The OpenAI Foundation Just Named 163 Grantees for $50M. Last Wave It Was 208 for $40.5M — and That Arithmetic Tells You Exactly Who the Fund Moved Toward.

On October 1, 2026, the OpenAI Foundation published the results of its 2026 People-First AI Fund: $50 million to 163 nonprofit organizations, selected from 1,669 applications, spread across 35 U.S. states plus Washington, D.C. The grants are unrestricted, one-time, and the Foundation says the full $50 million will be disbursed before the end of calendar 2026.

We covered the open call when it launched, including the budget-band eligibility and the July 15 deadline and the unrestricted-grant mechanism that made it unusual. Now there is outcome data — and outcome data is far more useful than a solicitation, because it tells you what the fund actually did rather than what it said it would do.

Three numbers do most of the work: 163, 1,669, and the comparison to the prior wave.

The Arithmetic: The Fund Climbed the Budget Band

Wave 1 of the People-First AI Fund awarded $40.5 million to 208 nonprofits — an average of roughly $195,000 per grant. The 2026 round awarded $50 million to 163 nonprofits — an average of roughly $307,000 per grant.

More money. Fewer grantees. Average grant up about 57 percent.

That shift is not a rounding artifact, and it is not a change in generosity. It is a direct consequence of the fund's sizing rule. Grants are set at up to 10 percent of an organization's annual operating budget. If the average grant rose from about $195,000 to about $307,000, and the sizing formula held, then the average grantee's operating budget rose from roughly $2.0 million to roughly $3.1 million.

The eligibility band was $500,000 to $10 million in annual operating budget, with stated priority for organizations in the $1 million to $8 million range. Wave 1's average implies a grantee pool clustered near the bottom of that band. The 2026 average implies a pool that moved decisively into the middle of it.

For anyone planning for a future round, that is the single most actionable finding in the announcement. The stated priority band was not decorative — it was applied, and it tilted upward. An organization with a $700,000 budget is eligible but is now demonstrably competing against a selected cohort averaging four times its size. An organization at $3 million to $5 million sits where the money actually landed.

A 9.8 Percent Selection Rate, and Why That Is Generous

1,669 applications for 163 awards is a 9.8 percent selection rate.

That number will read as brutal to anyone who has not spent much time in competitive philanthropy, and generous to anyone who has. For context, it is better odds than most NEH scholarly programs, better than NSF's flagship research competitions in several directorates, and vastly better than the single-digit hit rates typical of large open-call corporate funds with no eligibility floor. The budget-band restriction and the standalone-entity requirement did real work here: by fencing out university departments, fiscally sponsored projects, organizations under $500,000, and organizations over $10 million, the Foundation produced a field of 1,669 rather than the 10,000-plus an unfenced call would have drawn.

The Foundation also disclosed something most funders do not: every application received was reviewed by an experienced external grantmaker. Not triaged by staff, not screened algorithmically — read by an outside professional.

That disclosure matters strategically. In a fund where every application gets a human read, narrative quality converts into outcomes. In a fund where the first pass is a keyword screen, it does not. If you applied and lost, your application was read by someone qualified to judge it, which means the loss is information rather than noise. If you plan to apply next cycle, the writing is worth the investment — a rare thing to be able to say with evidence.

The stated selection criteria were community trust, demonstrated impact, and a thoughtful approach to exploring AI — in that order, and note which one comes last.

What the Named Grantees Reveal About What Won

The Foundation named examples across its three target categories — community support services, community arts and cultural organizations, and community journalism and media. Five are instructive:

• Greater Bakersfield Legal Assistance (California) — civil legal services
• Nashville Jazz Workshop (Tennessee) — exploring AI to make its archive searchable
• Honolulu Civil Beat (Hawaii) — making public records searchable
• California Black Media (California) — supporting a network of 30 Black-owned outlets
• Library Foundation of Los Angeles (California) — expanding equitable learning access

Look at what two of these have in common. Nashville Jazz Workshop wants a searchable archive. Honolulu Civil Beat wants searchable public records. These are the same project in different domains: an organization that already owns a valuable, unstructured corpus, applying AI to make that corpus findable.

