By Assaf Biton
Everyone keeps calling OpenAI a for-profit company now. That is only half the story. On paper, the part of OpenAI that owns ChatGPT answers to a charity called the OpenAI Foundation, a nonprofit that holds 26% of a business valued at $852 billion as of its March 31, 2026 funding round.
So the question I kept coming back to is simple. If a nonprofit "controls" one of the most valuable companies on Earth, who actually controls the nonprofit?
I read through the October 2025 restructuring, the March 2026 leadership announcement, the California Attorney General agreement, and the Elon Musk trial that just ended in May. The structure is real. The control is also real. But the independence that is supposed to make this work is the part nobody can point to. Let me show you why that gap matters, and why it gets sharper as OpenAI moves toward a possible $1 trillion IPO.
TL;DR: The OpenAI Foundation is a 501(c)(3) nonprofit that controls OpenAI Group PBC, the public benefit corporation running ChatGPT and the API (its Sora video app was discontinued in April 2026, with the API set to close in September 2026). The Foundation holds about 26% equity (Microsoft holds 27%) but keeps governance control through special voting shares: it appoints and removes the company's board. The catch is that 8 of its 9 board members also sit on the company board, the same chair runs both, and there is still no permanent executive director. It pledged $1 billion in grants for 2026, roughly 0.77% of its $130B paper stake, which critics call a pre-IPO public relations move rather than real philanthropy.
What the OpenAI Foundation Actually Is
The OpenAI Foundation is the renamed version of the original nonprofit that started OpenAI in 2015. Back then it was called OpenAI Inc. Today it is a 501(c)(3) tax-exempt charity, and it sits at the top of the whole structure.
Below it is OpenAI Group PBC, a Delaware public benefit corporation. That is the entity that ships products and takes investment. Microsoft, employees, and outside investors own shares in the PBC. The Foundation owns 26% of it.
Here is the part that trips people up. A public benefit corporation must "consider" the interests of society alongside shareholder returns. But it is still a for-profit. And unlike the old structure, there is no cap on how much money investors can make. The 100x profit cap from the 2019 setup is gone.
Pro Tip: When you read that the nonprofit "controls" OpenAI, mentally swap "controls" for "appoints the board." The Foundation does not run ChatGPT day to day. Its power is the authority to hire and fire the directors who oversee the people who do. That distinction explains roughly every criticism in this article.
The three entities, in plain English
- OpenAI Foundation: the 501(c)(3) nonprofit at the top. Holds 26% equity plus special controlling shares. Runs the grant programs.
- OpenAI Group PBC: the for-profit underneath. Operates ChatGPT and the API (it launched the Sora video app in September 2025 but shut it down in April 2026). Takes the investment, including Microsoft's 27% stake.
- The control link: the Foundation's special shares let it appoint and remove every PBC board member and veto major moves like mergers, mission changes, and new share issuances.
The flagship product operated by OpenAI Group PBC
Best for: General knowledge workers needing a capable all-in-one assistant, Developers wanting quick code generation with GPT Image and Codex integration
How a Nonprofit Controls a For-Profit Company
The control mechanism is not about owning the most shares. It is about owning the right shares.
The Foundation holds special voting and governance rights. As long as it keeps those shares, its board can appoint and remove all PBC directors. It also must sign off on the big stuff: governance changes, mission amendments, mergers and acquisitions, and any new shares the company wants to issue. There is also a Safety and Security Committee, chaired by Carnegie Mellon professor Dr. Zico Kolter, that can in theory veto model releases.
That sounds strong. And structurally, it is stronger than what most corporate foundations have. Google.org cannot fire Google's board. The Chan Zuckerberg Initiative is not even a nonprofit, it is an LLC. So on the org chart, the OpenAI Foundation looks like a genuine watchdog.
But a watchdog only works if it is a different animal from the thing it is watching. That is where this falls apart.
Eight of the Foundation's nine board members also sit on the for-profit OpenAI Group PBC board. Chair Bret Taylor runs both. Sam Altman is CEO of OpenAI and sits on the Foundation board. When the same people approve a decision on both sides of the table, "oversight" starts to look like a mirror.
The Governance Problem Critics Keep Circling
Let me be direct about the downsides, because this is where the research is loudest.
The board overlap is the headline issue. The Foundation is supposed to hold the company accountable to its charitable mission. Yet the people doing the holding are mostly the same people running the company. Robert Weissman, co-president of the watchdog group Public Citizen, put it bluntly: "Can anyone imagine OpenAI Foundation pursuing a strategy that contradicts OpenAI's business interests? The obvious answer is no."
Then there is the staffing. Wojciech Zaremba, an OpenAI co-founder who still works at the company, leads the Foundation's AI Resilience division. So a current OpenAI employee helps run the division meant to study OpenAI's risks. And as of mid-2026, the Foundation still has no permanent executive director. It has hired around it: Jacob Trefethen for life sciences grants, and Anna Makanju, formerly OpenAI's VP of Global Impact, who joined in April as Head of AI for Civil Society and Philanthropy. But the top independent seat is empty.
