The OpenAI Coup: How a Nonprofit Board Could Fire Its CEO
A 700-person company less than five years old, a six-member nonprofit board that exercised its power once and pulled off a coup — and its largest shareholder, Microsoft, isn't even on the board.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
A 700-person company set off a global uproar
Luo Yihang judges this unprecedented: Twitter was chaotic internally in its early years too, but that was a matter for a small Silicon Valley circle — it didn't get the whole world discussing and snacking on gossip. OpenAI as a company has existed for only a bit over 4 years, nearly 5, with a headcount of 700-plus, yet it caused this kind of storm, which is very rare. This contrast is itself the episode's question: why can a single board vote at such a small company become a global event.
— Luo YihangThe means were political; the motive wasn't necessarily a conspiracy
Luo Yihang repeatedly stresses: the coup happened very politically, but political things don't necessarily have bad motives — political is a characteristic and a means. The real cause is still a fairly large divergence in philosophy — when GPTs and the GPTs Store launched on November 6, he personally was excited and shocked, feeling they had taken a very radical step. He also reminds us that any decision, especially one by an outstanding figure, is often mixed with very strong personal values and private interests — the two can't be separated.
— Luo YihangOn safety, they say one thing and do another
Luo Yihang heard from people in the circle in Silicon Valley this May that Sam Altman actually says one thing and does another on alignment, safety, and governance — he is in fact very radical. This source is indirect, coming from an intermediary who is an AI entrepreneur, not directly from someone inside OpenAI. By contrast, Ilya has spoken in multiple interviews about three things: caring whether AI can possess human love, worrying about superintelligent AI in relation to human ethical order and safety, and that the most important thing he pushes internally is alignment.
— Luo YihangThe most prestigious independent directors all left
The board list was originally long, 9 people, and they dropped out one after another: Elon Musk exited in '18 or '19 over philosophical differences — at the time he opposed OpenAI becoming a nonprofit, and Luo Yihang believes that idea was genuine then; Elon Musk's ex-wife also exited a few months ago; most critically, LinkedIn founder Reid Hoffman, whose reason for leaving was that he was preparing to do similar startups and investments. Luo Yihang says, if Reid Hoffman were still on the board, would this matter have been less hard to wind down in the end — because this person is especially constructive, and is good friends with many people whose relationships are extremely opposed. Of the three remaining, he feels ‘the prestigious people have all left.’
— Luo YihangThey exercised power once, and it actually worked
Luo Yihang says two things shocked him. First, the company's key decisions were actually all made by these three people; the directors were neither informed, nor had they fully exercised power in the past — and then they exercised power once, and this thing actually got done. Second, 700 people signed in support — does this company have internal strife or not? It's practically a solid block. He believes Ilya later joining the signatories was a very important trigger for getting those 700-plus people to sign.
— Luo YihangA nonprofit controls a for-profit
The mechanism was built like this: OpenAI started as a nonprofit, and the people on the board held no shares, only voting rights, representing the public in supervising and managing the institution. In 2019 the situation changed — they were already using the Transformer architecture for training, burning money heavily, and the directors who had promised donations only put in money without taking shares — Elon Musk promised 100 million, Reid Hoffman promised several hundred million, and the money came in bit by bit. So OpenAI Inc was set up as the nonprofit, and OpenAI LP as the limited-partnership for-profit vehicle to raise funds. The result: OpenAI LP has no board of its own — its board is the board of OpenAI Inc.
— Luo YihangMicrosoft holds 49% but isn't on the board
Microsoft changed enormously that year, deciding on a route: AI goes toward artificial general intelligence, in partnership with OpenAI. Luo Yihang thinks this is related to Lu Qi's departure — Lu Qi represented another, different route for AI inside Microsoft, at least not wanting to do AGI, more about industry applications. Microsoft invested all the way to today, already holding 49% of OpenAI LP, just short of a controlling stake, but neither Microsoft nor Satia himself is on the board, which is why they were only told at the end. Luo Yihang says this is something the whole of Silicon Valley has simply never encountered.
