Last month, OpenMined attended the NAIRR Pilot Inaugural Annual Meeting in Arlington, VA. This incredible gathering brought together hundreds of attendees from nearly every state across the country to reflect on the NAIRR Pilot’s first year. As one of the non-governmental partners in this pioneering initiative, we wanted to share what we’ve learned and where we’re headed.
The NAIRR Pilot by the Numbers
The annual meeting report revealed the remarkable scale of what’s been built in just twelve months:
Research Impact:
- 250+ research projects supported across 40 states
- 22 classroom awards bringing AI resources directly into education
- 20 infrastructure projects expanding the nation’s AI capabilities
Demand Signals:
- Nearly 100 million GPU hours have been requested to date, which stands as a testament to the massive unmet need for computational resources in AI research
- Graduate students comprise the majority of active users, representing the next generation of AI talent
Partnership Scale:
- 14 federal agencies collaborating (NSF, DOE, NIH, and more)
- 26 non-governmental partners contributing resources, including OpenMined alongside Microsoft, NVIDIA, AWS, OpenAI, Anthropic, and others
What OpenMined Brings to the NAIRR
Among the diverse resources available through NAIRR, OpenMined addresses a critical gap: enabling research on data that cannot be shared directly.
The annual meeting’s Data Needs & Resources session highlighted a striking reality. There’s a 4-6 orders of magnitude difference between open and non-open datasets available to researchers. Healthcare data behind HIPAA restrictions. Proprietary research datasets. Information distributed across institutional boundaries. This “dark data” represents enormous untapped potential for AI advancement.
Through our NAIRR partnership, we’re offering:
- Free Software for distributed, privacy-preserving data science
- Compute credits for research consortia
- Training sessions teaching researchers how to perform federated learning across multiple data sources while keeping each institution’s data private
Our resources are designed for research consortia—groups of two or more researchers who want to collaborate on AI projects without centralizing sensitive data.
You can apply for these resources here.
Key Themes from the Annual Meeting
Several discussions at the meeting aligned directly with OpenMined’s mission:
The Data Access Challenge
The Data Needs & Resources breakout session identified major challenges:
- Researchers struggle to find relevant, available datasets
- Privacy concerns create significant barriers to data sharing
- There’s no standardized approach to data management for sensitive information
Participants explored “alternative approaches like ‘data visitation’ (sending computation to the data) and federated learning methods,” which align exactly with many of the software solutions OpenMined builds.
Privacy-Preserving AI as a Priority
Several lightning talks highlighted privacy-preserving approaches:
- “Democratizing AI for Cancer with Privacy-Preserving Synthetic Data Generation” from Oak Ridge National Laboratory, using NAIRR Secure resources to develop AI models to automate cancer report generation while addressing HIPAA restrictions
- “Virtual Pooling: Enabling Privacy-Preserving Clinical LLM Training” addressing the bottleneck of unstructured clinical data and privacy concerns in medical LLMs through virtual pooling across healthcare systems
- “LLMs and User Data Privacy: Leveraging Federated Learning with Homomorphic Encryption” from Grambling State University the project performs computations on encrypted data and is researching post-quantum encryption schemes for future security challenges
These projects demonstrate the growing recognition that privacy isn’t just a constraint. It’s a design requirement that model developers and builders should factor into system design.
Democratizing Access
A central theme throughout the meeting was equity in access to AI resources. Currently, 69% of NAIRR requests come from R1 research universities, while 22% come from non-R1 institutions and 5% from non-profits. Participants emphasized the need for institutional partnerships that help less-resourced institutions participate.
What We Learned
The meeting reinforced several insights for our work:
- The Infrastructure Must Support Collaboration David Bau’s keynote emphasized that NAIRR shouldn’t be “a shopping bag for GPUs” but rather a foundation for an AI research ecosystem. OpenMined’s focus on enabling cross-institutional collaboration fits this vision. We do not want to simply provide tools, but rather, we seek to enable fundamentally new ways of doing research
- User Support Matters as Much as Resources Multiple sessions highlighted that researchers need more than access. They need training, documentation, and ongoing support. Our commitment to providing training sessions with every deployment reflects this reality.
- The Demand for Privacy-Preserving AI is Growing From healthcare to agriculture to social sciences, researchers across domains are hitting the same wall: the data they need is locked behind legitimate privacy constraints. The 100 million GPU hours requested suggest the compute is available; the question is whether researchers can access the data to make use of it.
Looking Ahead
As the NAIRR Pilot enters its second year, we’re energized by what’s possible. Congressional efforts are underway to establish NAIRR as a permanent program, recognizing its importance to U.S. competitiveness in AI.
For OpenMined, this means continued investment in making our tools more accessible to researchers who may not have deep technical backgrounds, building case studies that demonstrate what’s possible when data barriers come down, and supporting the growing community of researchers using privacy-preserving AI through the NAIRR.
If you’re a researcher working with sensitive data, such as, healthcare records, proprietary datasets, information distributed across multiple institutions, we want to hear from you!
Whether you’re already part of NAIRR or considering applying, OpenMined can help you unlock insights from data you couldn’t access before.
Get started: Visit nairrpilot.org to submit a proposal, or reach out to us directly through the OpenMined NAIRR page.
