AI Policy @🤗: Response to the 2025 National AI R&D Strategic Plan
As part of the national effort to shape America's AI future, Hugging Face submitted its response to the RFI on the 2025 National AI Research and Development Strategic Plan. Just as roads and electricity served as foundational infrastructure for the industrial economy, open models, compute, and data must be recognized and supported as public infrastructure for the emerging AI economy. We believe that national leadership in AI depends on fostering an open, accessible, and responsibly governed ecosystem to boost innovation and economic growth that delivers public value, not just private returns.
This blog post provides a summary of our response, the full text is available here.
Why Open, Publicly-Supported AI Matters
Open AI systems are technically competitive and economically transformative. Recent models like Hugging Face's OlympicCoder, which outperforms Claude 3.5 on complex coding tasks with only 7 billion parameters, and AI2's transparent OLMo 2, which rivals the performance of similarly sized GPT4o and Qwen 2.5 models, showing that open approaches can rival proprietary alternatives with far fewer resources. Research has shown that open technical ecosystems are a boon for GDP growth, and they act as economic force multipliers by distributing AI benefits far beyond tech giants to startups, universities, and communities nationwide, creating a decentralized ecosystem that fosters transparency, strengthens local economies, and makes AI a driver of shared prosperity rather than concentrated wealth.
The Role of Federal Funding in Strategic AI Research
While private industry will continue advancing AI, markets alone cannot meet all societal needs. High-impact areas like public health, climate adaptation, and fundamental scientific research often deliver immense public value but limited financial return, making them unattractive to private capital. The federal government has historically filled this gap, backing foundational technologies like GPS and the internet. AI demands a similar commitment. Strategic federal support through funding, infrastructure, and open governance can ensure AI development reflects public interest by addressing market failures and investing where the private sector won't.
Four Strategic Priorities for National AI Leadership
1. Efficient, Transparent, and Accessible AI Research
To ensure AI benefits all Americans, we must move away from the trend toward ever-larger proprietary models. Smaller, efficient models are better suited for resource-constrained environments like rural clinics, community colleges, or small businesses. The open ecosystem already fosters this innovation, driven by necessity and creativity. Future research should focus on optimizing for specific use cases rather than generic performance benchmarks to bring AI within reach for far more people.
Examples from our recommendations:
- Develop compression techniques that drastically reduce model size and cost, enabling them to run on mobile and edge devices.
- Research alternative architectures like Mamba (state space models) and diffusion-based language models that offer better efficiency than transformers.
- Advance scaling laws and predictive tools to estimate what model size, data, and compute are needed for specific capabilities.
2. AI for Science, Health, Climate, and Resilience
AI's greatest potential lies in solving high-impact problems that often fall outside market incentives. From pandemic response to climate modeling, many of America's most pressing challenges lack sufficient AI investment because they don't promise immediate profit. This is where federal leadership is critical. With sustained funding, the government can support AI systems that work in solving problems for the American people. These systems aren't just useful but vital to national security, public safety, and long-term resilience.
Examples from our recommendations:
- Fund AI-powered materials discovery for strategic technologies like semiconductors and clean energy batteries.
- Develop domain-specific foundation models trained on scientific data for open access, such as in biology, medicine, chemistry, physics, and other basic sciences.
- Support open AI systems for early disease detection, drug discovery, privacy-preserving epidemiology, and climate modeling using satellite imagery and sensor data.
3. Trustworthy, Secure AI for Public Institutions
AI in critical infrastructure must meet much higher standards than consumer apps. Public systems used in transportation, emergency response, or healthcare require AI that is reliable under pressure, explainable in its decision-making, and secure against attack. Today's models often prioritize raw power over robustness or interpretability, making them a poor fit for the public sector. When AI influences who gets hired or what healthcare treatment is offered, the public deserves clear and accountable processes.
Examples from our recommendations:
- Build verifiable and auditable AI systems using chain-of-thought reasoning and neurosymbolic methods.
- Develop defenses against adversarial attacks, such as data poisoning or model theft.
- Implement AI system documentation standards that track model lineage, data provenance, and resource usage for full transparency and reproducibility.
4. Strengthening AI Infrastructure and Understanding Societal Impact
America’s leadership in AI depends not just on innovation at the frontier, but on who gets to participate and how we measure the consequences. Today, structural inequities in access to compute, data, and tooling restrict innovation to a narrow segment of institutions, while the broader societal effects of AI remain poorly understood.
Examples from our recommendations:
- Expand NAIRR into a permanent, publicly funded AI compute backbone, with allocations for high-risk and interdisciplinary research.
- Fund large-scale initiatives to digitize and publicly release data that is currently trapped in analog or inaccessible formats, including public records and cultural archives, to make them AI-ready for public benefit research and training.
- Fund longitudinal studies on AI's effects on employment, skills, and regional economies to inform proactive policy.
Looking Ahead
The 2025 National AI R&D Strategic Plan is a historic opportunity. By investing in open, secure, and accessible AI systems, the U.S. can lead in innovation, impact, and integrity. A future where AI works for everyone is within reach if we build it together.
Read our full response for detailed recommendations and technical guidance.
