NSF and NVIDIA Partner on Next-Gen AI Chips for Scientific Discovery and Open Models

August 18, 2025

By Silo AI Team

NSF and NVIDIA Partner on Next-Gen AI Chips for Scientific Discovery and Open Models

The National Science Foundation (NSF) and NVIDIA have announced a groundbreaking partnership to invest approximately $150 million in the development of open, next-generation AI models and advanced AI infrastructure designed to accelerate scientific discovery in the United States. This initiative centers on the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) project, led by the Allen Institute for AI (Ai2).

Key Elements of the Partnership

The partnership is structured around several key elements:

Funding and Leadership

The NSF is contributing $75 million, while NVIDIA is providing $77 million. The Allen Institute for AI (Ai2) will lead the effort, leveraging its expertise in large language models and multimodal AI.

Objective

The goal is to build a fully open suite of advanced AI models—specifically, multimodal large language models (LLMs) capable of understanding and processing various data types such as text, images, and scientific data.

Purpose and Impact

These open models will be trained on scientific literature and data, making them highly specialized for research domains like biology, energy, and materials science. They are intended to enable researchers to process information more rapidly, generate code and visualizations, and connect emerging insights to prior discoveries, significantly accelerating scientific progress.

Open-Source Commitment

All models and supporting software will be fully open-source, with source code and model weights made publicly available to ensure broad accessibility and foster innovation across the scientific community.

Silo Tech Article Banner - NSF and NVIDIA Partner on Next-Gen AI Chips for Scientific Discovery and Open Models

AI Infrastructure

NVIDIA’s contribution includes cutting-edge hardware, specifically the NVIDIA HGX B300 systems powered by Blackwell Ultra GPUs, and the NVIDIA AI Enterprise software platform. This infrastructure is designed to handle the large-scale training and inference demands of multimodal scientific models, offering high efficiency, scalability, and performance.

Policy Context

The partnership directly supports the White House’s AI Action Plan, which stresses the importance of open models and public-private collaboration to maintain U.S. leadership in AI and scientific innovation.

Broader Implications

The partnership has several broader implications for the scientific community:

Democratizing Access to AI: The rising costs of training and deploying advanced AI models have made it difficult for academic and public research institutions to keep pace with industry, which this partnership aims to address by democratizing access to powerful AI for science.
Lasting National Asset: By focusing on open and domain-specific AI tools, the initiative is expected to create a “lasting national asset” that strengthens both scientific discovery and the foundational science of AI itself.

Conclusion

In summary, NSF and NVIDIA, with Ai2, are jointly investing in open, multimodal AI model development and deploying powerful next-gen AI chips and infrastructure. The aim is to transform and democratize scientific research in the U.S., ensuring open access to state-of-the-art AI tools and reaffirming U.S. leadership in both science and AI innovation. This partnership has the potential to accelerate scientific discovery and create a lasting impact on the scientific community.