NERSC Opens Perlmutter Supercomputer for AI-for-Science Proposals
March 21, 2026 · 2 min read
Arthur Griffin
The National Energy Research Scientific Computing Center has issued its 2026 call for AI-for-science proposals, offering researchers up to 10,000 GPU node hours on the Perlmutter supercomputer — one of the most powerful machines available for scientific AI training. Proposals are due April 30, 2026.
What NERSC Is Offering
Each accepted project receives up to 10,000 Perlmutter GPU node hours (each node contains four NVIDIA A100 GPUs) along with associated storage quotas. Projects also receive up to 20,000 CPU node hours, with the possibility of additional allocation depending on resource availability.
NERSC specifically encourages teams to use Perlmutter's CPU-only nodes to generate AI-ready datasets, with the expectation that resulting datasets will be made publicly available — building a shared scientific data commons for the broader research community.
The allocations cover the NERSC 2026 Allocation Year, running through January 19, 2027, providing roughly nine months of active computing time for selected projects.
Who Should Apply
NERSC is looking for teams that combine deep learning expertise with strong domain knowledge in a scientific discipline. Applicants need demonstrated proof-of-concept results — this is not a program for exploratory work. The center wants projects ready to scale.
Critically, applicants do not need to be current NERSC users. This opens the door for university research groups, national laboratory teams, and independent researchers who have been developing AI models on smaller systems and need production-scale computing to validate or extend their work.
How This Fits the Federal AI-for-Science Pipeline
The NERSC call complements DOE's broader AI-for-science investments, including the recently announced $68 million in grants for AI research projects and the $320 million Genesis Mission. Together, these programs create a pipeline: researchers can develop AI methods with DOE grant funding and access the computing infrastructure to run them through NERSC allocations.
Teams interested in applying should visit NERSC's proposal page and submit by April 30, 2026. No prior NERSC relationship is required. Track AI research funding opportunities at grantedai.com.