics‑based methods, reducing the design cycle from years to hours and lowering both cost and risk for developing new catalytic processes across energy, chemicals, automotive and consumer goods sectors.
Dr. Adam Lewis, Head of Innovation at SandboxAQ, said, “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.” Jeff Graf, Global Head of Business Development, added, “With the combined power of NVIDIA DGX Cloud and SandboxAQ’s Large Quantitative Models, AQCat25 will transform catalytic discovery and optimisation processes, elevating the field of AI‑powered materials science to its highest potential.”
By removing barriers of proprietary software and costly computational resources, the free release of AQCat25 on Hugging Face empowers academic groups and industrial R&D teams to download and ingest the data directly into their machine‑learning workflows. The high‑resolution, spin‑resolved insights accelerate screening of catalysts for pharmaceuticals, detergents, consumer goods, exhaust treatment, fuel cell components and polymer degradation, fostering cross‑sector collaboration and positioning AQCat25 as a pivotal resource for the next wave of catalytic innovation.
Original Press Release
Source: View Original Source
