OpenAI Has Been a Public Benefit Corporation Since 2025: What It Teaches Brazilian Founders

One year later, what actually changed

On October 28, 2025, OpenAI formalized a restructuring that had been under negotiation for nearly a year with the offices of the Attorneys General of California and Delaware, according to the company's own official page about its structure. The nonprofit founded in 2015 was renamed OpenAI Foundation. The for-profit subsidiary created in 2019 became OpenAI Group PBC, a public benefit corporation.

The difference isn't just in the name. A PBC is, by legal definition, required to pursue a stated purpose and to consider the interests of multiple stakeholders, not just shareholder returns. In practice, OpenAI replaced an arrangement in which the for-profit was simply controlled by the nonprofit with a model in which the for-profit itself carries, in its bylaws, the legal obligation to advance the original mission: ensuring that artificial general intelligence benefits all of humanity.

Control separated from economics: the central design

The point most relevant to anyone thinking about corporate structure is how OpenAI separated two things that usually go together: who is in control and who makes money. OpenAI Foundation retained control of OpenAI Group through special voting and governance rights. It appoints all members of the Group's board and can replace them at any time (a structural veto power that no investor, no matter how large the check, holds).

At the same time, the Foundation moved from a purely control-based relationship to direct economic participation: it now holds 26% of OpenAI Group's capital, a stake valued at roughly $130 billion based on the company's current valuation, according to the official document. In addition, the Foundation holds a warrant that entitles it to additional shares if the Group's share price rises more than tenfold within a fifteen-year horizon.

This design makes the Foundation, according to OpenAI itself, the largest long-term beneficiary of the company's success, not merely its moral guardian. All other Group shareholders now hold the same class of common stock, which participates proportionally in any appreciation. There is no longer a differentiated return cap for investors: capital that goes into the Group grows (or shrinks) along with the company's value, the same for everyone.

Overlapping but not identical governance

The Foundation's board today is composed of Bret Taylor (chair), Adam D'Angelo, Paul Christiano, Sue Desmond-Hellmann, Zico Kolter, retired general Paul M. Nakasone, Adebayo Ogunlesi, Nicole Seligman, David Vélez, and Robin Vince, in addition to CEO Sam Altman. To avoid what the company calls "fragmented governance," practically all of the Foundation's directors also sit on the Group's board.

The exception is Zico Kolter, who chairs the Safety and Security Committee and now serves exclusively on the Foundation's board, acting as a non-voting observer on the Group's board. Within a year, a second director is expected to follow the same path. The safety committee remains subordinate to the Foundation, not the Group: in other words, risk oversight is deliberately kept out of the direct reach of shareholders who care only about returns.

The counterpoint: this is not a shortcut

The temptation to read this case as a manual for "how to lock in the mission and still raise a round" ignores the cost of the process. The restructuring took nearly a year of formal dialogue with two U.S. state attorneys general offices, required independent financial advice for the Foundation's board, and only closed after unanimous board approval. That's the weight of a company valued at hundreds of billions of dollars, not a seed-stage startup.

For most Brazilian founders negotiating a SAFE worth a few million reais, replicating a two-entity architecture with interlocking boards and fifteen-year warrants is over-engineering: the legal cost and negotiation time far outweigh the benefit of "protecting the mission" at a stage when the company is still validating its product. OpenAI's model is only justified because the company already had enough scale, capital, and regulatory exposure to sustain the arrangement.

What's useful for founders raising capital in Brazil

What's worth importing isn't the entire structure, but the principle: control and economics can be negotiated separately. A founder who fears losing the company's mission by accepting capital from funds with an aggressive return mandate doesn't have to choose between turning down the check or giving up governance: they can negotiate lighter instruments that already exist in Brazil, such as founder-retained plural voting shares, veto clauses on specific points of the corporate bylaws, or B Corp certification through Sistema B (the Latin American network that certifies B Corps), which imposes purpose commitments without requiring a second controlling legal entity.

It's also worth noting how OpenAI handled the upside question: by putting everyone on the same class of stock, it avoided the resentment typical of structures where some shareholders have a return cap and others don't. In SAFE negotiations with a lean cap table, this translates into a direct lesson: mission clauses that restrict an investor's financial return tend to raise the cost of capital, because the fund prices that risk into the valuation. Separating the governance lock from the economic lock, as OpenAI did, is more expensive to set up but cheaper to negotiate with whoever is putting up the money.

What remains open is whether this hybrid model withstands the pressure of an IPO or a round that requires aggressive dilution of the Foundation. OpenAI is betting that an interlocking board and proportional economic participation are enough to maintain de facto control even as external capital flows in at volume. For Brazilian founders, the real lesson isn't to copy the PBC, it's to understand that capital architecture is a strategic decision as central as the product itself, and that it needs to be designed before the pressure of the next round, not after.