The $50M Cities Deliver Call Has a Six-Week Hidden Gate: a Signed Letter From a City Government
October 15, 2026 · 7 min read
Granted Research Team · Editorial policy
On October 14, 2026, The Rockefeller Foundation and the OpenAI Foundation opened Cities Deliver: AI Solutions for Public Services — a $50 million global call for proposals administered by RF Catalytic Capital, Rockefeller's grantmaking affiliate.
Round 1 proposals are due November 29, 2026 at 11:59 PM US Eastern Time. Questions to the program close October 31, with the FAQ updated November 4 at 8:00 PM ET.
That is a six-week window. And buried in the eligibility structure is a requirement that will decide most outcomes before anyone reads a single technical narrative: you must produce a letter of intent from a city government that is not permitted to apply alongside you as a co-lead.
The money, precisely
| Item | Value |
|---|---|
| Program | Cities Deliver: AI Solutions for Public Services |
| Funders | The Rockefeller Foundation + OpenAI Foundation |
| Administrator | RF Catalytic Capital (RFCC) |
| Total commitment | $50,000,000 |
| Seed track | $500,000 to under $1,000,000 |
| Scale track | $1,000,000 to $2,000,000 |
| Grant term | 24 months |
| Disbursement | 50% upfront, 50% on mid-term milestones |
| Call opens | October 14, 2026 |
| Questions due | October 31, 2026 |
| Round 1 due | November 29, 2026, 11:59 PM ET |
| Interview notifications | December 2026 |
| Interviews | January 2027 |
| Round 2 invitations | February 2027 |
| Round 2 proposals due | March 1, 2027 |
| Final selections | April 2027 |
| Stated reach goal | Up to 100 million city residents |
Do the division. At the seed floor of $500,000, $50 million supports 100 grants. At the scale ceiling of $2 million, it supports 25. A realistic portfolio weighted toward the middle lands somewhere near 40 awards — which, against a stated goal of reaching 100 million residents, implies an average of roughly 2.5 million residents per funded project. That is not a rounding error in the program design. It is a statement about the size of city the reviewers expect to see named in your letter of intent.
The gate: a city partner that cannot apply
The eligibility architecture here is unusual and worth stating plainly.
Eligible lead applicants are nonprofits — including university labs and fiscally sponsored organizations — and public benefit corporations. For-profit companies may apply only in narrow circumstances where the project remains charitable in purpose and any business benefit is incidental.
City governments cannot apply directly. They also cannot be the lead. But every project must be delivered in partnership with "a city government partner committed to the outcomes and impact," and Round 1 requires a letter of intent from that city government (or an equivalent document) alongside proof of the lead organization's legal status and a short video demonstrating the AI solution.
This inverts the usual power dynamic in civic technology funding. Most government-technology money flows to the government, which then procures. Here the money flows to a civil-society intermediary, and the government's contribution is political commitment, operational access, and a signature — none of which are things a municipal legal department produces quickly.
Six weeks is a tight runway for a city attorney to clear a letter referencing a named AI deployment. If you are reading this and do not already have a working relationship with a city agency, the honest assessment is that Round 1 of this cycle is likely not yours. The more productive move is to treat the November 29 deadline as a forcing function for relationship-building and watch for a second cycle — the program's four-stage structure running through April 2027 suggests an institutional commitment rather than a one-off.
If you do have the relationship, the sequencing matters: get the letter of intent into the city's queue this week, and build the proposal around it rather than the reverse. A technically brilliant application with no signature is screened out at eligibility, before any reviewer evaluates the idea.
"Frontier AI must be core" is a disqualifier, not a flourish
The second structural filter is the AI requirement, and it is stricter than the usual grant-cycle gesture toward innovation.
The call requires projects to have a meaningful frontier AI component — recent AI models completing complex tasks — where those capabilities are core to the proposed solution. The review criteria reinforce this: the project must take advantage of the latest capabilities of frontier models, and those capabilities must be central rather than decorative.
This cuts a wide swath through the civic-tech proposal population. Dashboard modernization, data-warehouse consolidation, a decision tree with a chat interface bolted on top, a classifier trained in 2021 — these are useful projects that will likely fail this test. The question a reviewer is being asked to answer is whether the solution would collapse without frontier model capabilities, not whether AI appears somewhere in the architecture.