That is a pattern worth naming, because it is the most fundable AI use case in the nonprofit sector right now and it is dramatically underclaimed. Nearly every established community institution is sitting on a corpus — case files, oral histories, clip archives, meeting minutes, program records, donor correspondence, decades of local coverage. The institutional value of that corpus is currently near zero because nobody can search it. Retrieval is the one AI capability that is mature, cheap, verifiable, and immediately legible to a program officer. You do not have to promise a model will be right; you only have to promise it will help someone find the document.

California Black Media illustrates the second winning shape: the network intermediary. One grantee, 30 downstream outlets. The Foundation explicitly allowed regranting organizations with operating budgets under $15 million into the 2026 round, and the intermediary structure multiplies reach into organizations far below the $500,000 eligibility floor. If you are a state press association, a community foundation, or a sector coalition, that lane exists specifically for you and remains the highest-leverage position in this fund.

What is conspicuously absent from the named examples: anyone building a chatbot, anyone replacing staff, anyone proposing to train a model. The fund is product-agnostic — grantees choose their own tools and receive no OpenAI credits or accounts — and the selected work reflects augmentation of existing mission delivery, not technology adoption as an end.

Fifteen States Got Nothing

Thirty-five states plus D.C. means 15 states received zero awards.

Three of the five named example grantees are in California. That is a sample, not a distribution, but the geographic concentration is real and it is worth interpreting carefully rather than indignantly.

A 15-state shutout in an open national call with a 9.8 percent selection rate is not necessarily evidence of funder bias. It is more likely evidence of an application-volume gap. Open-call philanthropy flows along information channels — sector listservs, regional association newsletters, peer referrals, the handful of nonprofits in a state that track technology philanthropy closely. States with dense nonprofit infrastructure generate proportionally more applications and therefore more awards from a merit-blind process.

For organizations in the missing states, that reframes the problem usefully. The obstacle is probably not that your state is disfavored. It is that your state submitted few applications, which means the per-application odds there were likely better than average, not worse. The fix is distribution: get the next open call in front of your state association, your community foundation, and your regional funders' network the week it drops, not the week before it closes.

The Sentence to Read Twice

Buried in the fund's terms is a line that should govern how any grantee treats this money:

"Receipt of funding in one year should not create an expectation of future funding."

These are explicitly one-time grants. The fund has now run two waves with different sizes, different grantee counts, and a shifting budget band — it is not an annuity, and the Foundation has said so in writing.

Unrestricted money is the most valuable money in philanthropy precisely because it can pay salary, rent, and the costs no restricted grant will touch. It is also, for exactly that reason, the easiest money to accidentally build a recurring cost structure on. A $307,000 unrestricted grant that funds a new position becomes a $307,000 hole in 2027 unless somebody is already working the replacement. The disciplined use of a one-time unrestricted award is to buy something durable — capacity, infrastructure, a searchable archive, a reserve — rather than to buy a year of operating expense you will have to re-fund.

The Positioning Checklist for the Next Round

The Foundation has not announced a 2027 wave, and its own language discourages assuming one. But the fund has now run twice, the template is stable, and the cost of being ready is low. If a third wave opens, the organizations best positioned will have already handled:

1. Standalone 501(c)(3) status. Not a fiscally sponsored project, not a university department, not a program inside a larger institution. This is the single hardest disqualifier and the slowest to fix.
2. Budget-band awareness. The eligible range is $500,000 to $10 million; the selected cohort averaged roughly $3.1 million. Know which end of the band you sit on and calibrate expectations accordingly.
3. A named corpus. Identify the archive, case-file set, records collection, or clip library you already own that nobody can currently search. That is your strongest available proposal, and it is true before you write it.
4. Community-trust evidence first. The stated criteria put community trust and demonstrated impact ahead of AI thoughtfulness. No prior AI experience is required — and the fund says so explicitly — so lead with the thing you actually have.
5. The regranting lane, if you qualify. Under $15 million in operating budget excluding pass-through, and you can reach dozens of organizations too small to apply directly. That is the highest-multiplier application in the fund.

The broader signal in these results is that 2026's AI philanthropy wave is maturing from announcements into audited outcomes. The Citi Foundation's $25 million youth AI challenge closes on October 6, Humanity AI's $10 million open call closes October 21, and both now have a reference point for what a selected cohort in this space actually looks like. The organizations that read the outcome data, rather than only the solicitations, will write better applications than the ones that do not.