AI Catalyst Model Cracks Magnetic Code, Promises Green Manufacturing

PALO ALTO, CA – June 16, 2026 – In a significant leap for computational science and industrial manufacturing, SandboxAQ has unveiled an AI-powered platform that solves a decades-old problem in chemistry: accurately simulating the magnetic properties of catalysts. The research, published today in the peer-reviewed journal npj Computational Materials, introduces a method that could dramatically accelerate the discovery of new materials, paving the way for more sustainable fuels, greener fertilizers, and more efficient industrial processes.

The work centers on AQCat25, a massive dataset and a corresponding family of AI models. By incorporating the magnetic behavior—or “spin polarization”—of common industrial metals, the technology gives researchers a tool to design and screen next-generation catalysts with unprecedented speed and physics-based accuracy. This development addresses a critical blind spot that has long hampered virtual materials design.

“Catalysis drives the global economy, from the fuels that power our world to the materials that shape it,” said SandboxAQ CEO Jack Hidary in a statement. “With our AQCat model, industries can now simulate, screen, and optimize catalysts with physics-based accuracy, unlocking performance and sustainability breakthroughs at unprecedented scale.”

Cracking the Magnetic Code in Chemistry

Catalysts are the unsung heroes of the modern world, enabling or speeding up chemical reactions that are essential to producing over 80% of all manufactured goods. Many of the most important industrial catalysts rely on earth-abundant metals like iron, cobalt, and nickel. These elements have a crucial property that has been notoriously difficult and expensive to model computationally: magnetism.

This magnetic behavior, known as spin polarization, profoundly influences how molecules interact with a catalyst's surface, determining the efficiency and outcome of a chemical reaction. Historically, large-scale simulation datasets often omitted these effects to save on immense computational costs, forcing scientists to work with an incomplete picture and limiting the accuracy of predictions for many industrially vital materials.

SandboxAQ’s AQCat25 directly closes this gap. The dataset is a monumental achievement in itself, comprising 13.5 million high-fidelity calculations based on Density Functional Theory (DFT), the gold standard for quantum-level simulation. Generated using approximately 400,000 GPU-hours on NVIDIA DGX Cloud, the dataset spans 47,000 different catalyst systems, explicitly modeling spin polarization for 12 magnetic elements.

The AI models trained on this data are the real game-changer. They deliver DFT-level accuracy at speeds up to 20,000 times faster than running the first-principles simulations from scratch. This colossal speedup transforms catalyst discovery from a slow, painstaking process into a practical, high-throughput virtual screening endeavor. Researchers can now evaluate thousands of potential catalyst candidates in the time it once took to analyze a single one.

From Digital Blueprint to Global Factory Floor

The real-world implications of this breakthrough are vast. By providing a more accurate map of chemical possibilities, the AQCat25 models can guide scientists toward creating catalysts that are not only more efficient but also more sustainable. This has the potential to reshape industries that are fundamental to the global economy and have a significant environmental footprint.

Potential applications span numerous sectors:
* Energy: Accelerating the development of catalysts for producing green hydrogen, converting CO₂ into fuel, and creating sustainable aviation fuels.
* Agriculture: Designing more efficient processes for ammonia synthesis, the cornerstone of modern fertilizers, which could reduce the energy intensity of food production.
* Circular Economy: Creating novel catalysts that can break down plastics (depolymerization) or convert industrial waste streams into valuable new materials.

Furthermore, by enabling accurate modeling of earth-abundant metals, the technology encourages a shift away from expensive and rare precious metals like platinum and palladium, which are often used in catalytic converters and other applications. This move could reduce costs, mitigate supply chain risks, and lessen the environmental impact of mining.

For corporate R&D labs, the platform promises to dramatically shorten development cycles. The ability to pre-screen vast design spaces virtually before committing to expensive and time-consuming physical experiments can save millions of dollars and bring innovative products to market years earlier.

A New Era of Open-Source Materials Science

In a move aimed at accelerating progress across the entire scientific landscape, SandboxAQ has released the AQCat25 dataset and its associated models publicly on the Hugging Face platform under a Creative Commons license. This decision democratizes access to a powerful tool that would otherwise be out of reach for many academic labs and smaller companies, which lack the resources to conduct massive-scale computations.

This open-source approach positions AQCat25 to become a foundational resource for the materials science community. It complements other large-scale datasets, like the Open Catalyst 2020 (OC20) project, by providing higher-fidelity data and the critical missing piece of spin polarization. The company's research even details advanced joint-training strategies to combine insights from different datasets without a loss of accuracy, showcasing a path for collaborative scientific advancement.

By making its tools public, SandboxAQ is fostering an ecosystem where researchers globally can validate, build upon, and integrate this technology into their own work. This could ignite a new wave of innovation, leading to discoveries far beyond what a single organization could achieve alone and potentially setting a new standard for data sharing in computational chemistry.

The Intersection of AI and Quantum-Inspired Physics

This breakthrough is a flagship demonstration of SandboxAQ's core strategy, which operates at the intersection of AI and quantum science. The company develops what it calls Large Quantitative Models (LQMs)—physics-based AI systems designed to tackle complex simulation and optimization problems in sectors ranging from life sciences and materials to finance and navigation.

The AQCat25 models are a prime example of this approach. Instead of treating AI as a black box, they are built on the fundamental laws of quantum physics, allowing them to make predictions that are both fast and physically sound. By successfully modeling the complex quantum phenomenon of magnetism at an industrial scale, SandboxAQ has not only provided a powerful tool for chemists but has also offered a compelling glimpse into a future where physics-informed AI helps solve some of the world's most challenging scientific and engineering problems.