Orson Aguilar of the Eyes on OpenAI coalition asked the question that stuck with me: "Is this a true foundation or is it just a drawer on Sam Altman's desk?"
OpenAI Foundation: What Works and What Doesn't
What Works
- Real structural power: the Foundation can appoint and remove the entire PBC board, which is more authority than Google.org or most corporate giving arms hold.
- Veto rights over mergers, mission changes, and new share issuances are written into the structure, not just promised.
- The California and Delaware Attorneys General extracted 20 written concessions before approving the restructuring in October 2025.
- A credible grant hire: Jacob Trefethen oversaw $500M+ in science and health grants before joining.
- The 2026 grant budget is a 20x jump from the roughly $50M in 2025 activity.
What Doesn't
- 8 of 9 Foundation board members also sit on the for-profit board, so the watchdog and the watched share the same faces.
- No permanent executive director, so the Foundation launched its $1B plan without an independent leader running it.
- An OpenAI co-founder still employed at the company (Wojciech Zaremba) heads the AI Resilience division.
- The AG deal lacks specifics on enforcement, safety committee membership, and periodic review, per Stanford law professor Robert Bartlett.
- The first $40.5M in grants went to AI literacy and community groups, with no funding for AI safety researchers or accountability watchdogs.
Is $1 Billion a Lot, or Almost Nothing?
In March 2026 the Foundation pledged at least $1 billion in grants for the year, spread across four areas: life sciences and curing diseases, jobs and economic impact, AI resilience, and community programs. It is the largest single-year AI philanthropy commitment anyone has announced. That is genuinely notable.
But context changes the picture fast. The Foundation's 26% stake was worth roughly $130 billion at the October 2025 valuation. So $1 billion is about 0.77% of its assets.
Why does that fraction matter? Standard private foundations in the United States must pay out around 5% of their assets every year. At 5%, a $130B endowment would give away roughly $6.5 billion to $7.5 billion annually. The OpenAI Foundation is giving about one-seventh of that.
The Foundation may not legally be a "private foundation," which means it might be exempt from the 5% minimum payout rule entirely. That is a technical relief for OpenAI and a real worry for critics: a $130B charity with no mandatory floor on how much it must give away.
There is also a separate $25 billion long-term commitment announced in October 2025. The catch? No timeline. It could play out over decades. The 2026 grants are described as "early investments" toward that number, which is another way of saying the big promise has no deadline.
How the Foundation compares to other tech foundations
I find this table the fastest way to see the gap. Pay attention to the last column.
| Foundation | Assets | Annual Giving | Independent Board? |
|---|---|---|---|
| OpenAI Foundation | ~$130B (equity) | $1B (2026 target) | No (8 of 9 also on company board) |
| Gates Foundation | ~$86B | ~$9B/year | Yes |
| Novo Nordisk Foundation | ~$140B+ | ~$1.6B/year | Partially |
| Chan Zuckerberg Initiative | ~$100B+ (pledged) | ~$1B/year | N/A (it's an LLC, not a nonprofit) |
| Google.org | N/A | ~$300M/year | No |
The Gates Foundation holds fewer assets and gives away nine times more. It also has no commercial product to protect and a fully independent board. The Novo Nordisk Foundation is the closest structural cousin, a foundation controlling a company, but it has 80+ years of independent operation behind it. The OpenAI Foundation has none of that track record yet.
To be fair, the two are not pure rivals. The Gates Foundation and OpenAI are collaborating on Horizon1000, a $50M effort to bring AI to 1,000 primary healthcare clinics in Africa by 2028. So this is not all theater.
Why the Timing Looks Suspicious
Here is the pattern that made me raise an eyebrow. Every major philanthropy announcement has landed on top of a business milestone.
The $25B commitment was announced the same week the restructuring closed in October 2025. And the $1 billion pledge plus the leadership reveal? That dropped on March 24, 2026, the exact same day OpenAI secured an extra $10 billion in funding. Same day. Not the same week.
The IPO context: OpenAI confidentially filed for a U.S. IPO on June 8, 2026, with a listing window from September into the fourth quarter and a target valuation near $1 trillion. Its last confirmed primary valuation was $852 billion after the $122 billion round closed on March 31, 2026. The company reported roughly $2 billion in monthly revenue in March 2026, while still projecting a $14 billion loss for the year. A $130B foundation is a useful story to tell investors, regulators, and the public right before the biggest tech IPO in history.
Nick Moes, executive director of The Future Society, called the restructuring "a dark day for philanthropy," warning the funding would mostly serve OpenAI's commercial interests by promoting a positive view of AI rather than funding real scrutiny. Whether you buy that or not, the calendar is hard to ignore.
The Musk Trial and the Soul of OpenAI
For two years, the loudest challenge to this whole structure came from one person: Elon Musk. He helped fund OpenAI early (roughly $38 million) and sued in 2024, arguing Altman and Greg Brockman betrayed the founding promise to keep OpenAI a nonprofit. He sought up to $134 billion in damages.
