NERSC Issues 2026 Call for AI for Science Proposals

Why It Matters

Providing dedicated AI‑focused HPC resources accelerates breakthroughs across climate, materials, biology and other DOE‑priority fields, while the requirement for publicly available datasets expands the community’s shared data pool.

Key Takeaways

• Up to 10,000 Perlmutter GPU node hours offered.
• Additional 20,000 CPU node hours for AI‑ready datasets.
• Open call; non‑users may apply.
• Proposals due April 30 2026; resources allocated through Jan 2027.
• NERSC staff will advise on HPC usage, not model building.

Pulse Analysis

NERSC’s 2026 AI for Science initiative reflects a broader shift toward integrating large‑scale artificial intelligence with high‑performance computing. By earmarking thousands of GPU hours on Perlmutter—a system built around four NVIDIA A100 GPUs per node—researchers can train sophisticated deep‑learning models that were previously constrained by compute limits. This infusion of AI capability into a national user facility not only democratizes access but also aligns with the Department of Energy’s push to modernize scientific discovery pipelines, from climate modeling to quantum materials design.

The program’s dual‑track resource allocation—GPU time for model training and CPU time for generating AI‑ready datasets—encourages a full lifecycle approach. Teams are expected to produce datasets that will be openly shared, amplifying the impact beyond the original project and fostering reproducibility. By allowing non‑users to apply, NERSC widens its user base, inviting fresh perspectives from academia, industry, and emerging research groups. The structured proposal requirements, including clear deliverables and timelines, ensure that awarded projects can demonstrate tangible scientific outcomes within the 2026 allocation year.

Strategically, the call supports DOE’s mission to maintain U.S. leadership in computational science. Access to Perlmutter’s massive parallelism enables experiments that blend physics‑based simulations with data‑driven inference, a combination increasingly essential for tackling grand challenges like fusion energy and climate resilience. Prospective applicants should emphasize alignment with DOE priorities, showcase prior AI successes, and outline robust data‑management plans to maximize the likelihood of funding. Successful projects will not only advance their own research agendas but also enrich the national scientific ecosystem through shared tools and datasets.