NVIDIA and the National Science Foundation Partner to Advance Open AI Models for U.S. Scientific Leadership

NVIDIA is collaborating with the U.S. National Science Foundation (NSF) to advance scientific discovery through artificial intelligence, supporting the Allen Institute for AI (Ai2) in developing open, multimodal AI models under the OMAI initiative—Open Multimodal AI Infrastructure to Accelerate Science. This project is part of the NSF’s Mid-Scale Research Infrastructure program and aims to establish a national, fully open AI ecosystem that empowers researchers across the country. Brian Stone, acting director of the NSF, highlighted the transformative potential of AI in science, stating that the partnership equips American scientists with cutting-edge tools to accelerate breakthroughs and secure U.S. leadership in science and technology. The initiative is designed to tackle complex challenges once considered insurmountable by providing researchers with access to powerful AI infrastructure and open models. NVIDIA’s contribution includes state-of-the-art NVIDIA HGX B300 systems, powered by Blackwell Ultra GPUs, which deliver unprecedented performance in model training and inference. These systems feature high-bandwidth memory and advanced interconnect technologies, enabling efficient handling of the largest AI models and most demanding workloads. NVIDIA also provides the AI Enterprise software platform, which supports scalable, secure, and optimized AI development. The OMAI project will support research teams at the University of Washington, the University of Hawaii at Hilo, the University of New Hampshire, and the University of New Mexico. By combining public and private investment, the initiative strengthens U.S. technological competitiveness and aligns with the White House’s AI Action Plan, which prioritizes AI-driven innovation, faster data center deployment, and global leadership in AI. Noah Smith, senior director of natural language processing research at Ai2, emphasized the importance of access to compute resources, noting that without powerful infrastructure, even the most ambitious AI models cannot be built. He added that open access to training data, code, documentation, and tools allows researchers to trace model behaviors back to training instances and better understand how AI systems learn and evolve. The OMAI initiative will provide fully open models, open-source data analysis tools, and comprehensive training materials to early-career researchers. By making these resources available at low or no cost, the project mirrors the principles of open-source software and scientific digital libraries, ensuring broad and equitable access. This effort supports the broader goal of advancing U.S. leadership in science and engineering by democratizing access to frontier AI. As part of the White House’s “Winning the AI Race” strategy, OMAI strengthens the foundation for innovation in academia, industry, and government, positioning America at the forefront of the next wave of scientific and technological progress.