NVIDIA’s AI advances revolutionise materials discovery for sustainable technologies

To drive innovation across a spectrum of future technologies, from liquid-cooled data centers and high-resolution digital displays to long-lasting batteries, scientific research is intensifying its search for novel chemicals and materials optimised for energy efficiency, durability, and performance. Central to this effort are advanced computational methods and AI-driven simulations, which enable researchers to predict and analyse material properties at an atomic scale with remarkable precision and speed.

At the SC25 conference held in St. Louis, NVIDIA unveiled a suite of accelerated data processing pipelines and AI microservices designed specifically to bolster chemistry and materials science. These technologies promise to transform sectors such as aerospace, energy, and manufacturing by drastically accelerating the pace of materials discovery and validation.

One of the highlights at the conference was a demonstration by the U.S. Department of Energy’s Brookhaven National Laboratory. Using the NVIDIA Holoscan AI sensor processing platform, Brookhaven researchers can visualize materials at resolutions below 10 nanometres. The platform is deployed at Brookhaven’s National Synchrotron Light Source II (NSLS-II), a facility renowned for its powerful X-ray beamlines that probe the structure of complex material systems, ranging from batteries to microelectronic components and nanoparticle superlattices.

Historically, data generated from these imaging experiments posed significant challenges due to sheer volume and complexity. Integration of the NVIDIA Holoscan platform, specifically designed for real-time, high-throughput edge computing of streaming data, allows researchers to receive near-instant feedback as scans are conducted. Hanfei Yan, lead beamline scientist at NSLS-II, emphasised to Nvidia’s blog that this immediate insight enables identifying regions of interest on the fly and tracking property changes in real time, crucial for making informed decisions during experiments. This real-time processing reduces costly instrument downtime and opens the door for AI-assisted autonomous experimentation, according to Daniel Allan, NSLS-II’s group leader of data engineering.

The Holoscan platform’s architecture, originally designed to handle multimodal sensor data in domains like medical imaging and industrial inspection, features a modular, scalable pipeline that supports rapid development and deployment. Its integration into scientific workflows underscores the expanding role of edge AI in accelerating research outputs.

In parallel, NVIDIA showcased ALCHEMI, a suite of microservices on the NIM (NVIDIA Industrial Metaverse) platform tailored for high-throughput atomistic simulations. These microservices include Batched Conformer Search (BCS) and Batched Molecular Dynamics (BMD), which apply machine learning models to predict and simulate chemical properties with quantum chemistry-level accuracy but at a fraction of the time traditional methods require. The BCS microservice utilises AIMNet2, an interatomic potential model, to identify and rank low-energy molecular conformers rapidly, while BMD accelerates molecular dynamics simulations by dynamically batching workloads and leveraging GPU acceleration.

Japanese energy giant ENEOS has leveraged these ALCHEMI microservices for two pivotal energy research streams: discovering new liquids for immersion cooling in next-generation data centers, a critical technology to manage thermal loads in increasingly powerful computing environments, and identifying catalysts for processes like hydrogen fuel production. ENEOS’s AI innovation lead, Takeshi Ibuka, noted that these tools allowed the evaluation of tens of millions of liquid candidates and ordinally more catalytic candidates within weeks, achieving throughput roughly an order of magnitude higher than previous methods. This extensive computational prescreening enables significant cost savings and truncates the timeline from concept to commercialisation.

Similarly, Universal Display Corporation (UDC), a leader in OLED material innovation, has applied NVIDIA’s ALCHEMI NIM services to accelerate the discovery of next-generation OLED compounds that promise enhanced energy efficiency, colour precision, and device performance. Faced with the daunting molecular design space, estimated to encompass as many as 10^100 potential molecules, UDC researchers can now evaluate billions of candidates up to 10,000 times faster than traditional CPU-bound approaches allowed. Julie Brown, UDC’s executive vice president and chief technical officer, highlighted that GPU-accelerated discovery combined with their scientific expertise has dramatically scaled both the speed and scope of their research. By running molecular dynamics simulations concurrently across multiple GPUs, simulation times have been compressed from days to mere seconds, empowering scientists with immediate feedback and greater creative freedom.

These advances do not merely enhance OLED technology but also contribute towards broader goals of energy efficiency and sustainability in digital displays worldwide, Brown emphasised during a discussion featured by NVIDIA.

Overall, the convergence of real-time AI-powered sensing through Holoscan and high-throughput molecular simulations from ALCHEMI exemplifies a pivotal shift in industrial decarbonisation and advanced materials science. These technologies deliver critical capabilities for predicting and engineering materials that can meet the demanding performance and sustainability criteria of tomorrow’s technologies.

Such integration of edge AI and cloud-scale computation positions these tools not only as accelerators of scientific discovery but also as enablers of cost-effective, scalable innovation across the energy, manufacturing, and aerospace sectors. As these platforms mature and broaden their adoption, they are poised to play a foundational role in the industrial transition to cleaner, smarter materials and technologies.