The Simons Math+AI Open Call Is Here, and It Is Only for Institutes With a Building: $500,000, Up to Five Awards, December 1

When the Simons Foundation, Anthropic and OpenAI announced a combined $15 million commitment to help the mathematics community deal with AI on October 15, 2026, the release promised "an open call for proposals" but published no guidelines. The guidelines are now live, and they answer the question every mathematician asked that week: can I apply?

For almost everyone, the answer is no. The call is titled Targeted Grants to Math+AI Institutes — 2027 Request for Applications, and the word that matters is institutes. The money goes to established research institutes and centers with a physical space, to run workshops, visitor programs and other shared activities that help the field evaluate and absorb AI-assisted mathematics. Individual faculty, research groups, virtual networks and university centers built around one department's research are all out.

The deadline is December 1, 2026, at 12 p.m. Eastern. With the foundation's office closed November 25–27 for Thanksgiving, eligible institutes have about three and a half working weeks to write one.

The Call in Brief

Program: Targeted Grants to Math+AI Institutes, 2027 RFA. The Simons Foundation calls it a one-time call, created in partnership with Anthropic and OpenAI.

Purpose: Workshops and other initiatives that help researchers evaluate AI-assisted results against the mathematical community's standards, learn and use new tools, and shape shared standards for AI use in mathematics. The RFA says plainly that it does not provide primary support for operating or founding an institute.

Award size: Up to $500,000 total over two years, including 20% indirect costs. Annual budgets should be around $250,000, though the split between years can vary as long as the total stays within $500,000.

Number of awards: Up to five from the Simons Foundation. Anthropic and OpenAI may each make additional awards under their own policies, and with the applicant's consent, applications may be shared with them.

Period: Two years, starting no earlier than March 1, 2027 and no later than June 1, 2027. Not renewable.

Deadline: Full proposals due December 1, 2026, 12 p.m. ET, through the Simons Award Manager (SAM). There is no letter of intent stage. Notification is expected by December 31, 2026.

Review: Simons Foundation scientific staff, the Mathematics and Physical Sciences (MPS) Scientific Advisory Board, and an external expert review committee.

Do the Math on the Headline

The October announcement said $15 million, with $5 million each from Anthropic and OpenAI and the foundation managing and awarding the money. This call, though, can award at most five grants of $500,000 each, or $2.5 million: one sixth of the headline.

That isn't a bait-and-switch. The October release described the $15 million as "funding and resources." It listed workshops, conferences and special programs alongside the open call, and the companies' contribution was never described as all cash for outside grantees. But applicants should set their expectations accordingly. This is a selective, small-n competition, not a broad funding program. If the Simons Foundation draws applications from even a modest share of the world's mathematics institutes, odds of one in four or worse are plausible.

The partner clause is the wildcard. Anthropic and OpenAI "may" fund additional applicants "under their own policies." Nothing on the RFA page commits either company to a number, a dollar figure or a timeline. Applicants should consent to sharing, because there is little downside, but should budget their effort as if only the five Simons slots exist.

Who Is Actually Eligible

The RFA and its FAQ draw the line in several places.

In:

• Established U.S. and foreign public and private educational and nonprofit institutes and research centers.
• Centers or institutes inside larger universities, if their main purpose is to serve a field by strengthening contacts across the scientific community.
• Institutes that already hold, or applied for, a Simons Targeted Grant to Institutes, provided the Math+AI proposal does not overlap that award's scope.
• Joint proposals from multiple institutes within one university.

Out:

• Initiatives, consortia or organizations without a physical space.
• Centers whose primary purpose is to focus the research of one university's own faculty.
• National labs or equivalents with a primarily experimental focus.
• Institutions in China, along with their fiscal sponsors, agents, subcontractors and consultants.
• Simons Foundation and Flatiron Institute W-2 employees, as PI or as funded personnel.

Limit: One application per institute or center. A university may submit one from each distinct institute it hosts.

In practice, the profile is a visitor-driven mathematics institute: a place with offices, seminar rooms and a stream of programs that bring outside mathematicians together. Think of the U.S. institutes that run thematic semesters and workshop series, their counterparts in Europe, Canada and Asia, and the smaller university-hosted centers that run summer schools and research-in-residence programs for the whole field, not just their home department.

The FAQ adds one more signal: for institutes in developing countries, the foundation also considers how the grant can stimulate science there and connect researchers to the mainstream. An institute in, say, Latin America, Africa or South Asia that can position itself as a regional hub for AI-literate mathematics has an argument that a well-funded North Atlantic institute cannot make.

What the Money Buys, and What It Doesn't

The allowable-cost list tells you the shape of a winning proposal.

Allowed:

• Salary and benefits for visiting scholars, postdocs, research associates and graduate students.
• Visitor travel, meals and lodging.
• Conferences, workshops, summer schools and meetings, including facility fees.
• Administrative professional staff who support institute activities.

Not allowed:

• PI summer salary for administering the program.
• Salary for scientific leadership. The RFA expects that to come from the institute's existing structure.
• Insurance against exchange-rate costs, per the FAQ.

