If Moore’s Law had a chemistry cousin, this might be it. SandboxAQ has unveiled AQCat25-EV2, a powerful new quantitative AI model that promises to reshape how researchers discover and design catalysts—the unseen engines driving over 80% of global manufacturing.
Developed using the AQCat25 dataset, the model brings a level of computational speed and precision once thought impossible. It’s capable of predicting chemical energetics at near quantum-mechanical accuracy but up to 20,000 times faster, marking a major leap forward for computational materials science.
The Catalyst Bottleneck—And Why It Matters
Catalysts sit at the heart of everything from fertilizers and fuels to plastics and pharmaceuticals. Yet despite their ubiquity, designing better ones remains painfully slow. Traditional laboratory methods can test fewer than 100 catalysts a week, making most breakthroughs incremental rather than transformative.
That’s where AQCat25-EV2 comes in. By integrating quantum spin polarization—a magnetic property long neglected by prior AI models—SandboxAQ’s system can now model all industrially relevant elements in the periodic table. Translation: researchers can run massive, accurate virtual screenings across chemical systems that were previously off-limits.
“This model allows in silico screening across an unprecedented range of chemistries,” said Dr. Bob Maughon, former CTO of SABIC. “It’s a powerful step toward solving critical problems like CO₂ reduction, plastic recycling, and advanced battery materials.”
Quantum Meets AI—At Scale
The model’s foundation, the AQCat25 dataset, includes 13.5 million high-fidelity quantum chemistry calculations across 47,000 catalytic systems. That dataset makes AQCat25-EV2 not just another machine learning model—it’s one of the first AI systems grounded in physical chemistry at scale.
SandboxAQ trained the model using NVIDIA DGX Cloud, consuming more than 500,000 GPU-hours on NVIDIA H100 Tensor Core GPUs. The company plans to make it available through NVIDIA’s ALCHEMI platform, expanding access to researchers worldwide.
Beyond Incremental Progress
Historically, catalyst discovery has been limited by cost and time—researchers often stopped testing after finding one viable candidate, leaving potentially better ones undiscovered. AQCat25-EV2 changes that dynamic, allowing entire chemical spaces to be explored virtually.
For industries struggling with sustainability and energy challenges, this could be game-changing. SandboxAQ expects the model to aid breakthroughs in methane-to-methanol conversion, hydrogen fuel cell catalysts, CO₂-to-fuel processes, and depolymerization for plastic recycling.
According to Aayush Singh, Head of Catalytic Sciences at SandboxAQ, the goal is to “de-risk R&D across materials science” by giving researchers a broader, faster, and more confident discovery toolkit.
The Bigger Picture
SandboxAQ’s move comes amid a broader surge of AI-meets-physics research, where quantum-informed models are being used to accelerate materials innovation. Rivals like DeepMind’s GNoME and Microsoft’s Materials Acceleration Platform are exploring similar territory—but AQCat25-EV2’s spin-polarized approach sets it apart for its realism and industrial applicability.
If successful, the model could make quantum chemistry simulation at scale not just a research milestone but a commercial norm—opening the door to greener, cheaper, and more efficient materials design across sectors.
In short, SandboxAQ isn’t just speeding up chemistry; it’s redefining how science gets done.
If Moore’s Law had a chemistry cousin, this might be it. SandboxAQ has unveiled AQCat25-EV2, a powerful new quantitative AI model that promises to reshape how researchers discover and design catalysts—the unseen engines driving over 80% of global manufacturing.
Developed using the AQCat25 dataset, the model brings a level of computational speed and precision once thought impossible. It’s capable of predicting chemical energetics at near quantum-mechanical accuracy but up to 20,000 times faster, marking a major leap forward for computational materials science.
The Catalyst Bottleneck—And Why It Matters
Catalysts sit at the heart of everything from fertilizers and fuels to plastics and pharmaceuticals. Yet despite their ubiquity, designing better ones remains painfully slow. Traditional laboratory methods can test fewer than 100 catalysts a week, making most breakthroughs incremental rather than transformative.
That’s where AQCat25-EV2 comes in. By integrating quantum spin polarization—a magnetic property long neglected by prior AI models—SandboxAQ’s system can now model all industrially relevant elements in the periodic table. Translation: researchers can run massive, accurate virtual screenings across chemical systems that were previously off-limits.
“This model allows in silico screening across an unprecedented range of chemistries,” said Dr. Bob Maughon, former CTO of SABIC. “It’s a powerful step toward solving critical problems like CO₂ reduction, plastic recycling, and advanced battery materials.”
Quantum Meets AI—At Scale
The model’s foundation, the AQCat25 dataset, includes 13.5 million high-fidelity quantum chemistry calculations across 47,000 catalytic systems. That dataset makes AQCat25-EV2 not just another machine learning model—it’s one of the first AI systems grounded in physical chemistry at scale.
SandboxAQ trained the model using NVIDIA DGX Cloud, consuming more than 500,000 GPU-hours on NVIDIA H100 Tensor Core GPUs. The company plans to make it available through NVIDIA’s ALCHEMI platform, expanding access to researchers worldwide.
Beyond Incremental Progress
Historically, catalyst discovery has been limited by cost and time—researchers often stopped testing after finding one viable candidate, leaving potentially better ones undiscovered. AQCat25-EV2 changes that dynamic, allowing entire chemical spaces to be explored virtually.
For industries struggling with sustainability and energy challenges, this could be game-changing. SandboxAQ expects the model to aid breakthroughs in methane-to-methanol conversion, hydrogen fuel cell catalysts, CO₂-to-fuel processes, and depolymerization for plastic recycling.
According to Aayush Singh, Head of Catalytic Sciences at SandboxAQ, the goal is to “de-risk R&D across materials science” by giving researchers a broader, faster, and more confident discovery toolkit.
The Bigger Picture
SandboxAQ’s move comes amid a broader surge of AI-meets-physics research, where quantum-informed models are being used to accelerate materials innovation. Rivals like DeepMind’s GNoME and Microsoft’s Materials Acceleration Platform are exploring similar territory—but AQCat25-EV2’s spin-polarized approach sets it apart for its realism and industrial applicability.
If successful, the model could make quantum chemistry simulation at scale not just a research milestone but a commercial norm—opening the door to greener, cheaper, and more efficient materials design across sectors.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
