OpenAI splits into two: non-profit holding company + public welfare enterprise making money (same as Anthropic)

OpenAI issued a "maintaining non-profit" announcement today, which is quite important:

Prior to this, OpenAI planned to separate its for-profit subsidiary from the control of the nonprofit organization and transform it into an independent for-profit company. After facing resistance and legal challenges from many parties, OpenAI abandoned the idea.

Key points:

• •  Maintaining non-profit , OpenAI still flies the banner of “non-profit controlled”.
• •  The LLC under the subsidiary will be changed to a PBC , and the original capped-profit restrictions (return cap) will be eliminated and replaced with a regular equity structure. ( Note: Anthropic is a PBC . In addition, I feel that in name, it is for a clearer governance structure, but in fact it is also to prepare for future financing, stock issuance, and even IPO. And why do I always feel that this subsidiary is the main body...)
• •  Resources will flow back to non-profits . Under the new structure, non-profits will receive a portion of the share profits and have more money to invest in public welfare-oriented AI applications - medical care, education, public services, scientific research, etc.
• •  The story of democratized AI is still being told . OpenAI continues to emphasize "putting powerful tools in the hands of everyone." It is neither a closed approach nor an "authoritarian AI." The model will continue to be open, and the freedom of use will also be expanded.

what happened

OpenAI published a long "Letter to Employees" on its homepage:

Here I translated it into Chinese:

Sam's Staff Letter

OpenAI was never a traditional company from the beginning, and it never will be.

Our mission is clear: to ensure that artificial general intelligence (AGI) can truly benefit all of humanity .

When we founded OpenAI, we didn’t know how to achieve this ambitious goal. We sat around the kitchen table, not knowing what kind of research to start with, and never thought about developing products or designing business models. We couldn’t have imagined that AI would be widely used in medicine, education, and work efficiency in the future, and that it would require hundreds of billions of dollars in computing resources.

Initially, many believed that powerful AI should only be in the hands of a few “trusted people.”

But now, we see new possibilities - AGI can directly empower everyone and become the most powerful tool ever created by mankind. If we succeed, it will promote social development and improve the overall quality of human life. Although it will not be used entirely for good purposes, we believe that the overall goodwill of mankind will eventually outweigh the negative risks.

We are committed to taking a path of " democratized AI ". We want to put powerful AI tools in the hands of everyone. We are amazed by the creations made by users, and are willing to give them greater freedom to use them - even if there are differences in values ​​between us, we still respect their right to decide how ChatGPT behaves.

Our goal is to build a "world brain" that everyone can use to do what they want (without interfering with the freedom of others).

Now, thousands of people are using ChatGPT to conduct scientific research, write code, treat diseases, learn new knowledge, and even solve life problems. We are very proud of this, which is the most authentic response to our mission.

However, this is not enough. The global demand for AI far exceeds our current supply. We have had to set usage limits and slow down service. In the future, as models become more powerful, this demand will only increase and become more widespread.

Looking back nearly a decade ago, when we founded OpenAI, we could not have imagined that we would get to this point. And now, we are excited to see this picture.

To address this reality, we decided to adjust our organizational structure with three goals:

1. 1. In order to provide services to the world at large, we need hundreds of billions or even trillions of dollars in resources.
2. 2. We want to build the most effective AI nonprofit organization in history and bring the greatest positive impact to humanity.
3. 3. We need to achieve “beneficial AGI” — both promoting technological progress and ensuring its safety and controllability.

We will transform the LLC into a Public Benefit Corporation (PBC) under the leadership of a nonprofit organization, retaining our original mission. This will give us a simpler and clearer equity structure, allowing for more efficient development while maintaining a public welfare orientation.

Non-profit organizations will obtain sufficient shares and have a leading role, which will provide them with considerable resources to promote AI public welfare projects such as medical care, education, science, and public services. As PBC grows, non-profit organizations will also obtain more resources to invest in the social benefits of AI.

We look forward to hearing the Nonprofit Commission’s next recommendations. They will help us ensure that the benefits of AI are not just for a few, but truly benefit everyone.

This structural adjustment will help us continue to accelerate progress while ensuring safety and put excellent AI in the hands of more people. Building AGI is one brick we have laid for human society, and we look forward to seeing the next brick you lay.

—Sam Altman , May 2025

How to interpret

OpenAI announced that it would continue to be non-profit, which sounds like a "compromise" with Musk. But if you look closely, you will find that this is actually a re-centralization of power by Sam Altman.

Originally, OpenAI wanted to separate its for-profit business from its non-profit structure and turn it into an independent company. The goal was to simplify the governance structure, release equity, and facilitate listing and financing. Musk sued, and state regulators began to intervene, so the structure could not be dismantled.

So Altman went with the flow: he would not dismantle the company and would keep the shell of "non-profit holding". But in reality, the non-profit organization was already in his hands.

Don't forget that after the dismissal storm in 2023, Altman returned and the board of directors was reorganized. The current board of directors has 10 people, most of whom joined after Altman returned. In the future, non-profit and for-profit will have separate boards of directors, and the new non-profit board members are said to be selected by the existing board of directors. In other words, whoever controls the non-profit board controls OpenAI, and Altman controls the right to select people .

The structure remains the same, but the control is changed.

For Altman, this is safer than turning the company into a normal for-profit enterprise: • The non-profit shell can block regulation and resist litigation risks; • Investors can still enter the market and get shares, no longer in a “return cap” model, but directly hold shares; • The non-profit is still the controlling shareholder and can continue to lead the direction, control the technology path and public communication rights.

If Musk's goal was to limit OpenAI, the effect was the opposite

The PBC (Public Benefit Corporation) structure is already the default configuration for AGI startups such as Anthropic and xAI. It allows companies to pursue profits legally while also being included in their mission statements, continuing to promote the banner of “benefiting mankind.”

Now OpenAI is also a PBC, it can issue shares, tell stories, and receive continuous investment; on another dimension, it is also preparing for future financing, issuing shares, and even IPO.

Next step? OpenAI will issue regular shares to replace the "profit certificates" (original capped-profit agreement) held by early investors to avoid the "control dilution" problem.

Investors have no board seats and no voting rights. This is similar to the two-tier voting rights structure of Google and Meta: you can make money, but you don’t care about the business.

For Altman, this set of actions is " nominal concession + substantive enhancement ". For investors, it is "losing control but being able to get on board". For regulators, it is "you win the structure but cannot control the people".

Qin Shi Huang ate peppercorns and won. Altman did not take back OpenAI directly. He designed a system so that he would never need to take it back again.