Indirect costs are fixed at 20% and included in the $500,000 cap. The FAQ calls the rate nonnegotiable.

Read together, the rules say this is money for people moving through a building. The foundation wants to pay for a postdoc who becomes the institute's resident expert on formal verification, a semester program where number theorists and Lean developers work side by side, a summer school that teaches graduate students to check machine-generated proofs, and the coordinator who makes all of it happen. It does not want to pay the director, buy GPUs, or fund the institute's own research agenda.

Compute is conspicuously absent from the allowable-cost list. If your proposal depends on running models at scale, you will need to show where that capacity comes from. One plausible route is in-kind access from the AI partners, but the RFA promises nothing of the kind.

Why the Design Makes Sense

The October trigger was OpenAI's October 6 release of 722 AI-generated manuscripts claiming results on hundreds of open problems, which landed on a field with no shared process for checking them. Our earlier analysis argued that the $15 million funds checking the work, not doing it. The RFA confirms that reading and adds a structural choice: route the money through the institutions that already set mathematics' social norms.

Mathematics institutes are where the field decides, informally, what counts. Their workshops are where a new technique gets stress-tested by people who did not invent it. Their programs decide which subfields get a semester. Funding them to build AI-evaluation capacity is a bet that standards will emerge from convenings, not from journals or from the companies themselves. It also keeps Anthropic and OpenAI at arm's length: the Simons Foundation and its advisory board pick the grantees, and the companies' optional add-on awards sit outside that process.

For an institute director, that framing is the pitch. Proposals that read as "we will host an AI workshop" are generic. Proposals that read as "we will become the place where subfield X builds its verification standard, and here is the two-year program that gets there" answer the question the review committee is asking.

A Four-Week Proposal Plan

Week of November 2: decide and register. Confirm your institute meets the physical-space and community-service tests. If you sit inside a university, get your sponsored-programs office to confirm that only one proposal is going out from your center, and coordinate with sister institutes if a joint proposal makes more sense. Set up or confirm access to SAM now; the FAQ warns that records lock when two users edit at once and that nothing can be submitted after the deadline.

Week of November 9: design the program. Pick one or two concrete Math+AI problems your institute is positioned to own. Good candidates map directly to the RFA's four verbs:

• Evaluate: structured review of AI-claimed results in a subfield, with a public output.
• Learn: summer schools or bootcamps on proof assistants and AI tools for graduate students and early-career faculty.
• Share: workshops that bring together mathematicians on opposite sides of the AI debate.
• Shape standards: working groups that draft attribution, disclosure or verification norms for journals and hiring committees.

Week of November 16: budget and personnel. Build a two-year budget near $250,000 per year. Weight it toward postdocs, visitors and program costs, with a part-time coordinator if you need one. Make sure no line item pays the director or scientific leadership. Line up named participants for at least the first workshop; letters from mathematicians outside your institution show the community-serving function the eligibility rules require.

November 23–30: write, review, submit. The foundation's office is closed November 25–27, so get technical questions in before November 24. Submit by November 30 to leave a day of margin. Deadlines are U.S. Eastern regardless of your time zone.

The Other December 1 Simons Deadline

The same day, the foundation closes a second AI-related call with a very different audience: Utilizing the Full Power of Empire AI, 2027 RFA. It funds mathematics and basic science research that uses Empire AI's multinode GPU clusters, New York State's academic AI computing consortium.

• Total budget: $6 million over two years, across six to 12 awards. There is no per-award cap; budgets should match the proposed research, and indirect costs are included.
• Eligibility: The lead PI must hold a tenured or tenure-track faculty position at an Empire AI member: Columbia, Cornell, CUNY, SUNY, Rensselaer, NYU, the University of Rochester, RIT, or the Icahn School of Medicine at Mount Sinai. Co-PIs at other institutions, including outside the U.S., can join by subcontract.
• Timeline: Applications opened September 1, 2026, and close December 1 at noon ET. Decisions are expected March 1, 2027, and the RFA page lists an award period of June 1, 2027 to May 31, 2029. One secondary listing cites an August 1 start, so confirm with the foundation.

The two calls are complements. Math+AI Institutes pays for the community to evaluate AI-generated mathematics. Empire AI pays individual researchers to use frontier-scale compute themselves. A New York institute could plausibly pursue both, but the overlap rule for Simons institute grants suggests keeping the scopes cleanly separated.

What to Watch

• Partner awards. Whether Anthropic or OpenAI fund declined applicants, and how many, will show whether the companies' $10 million is flowing to outside institutions or mostly into their own programming.
• The December 31 notification. It is a fast turnaround for a call reviewed by staff, an advisory board and an external committee. A slip would hint at heavy volume.
• Spillover. If AI-generated results start hitting physics, chemistry or biology at the same scale, expect copycat "absorption" calls in those fields. Institutes that win here will have a track record to cite.

The October announcement looked like money for the whole mathematics community. The RFA reveals that it is money for the community's institutions, with a short list of perhaps five winners. If you run one of those institutions, the next four weeks are worth the effort. If you don't, the most useful thing to do is call the director of the institute where you'd most like to spend a week, and offer to help write the workshop they'll need you at.