Simons Foundation's $6 Million Empire AI Call: A December 1 Deadline for New York Scientists Who Need Multinode GPUs
November 19, 2026 · 6 min read
Granted Research Team · Editorial policy
The Simons Foundation's newest competition has one requirement that will rule out most machine-learning-for-science proposals. Every submission, the request for applications says, "should detail how their work will use multinode clusters of GPUs in Empire AI." A model that trains on one workstation card, or on a single node of a campus cluster, does not fit this call.
The program is Utilizing the Full Power of Empire AI – 2027. Its budget is about $6 million over two years, and Simons expects to fund six to 12 projects. Applications are due at 12 p.m. Eastern on December 1, 2026, which is 12 days from today. The award period runs from June 1, 2027 to May 31, 2029, and decisions arrive March 1, 2027.
The call has drawn more attention because of what happened two days ago. On November 17, Simons, Simons Foundation International, the Secunda Family Foundation and Bloomberg Philanthropies announced a $70 million effort to create 120 research positions across New York for AI-enabled science. Together, the two programs give a clear picture of what private philanthropy is building around the state's public AI computing consortium.
Who Can Apply: Nine Institutions, Faculty Only
Eligibility is narrow. The principal investigator must hold a tenure-track or tenured faculty position at one of the nine Empire AI member universities:
- Columbia University
- Cornell University
- City University of New York (CUNY)
- State University of New York (SUNY), across its campuses
- New York University
- Rensselaer Polytechnic Institute
- University of Rochester
- Rochester Institute of Technology
- Icahn School of Medicine at Mount Sinai
The Flatiron Institute, Simons' own research arm, belongs to the consortium, but its employees can serve only as unfunded collaborators. No Simons W-2 employee can be PI.
Collaborators can come from much farther away. Additional PIs at outside institutions, including non-U.S. nonprofits, can be funded through subcontracts, and the RFA says "there are no citizenship requirements." Every applicant and key collaborator must hold a Ph.D., M.D. or equivalent and a faculty-equivalent position. Simons will not fund institutions in China or Hong Kong, including through subcontracts.
The public pages set no per-institution limit. Several member schools, including the University at Albany and the University of Rochester, run their own internal routing or review, so check with your sponsored-programs office this week. With only 12 days left, any internal deadline may already have passed.
Scientific Scope: Foundation Models in Simons' Home Fields
The RFA lists five eligible fields: "computational astrophysics and physics, computational biology, mathematics, neuroscience and computational physical chemistry," along with their intersections. The examples it gives show how wide the call reaches inside those fields: "computational biophysics, neuroscience, ecology and evolution, plant biology, optimal transport, multiscale physics, fluid dynamics and numerical general relativity."
There are two more conditions. Proposals must "rely on the development and utilization of foundational machine learning models." Using an off-the-shelf classifier as an analysis tool is not enough. Pure data-curation projects qualify only "in exceptional cases," and only when the same team will use the data to solve the scientific problem in the proposal.
That rules out clinical medicine, engineering and most social science, even at member institutions. A Mount Sinai investigator working on a protein-structure foundation model fits. A Mount Sinai investigator building an imaging model for hospital triage probably does not.
No Award Cap, So Size the Budget to the Compute
There is no maximum per award. "The total amount requested should reflect the proposed research," the RFA says, and awards include indirect costs at Simons' nonnegotiable indirect rate. With $6 million spread across six to 12 projects, the arithmetic implies an average of about $500,000 to $1 million per project over two years. Asking for $2 million amounts to asking reviewers to fund you instead of two other teams, so that request needs a matching scientific case.
Simons does not pay for the GPUs. Empire AI provides the hardware to member researchers through shared infrastructure housed at the University at Buffalo. The grant pays for the people and the work that make use of that hardware. The strongest budgets will match the size of the compute: postdocs and graduate students who can write distributed training code, data engineering, and whatever time the PI needs to run a project at that scale. The RFA and FAQs don't say how GPU time is allocated, so applicants should confirm their expected Empire AI access with their institution's consortium contact before claiming it in the narrative.
What the Application Requires
Submissions go through Simons' SAM portal (sam.simonsfoundation.org), and the institution's Signing Official must sign off before noon on December 1. The components:
- Proposal narrative on the SAM template, which holds the page limits and formatting rules
- Biosketches in SciENcv, NIH or NSF format
- Resources and research environment statement, also on a template. This is where you describe your multinode workflow and your Empire AI access
- Disclosure of generative AI use in preparing the application
- Disclosure of any Simons or Flatiron affiliation
- A data and renewable-reagents plan where applicable
- A whole-dollar budget for each period, plus publications and other support
Simons scientific staff and an expert committee will review the applications. The public pages publish no formal scoring rubric. Given the multinode requirement, reviewers will almost certainly look first for evidence that the team has run distributed training before. Preliminary results from a smaller multi-GPU run, or a scaling curve, will count for more than a sentence of ambition.
Note the timing of the foundation's office closures: December 24 to January 1, and January 18. Those dates fall after the deadline, but they mean questions sent during the holidays will not get a reply.
How the $70 Million Fellowship Push Fits In
The November 17 announcement is a separate program, and applicants should not confuse it with the RFA. It has two parts:
- The Simons Empire Faculty Fellowship Program, funded by Simons and SFI, pays for 58 early-career faculty positions at 18 institutions across New York, a wider group than the nine consortium members.
- The Empire Postdoc Fellows Program, funded by the Secunda Family Foundation and Bloomberg Philanthropies, places more than 60 computationally focused postdocs in multidisciplinary labs at Empire AI schools.
"Empire AI is giving New York state's researchers access to computing power that no single institution could build on its own," Empire AI director Kiran Keshav said. Board chair Tom Secunda gave the reason behind the fellowships: "we also need to invest in the scientists who will put it to work."
For PIs, the two programs feed each other. A lab that wins an RFA award this spring will be well placed to host an Empire Postdoc Fellow. A newly hired Simons Empire faculty fellow will be a natural co-PI on the next RFA cycle. Philanthropy is building a full staffing pipeline around a state-backed computing consortium, which is still unusual. The bet is that the people will matter as much as the hardware.
For researchers outside New York, the model is worth watching. Federal AI compute access through the NSF NAIRR pilot and the national labs is competitive and allocated in smaller slices. A regional consortium backed by private science funders offers faster access, but only to researchers at member schools.
If you are at a member institution with a foundation-model project that already needs more than one node, the December 1 deadline is tight but realistic. Teams pulling together the narrative and resources statement this week can use Granted to draft and tighten each section before the portal closes.