Simons Foundation Opens $6 Million Empire AI Research Program

Key Takeaways

• The Simons Foundation is offering $6 million for two-year basic science and mathematics research programs using Empire AI infrastructure.
• Eligible proposals will develop or use foundational machine learning models on Empire AI’s multinode GPU clusters.
• Applications are due December 1, 2026, at noon Eastern time, with participation centered on Empire AI member institutions.

The Simons Foundation has opened a $6 million funding opportunity designed to push New York State researchers toward more computationally ambitious uses of artificial intelligence.

Called “Utilizing the Full Power of Empire AI,” the request for applications will support two-year programs in computational astrophysics and physics, computational biology, mathematics, neuroscience, and computational physical chemistry. Projects can focus on one field or work across several disciplines.

There is an important condition: proposals are expected to rely on the development and use of foundational machine learning models. Applicants also need to explain how their research will use Empire AI’s multinode GPU clusters, rather than treating computing capacity as a generic project expense.

That focus distinguishes the program from a conventional science grant. The Simons Foundation is effectively asking researchers to propose work that benefits from distributed, accelerator-intensive computing at a scale that can be difficult for individual laboratories to obtain. What scientific questions become practical when a research team gains access to substantially more training, inference, and storage capacity?

Empire AI was launched in 2024 as a statewide consortium intended to advance AI and AI-supported technologies serving the public good. It now connects 10 New York research institutions and is backed by more than $500 million in combined public and private investment, according to information associated with Governor Hochul’s Empire AI Initiative.

The Simons Foundation has already committed $75 million to support participation by the City University of New York and the Flatiron Institute, its in-house computational research division. The new $6 million program adds a project-level funding mechanism to that broader infrastructure commitment.

Compute is the practical center of the initiative. The $40 million NVIDIA-powered Empire AI Beta system at SUNY Buffalo offers an 11x increase in AI training capacity, a 40x boost in AI inference, and an 8x expansion in storage over the initial Alpha system. Hundreds of research projects across health care, climate, advanced manufacturing, and other fields are already queued to use Empire AI resources.

Access to GPUs alone does not produce useful science. Research groups also need suitable data, scalable model architectures, reproducible workflows, experienced personnel, and a clear reason for using a foundational model in the first place. The RFA’s two-year structure may encourage applicants to connect those technical and personnel requirements around defined scientific problems rather than open-ended infrastructure experimentation.

Eligibility is broad within the consortium. Tenure-track and tenured faculty at the City University of New York, Cornell University, Columbia University, the Icahn School of Medicine at Mount Sinai, New York University, Rensselaer Polytechnic Institute, the Rochester Institute of Technology, the State University of New York, and the University of Rochester can apply.

Flatiron Institute employees may join proposals as unfunded collaborators. Additional principal investigators can work through subcontracts from institutions outside Empire AI, including U.S. and non-U.S. nonprofit organizations, colleges, universities, hospitals, laboratories, institutes, state and local government units, and eligible federal agencies. The program has no citizenship requirement.

That flexibility could be particularly useful for interdisciplinary projects. A computational biologist at an Empire AI institution, for example, could bring in specialized collaborators from a hospital or overseas nonprofit while retaining the eligible consortium-based lead investigator.

Governance will matter as projects move from proposal to execution. The NIST AI Risk Management Framework provides a voluntary structure for identifying and managing AI risks, while ISO/IEC 22989:2022 establishes common terminology and concepts for artificial intelligence. Neither substitutes for discipline-specific scientific controls, but both can help teams describe model behavior, oversight, documentation, and accountability consistently.

For technology vendors and research administrators, the program also illustrates how shared AI infrastructure is becoming an institutional service rather than a collection of isolated laboratory purchases. NVIDIA supplies the underlying GPU technology, but the larger operational challenge includes scheduling, storage, data movement, software environments, model evaluation, and equitable allocation among competing projects.

Applications are due December 1, 2026, at noon Eastern time. The strongest proposals are likely to pair a fundamental scientific question with a credible plan for multinode computation, collaboration, and responsible model development. The funding is meaningful, but access to Empire AI’s expanding infrastructure may prove just as consequential for the selected research teams.