The National Science Foundation has awarded Stony Brook University a grant, in collaboration with the University at Buffalo, to deploy and operate a new national supercomputing resource for artificial intelligence and research computing. The system, Kitsune, will be the flagship cluster of the Stony Brook Supercomputing Center, the university's new Tier 1 supercomputing center (award OAC-2505376).
The initial $13.77 million goes primarily toward purchase of the system and first-year activities, with additional funds for operations over five years, subject to external review. The NSF award record shows $21.95 million obligated to date. The award was made on July 24, 2025 through NSF's Advanced Computing Systems and Services program (Category I, capacity resources), under the title Sustainable Cyber-infrastructure for Expanding Participation.
Much of the work researchers actually run is scalar code, code that vectorizes imperfectly, or, increasingly, AI inference: workloads that machines built for large parallel simulations handle poorly and at high cost. Kitsune is designed for that long tail of science, the high-throughput, memory-hungry computing common in genomics, bioinformatics, statistics, and computational linguistics, as well as mainstream parallel work. After an initial testing period it will be allocated through the NSF ACCESS program, serving researchers, students, and educators across the country, including institutions that have never had this kind of computing within reach.
The award is led by principal investigator Robert J. Harrison, director of the Institute for Advanced Computational Science, with co-principal investigators Alan Calder at Stony Brook and Matthew Jones and Nikolay Simakov at the University at Buffalo's Center for Computational Research.
"This project employs a comprehensive, multilayered strategy, with regional and national elements to ensure the widest possible benefits," said Harrison. "The team will collaborate with multiple initiatives and projects, to reach a broad audience that spans all experience levels from high school students beginning to explore science and technology to faculty members advancing innovation through scholarship and teaching."
"The University at Buffalo is excited to partner with Stony Brook on this new project that will advance research, innovation and education by expanding the nation's cyber-infrastructure to scientific disciplines that were not high performance computing-heavy prior to the AI boom," said Simakov.
As planned, Kitsune comprises 162 compute nodes, each with two AMD EPYC 9965 processors (Turin, Zen 5) for 384 cores per node, more than 60,000 cores in total and a peak of 6.7 petaFLOPS in double precision, with 126 terabytes of memory, six petabytes of usable parallel storage, an InfiniBand NDR-400 fabric, and Rocky Linux 10. The design favors versatility and throughput: scalar, vectorized, and memory-intensive workloads; many independent jobs alongside large parallel ones; AI inference on the CPU; and strong energy efficiency and performance per dollar.
The design has evolved since the award was made. The project began under the working name AMA27; the system is now Kitsune (狐), Japanese for fox. The name continues a tradition begun with Ookami (狼, wolf), the center's earlier NSF-funded ACCESS resource.
Kitsune is planned to come online in the first quarter of 2027. Timeline, specifications, and the launch-update sign-up are on the Kitsune page.