NSF and NVIDIA Invest $152M in Open-Source AI for Scientific Research

The U.S. National Science Foundation (NSF) and NVIDIA have launched a $152 million partnership to develop open-source AI infrastructure for the scientific community. The initiative ensures that high-performance AI tools remain accessible to academic researchers who lack the massive budgets of private industry.

Led by the Allen Institute for AI (Ai2), the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) project will create a suite of fully open-source, multimodal large language models (LLMs) trained specifically on scientific data. The total investment is split nearly evenly, with the NSF contributing $75 million through its Mid-Scale Research Infrastructure program and NVIDIA providing $77 million. The project supports a consortium of academic institutions, including the University of Washington, the University of New Mexico, the University of New Hampshire, and the University of Hawaii at Hilo.

Closing the Compute Divide

For years, the escalating cost of developing frontier AI models has created a significant gap between corporate labs and university researchers. While private companies can afford the immense compute power required to train state-of-the-art models, traditional federal grants and university budgets have struggled to keep pace. This "compute divide" threatens to stall academic innovation in critical fields such as biology, energy, and materials science.

By providing open-weight models and the necessary infrastructure, the OMAI project aims to democratize access to these tools. This allows researchers to fine-tune models on specialized scientific literature and proprietary datasets without needing to build a foundation model from scratch.

Implications for Global Science

This shift toward "fully open" AI is a strategic move to prevent the foundational tools of scientific discovery from being locked behind proprietary APIs. When models are open-source, the global research community can audit, verify, and improve them, accelerating the pace of discovery across disciplines.

"AI is the engine of modern science — and large, open models for America's researchers will ignite the next industrial revolution," said NVIDIA founder and CEO Jensen Huang. Ali Farhadi, CEO of Ai2, echoed this sentiment, stating that fully-open AI is "not just a preference — it's a necessity."

Future Outlook

As the OMAI project rolls out its infrastructure, the focus will shift to how these multimodal models handle complex scientific data, such as imagery and chemical structures, alongside text. While the partnership establishes the financial and technical framework, the long-term success of the initiative will depend on the adoption rate among academic labs and the ability of the open-source community to maintain and evolve the models as AI capabilities continue to scale.