Supporting this infrastructure, cloud services from Cirrascale and hardware from Supermicro will be utilized, ensuring robust development and deployment of advanced AI models.

NVIDIA’s infrastructure features Blackwell Ultra GPUs with high-bandwidth memory, designed to handle large models and demanding workloads efficiently, enabling researchers to process enormous datasets and generate actionable insights.

Ai2 highlights the importance of open AI models that allow analysis, modification, and training from scratch, contrasting with proprietary models that rely on closed data.

The U.S. National Science Foundation (NSF) has partnered with NVIDIA in a groundbreaking initiative to develop open-source AI models aimed at transforming scientific research across the country, with a combined investment exceeding $152 million supporting the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) project led by the Allen Institute for AI (Ai2).

This project emphasizes open models and data transparency to enable systematic study of AI behaviors, improve reproducibility, and verify scientific applications, setting a new standard for openness in AI-driven research.

NVIDIA will supply high-performance AI infrastructure, including HGX B300 systems and AI Enterprise software, to accelerate model training and deployment at scale, facilitating faster scientific insights.

The initiative will be implemented in stages, with initial models expected to be available approximately 18 months into the project, and outputs will be released progressively to the scientific community.

The project will support research teams from multiple U.S. universities, including the University of Washington, Hawaii at Hilo, New Hampshire, and New Mexico, aligning with national efforts to strengthen AI leadership.

This initiative supports the White House's AI Action Plan, emphasizing the acceleration of AI-enabled science, fostering innovation, and maintaining U.S. leadership in global AI research and development.

It marks NSF’s first major investment in AI software infrastructure for science, focusing initially on areas like new materials discovery, protein function prediction, and improving large language models.

The project centers on large language models with billions of parameters, with a strong emphasis on open access to datasets, training data, code, and documentation to promote transparency and systematic study of AI behaviors.

Funding will primarily support computational resources to develop larger, more advanced AI models, aiding researchers in data processing, visualization, code generation, and pattern recognition to accelerate scientific breakthroughs.

Ai2 plans to release key artifacts during development for transparency and community engagement, with a major release expected around 18 months after project start.