One genuinely important detail: there is no requirement or preference to use OpenAI technology. Applicants choose their own provider. For a program co-funded by the OpenAI Foundation, this is a meaningful piece of program hygiene, and it means an Anthropic- or Google-based architecture is not disadvantaged on its face. Do not contort your stack to flatter a funder that has explicitly said it does not want that.
Projects related to voting or elections are excluded. Full stop. Civic engagement is one of the four focus areas, but it is scoped to service channels, participatory decision-making, permitting, court services, and digital connectivity — not electoral administration.
The four focus areas, and what they actually signal
| Area | Named examples |
|---|---|
| Public health | Frontline care delivery, disease surveillance, disability services, food and nutrition access |
| Economic mobility | Benefits enrollment, career navigation, skills-based hiring, employer training |
| Environmental resilience | Emergency services, energy infrastructure, land management in at-risk communities |
| Civic engagement | Service channels, participatory decision-making, permitting, court services |
Read these as a list of administrative bottlenecks, not research domains. Benefits enrollment, permitting, and disease surveillance share a structure: a queue of human-mediated determinations where the cost per determination is high and the error rate from under-resourcing is the actual public harm. That is the shape of problem this portfolio is built to absorb.
What the three pilots reveal about the selection taste
Cities Deliver did not emerge from a strategy memo. It generalizes three pilot projects run between roughly 2024 and 2026:
- Cali, Colombia — Dengue.AI. Forecasts dengue outbreaks up to three weeks in advance, letting public health officials target prevention geographically before cases appear.
- Cape Town, South Africa. An AI-powered messaging platform for two-way communication between residents and the city.
- Rio de Janeiro, Brazil. A WhatsApp chatbot that connects flood alerts to emergency response, with planned expansion to additional extreme weather events.
All three share a pattern worth internalizing: each sits on a communication or prediction channel the city already owns, and each produces an output a city employee acts on. None of them replaced a government function. None required residents to adopt a new app — Rio's runs on WhatsApp, which Brazilians already use.
If you are writing a Round 1 proposal, the implicit benchmark is this: can you name the existing city channel your solution rides on, and the specific municipal decision it changes?
The six review criteria, in order
The published criteria are a sequence, and the order is informative:
- Start with the problem. A significant problem the city government has prioritized, affecting vulnerable communities or promoting social welfare.
- Then plan a solution. Frontier AI capabilities core to the approach.
- Focus on readiness. Your team and your city partner can integrate technology into public-sector settings.
- Build for sustainability. Positioned to survive in terms of follow-on funding, city capacity, and integration with government systems.
- Act responsibly. Demonstrated ability to build and manage AI solutions safely.
- Plan to learn. A clear learning agenda — evaluation, user feedback, and a path to social impact.
Criterion 1 before criterion 2 is not accidental. Proposals that lead with the model and work backward to a problem are the dominant failure mode in AI philanthropy, and this ordering is a deliberate trap for them. The phrase "a problem your city government has prioritized" is the operative language: it means the problem statement should be traceable to a published municipal plan, a budget line, or a mayor's stated agenda — something external to your proposal that a reviewer can verify.
Criterion 4 deserves particular attention because it sits in tension with the grant mechanics. Grants run 24 months and do not guarantee follow-on funding. You are being asked to demonstrate sustainability for a capability whose funder has explicitly declined to commit to continuing it. The resolution is to make the city the sustainability plan — integration into a municipal system with a budget owner, not a promise to raise more philanthropy in year three.
What grantees get besides money
Beyond the cash, grantees receive peer-learning cohorts, occasional AI and engineering support, network access, and invitations to convenings. The engineering support is the line item worth weighing. For a small nonprofit, access to implementation help can be worth more than the marginal $200,000 — and it suggests the funders expect some technical shortfall in the grantee pool, which in turn suggests that a proposal honest about its engineering gaps may fare better than one claiming none.
The short version
If you have a city government relationship: this is a six-week sprint to a letter of intent, a video, and a problem statement anchored in something your city has already published. Start with the letter.
If you do not: Round 1 closing November 29 is almost certainly out of reach, and the better investment is a documented municipal partnership before the next cycle. The stage structure running to April 2027 suggests there will be one.
If you are a for-profit civic tech company: read the eligibility language closely. Limited circumstances, charitable purpose, incidental business benefit. A fiscally sponsored structure or a nonprofit lead with you as a subcontractor is the cleaner path.
If you are watching AI philanthropy generally: the significant design choice here is provider neutrality from a foundation funded by a frontier lab. That is the precedent to watch, and the one other AI-adjacent funders will be measured against.