The trial opened on April 27, 2026, in Oakland before Judge Yvonne Gonzalez Rogers. It ran 11 days. The most damaging piece of evidence was a 2017 handwritten journal entry from Brockman: "I cannot believe that we committed to non-profit if three months later we're doing b-corp then it was a lie."
And then, on May 18, 2026, it ended fast. The nine-member advisory jury deliberated for less than two hours and dismissed all of Musk's claims, finding he filed outside the three-year statute of limitations. Judge Gonzalez Rogers adopted the verdict. The court never ruled on whether the underlying betrayal claim was true. It ruled that Musk waited too long.
Musk called it a "calendar technicality" and said he would appeal. So the legal fight is not fully over. But the immediate threat to the restructuring is gone, which removes one of the few external forces that could have forced real change on the Foundation's structure.
"The court did not decide that the nonprofit kept its promise. It decided that Musk asked the question too late."
What This Means If You Use ChatGPT
You might be thinking: I just use ChatGPT to write emails, why should any of this matter to me? Fair question.
It matters because governance shapes incentives, and incentives shape the product. Critics point to the September 2025 launch of Sora as a sign the nonprofit is not slowing anything down. Public Citizen argues there has been "no evidence whatsoever of the nonprofit exerting control over the for-profit" since the November 2023 board crisis, when Altman was fired and reinstated within five days.
So when OpenAI tells you a charity is keeping it honest, this article is your reason to ask one more question: honest according to whom, and who checks? If you compare AI assistants on safety culture, governance belongs on the list next to speed and price. You can see how the major models stack up in our ChatGPT vs Claude comparison and our roundup of the best AI chatbots in 2026.
Pro Tip: Want to track whether the Foundation is real or for show? Watch one number across 2027: total grants disbursed versus the 5% benchmark (~$6.5B). If it stays near $1B while the company's valuation climbs past $1T, the gap between the promise and the payout is your answer. Bookmark our AI news category for the updates.
The Verdict: Real Structure, Missing Independence
Our Take on the OpenAI Foundation
What's genuinely real: The control mechanism. The Foundation can appoint and fire the company's board and veto major moves. That is more power than any other tech foundation holds over its parent company.
What's still missing: Independence. With 8 of 9 board seats shared with the for-profit, no permanent executive director, and a $1B payout that is 0.77% of assets, the Foundation looks more like a department of OpenAI than a check on it.
The bottom line: This is the most powerful nonprofit-over-company structure in tech, attached to the least independent oversight. Both things are true at once, and that tension is the whole story.
So what should you actually do with this? Don't take "a nonprofit controls OpenAI" at face value, and don't dismiss it as pure theater either. Track three concrete signals over the next year: whether the Foundation hires an independent executive director, whether it appoints board members who do not also work for OpenAI, and whether 2027 grants climb toward the 5% benchmark. If two of those three happen, the watchdog is growing teeth. If none do, you have your answer about the drawer on Sam Altman's desk.
Frequently Asked Questions
What is the OpenAI Foundation?
The OpenAI Foundation is a 501(c)(3) nonprofit, formerly named OpenAI Inc., that controls OpenAI Group PBC, the for-profit public benefit corporation that operates ChatGPT and the API. (The PBC also ran the Sora video app, which it discontinued in April 2026.) It holds about 26% equity in the company plus special voting shares that let it appoint and remove the company's board. It also runs OpenAI's grant programs, with a $1 billion budget for 2026.
Does the nonprofit really control OpenAI?
Legally, yes. The Foundation's special shares give it the power to appoint and remove every member of OpenAI Group PBC's board and to veto major actions like mergers, mission changes, and new share issuances. In practice, critics question that control because 8 of the Foundation's 9 board members also sit on the for-profit company's board, so the same people oversee both sides.
How much does the OpenAI Foundation give to charity?
The Foundation pledged at least $1 billion in grants for 2026 and a longer-term $25 billion commitment with no set timeline. Its first grants in December 2025 totaled $40.5 million to 208 nonprofits. For scale, $1 billion is roughly 0.77% of its estimated $130 billion stake, well below the ~5% annual payout standard that applies to private foundations.
What happened with Elon Musk's lawsuit against OpenAI?
On May 18, 2026, an advisory jury in Oakland dismissed all of Elon Musk's claims after deliberating less than two hours, finding he filed his 2024 lawsuit outside the three-year statute of limitations. Judge Yvonne Gonzalez Rogers adopted the verdict. The court did not rule on whether OpenAI broke its founding nonprofit promise. Musk called the ruling a "calendar technicality" and said he would appeal.
Is the OpenAI Foundation independent from OpenAI the company?
Not yet, by most measures. As of mid-2026 it has no permanent executive director, 8 of its 9 board members also serve on the for-profit board, the same chair (Bret Taylor) runs both entities, and an OpenAI co-founder still employed at the company leads its AI Resilience division. These overlaps are the main reason a coalition of 60+ nonprofits continues to push the Attorneys General for stronger enforcement.
