NSF commits $100M to make scientific datasets AI-ready

Federal push to retrofit research data for machine learning

The National Science Foundation has launched a $100 million program designed to make existing scientific datasets more accessible to artificial intelligence systems, shifting focus from new data collection to maximizing the value of research already conducted.

The NSF Unlocking Dataset Value for AI-Enabled Scientific Discovery program will distribute awards ranging from $2 million to $5 million to research teams that enhance datasets for automated analysis. Planning grants of up to $200,000 will also be available, according to details first reported by NSF.

The initiative addresses a persistent challenge in computational research: many valuable datasets created through prior NSF-funded work remain difficult for AI systems to process. They may lack proper metadata, use incompatible formats, or sit in isolated repositories that prevent cross-disciplinary analysis.

What the program will fund

Rather than supporting fresh data gathering, the program targets improvements to datasets already in hand. Eligible projects include developing automated methods for feature extraction and metadata generation, building data pipelines that feed AI analysis tools, and creating frameworks to integrate multiple datasets from different sources.

Researchers are encouraged to build on existing infrastructure, including NSF data platforms, the NSF Integrated Data Systems and Services program, and the National AI Research Resource. The program also connects to the American Science and Security Platform developed by the Department of Energy.

"High-quality scientific data are foundational to advancing an AI-enabled research and innovation ecosystem," said Ellen Zegura, senior science and engineering advisor in the NSF Office of the Director. "By helping research communities unlock the value of existing data, we can accelerate breakthroughs, drive innovation and strengthen America's global leaderships."

Why it matters

This represents a strategic shift in how federal science funding approaches the AI era. Instead of treating datasets as single-use products tied to individual studies, NSF is investing in making research outputs reusable assets that can power machine learning across disciplines. The economics are compelling: retrofitting existing data costs less than new collection while potentially multiplying the return on prior research investments. For institutions with deep data archives, this creates an opportunity to extract new scientific value from work already completed.

Alignment with broader AI strategy

The program supports objectives outlined in America's AI Action Plan, particularly around AI-enabled discovery and the creation of structured datasets suitable for training machine learning models. It also complements the Genesis Mission, a White House initiative established by executive order in November 2025 to build an integrated platform for scientific AI applications.

The Genesis Mission aims to train scientific models on curated datasets, deploy AI agents to test hypotheses, and automate research workflows across national science and technology priorities.

Details of the NSF Unlocking Dataset Value for AI-Enabled Scientific Discovery program were announced by the National Science Foundation.