— Luo YihangWhen something is abnormal, there must be a demon
Luo Yihang says from day one he didn't think this mechanism was good: the founder holds no shares, his voting power is equal to others', a person close to being the largest shareholder has no voting rights — this is simply bizarre. He contrasts it with Silicon Valley's past very creative founder AB-share mechanism — Meta has so many outside shareholders that by equity they already exceed Zuckerberg, but Zuckerberg has AB voting rights to control the board. OpenAI instead went the opposite direction, moving toward greater dispersion in the name of a nonprofit. At the start of the year when reviewing, everyone thought this design was original and a key step to success; today everyone says this mechanism is very problematic.
— Luo YihangIn the past, YC wouldn't even take founders with accents
In 2010, at a YC event at the Computer History Museum, Luo Yihang personally heard Paul Graham tell an Eastern European founder one of the problems is that your English is not very good; the other said my English is really good, and Paul Graham replied you have a very strong accent. Luo Yihang says that person's accent was absolutely lighter than Satia's and Sundar Pichai's. He personally dislikes this, believing accent is part of diversity. But on Sam Altman you don't see this kind of baggage and prejudice — in 2017 YC broke precedent by holding an event for Chinese LPs during the National Day holiday, and even provided simultaneous interpretation.
— Luo YihangHe looks nerdy, but does the work of a networking king
Luo Yihang says Sam Altman is simply another version of Silicon Valley's networking king: he looks extremely nerdy, but all the work he does is the work people like Reid Hoffman do — convening groups, persuading people, asking for money, recruiting, presenting a vision. He has basically hardly ever done technology. Luo Yihang sees this as one of his traits: deep and broad connections, skilled at expressing and conveying a mission, skilled at fundraising, skilled at bringing different people together, and pushing things to happen in a more radical way.
— Luo YihangHe didn't write a sob story; he talked a lot about love
Luo Yihang thinks Sam Altman's handling over these days deserves praise: first, he didn't choose to write a sob story or publicly face off in a PK, because that might hurt his own image, force the other side to show many cards, and more importantly would hurt the organization; he has so far not revealed a single word externally about what the disagreement was or how the process went, nor publicly protested injustice. Second, he talked a lot about love, saying love the team. Third, he can clown around now, with great energy. Fourth, after Ilya tweeted that he regretted participation in the board, Sam Altman liked three hearts — Luo Yihang thinks this doesn't mean he fully agrees with every word Ilya said, but is a gesture to gather everyone back together.
— Luo YihangChinese people aren't incapable — they just weren't taught
Luo Yihang says Chinese people feel somewhat out of place after arriving in Silicon Valley, because from childhood we were never educated to play these political things. Everyone complains that Indians form cliques, but he thinks this isn't clique-forming — Indians arrived in America earlier, and apart from accent they tolerated everything else well, and they know better how to find people to endorse them, platform them, support their positions, and defend their interests. White people do the same, it's just that no one says they form cliques. He gives Meta as an example: after '14, '15 Chinese people increased, especially after some Chinese reached VP level and above, Chinese at Meta began consciously doing similar things — mentoring and passing on, supporting their own people, actively expressing that not all of us can be engineers, we need to be PMs and enter the business sector.
— Luo YihangSilicon Valley truly favors these two types
Luo Yihang summarizes: two types of people are especially welcomed by Silicon Valley, and behind this is a very interesting tension. One is white men who graduated from Ivy League science and engineering schools, with senior but not particularly high-level experience at big companies — they most easily win the favor of VCs and the environment, represented by Luminar founder Austin Russell — Ivy League graduate, rich second generation, white, with a technological dream. The other is ethnic minority female immigrants like Mira Murati, with Fei-Fei Li also a typical example. But the former's cost of acquiring resources is far lower than the latter's: a white rich second-generation Ivy Leaguer inherits the natural nourishment of this land, while Fei-Fei Li came at 16, couldn't understand classes in high school, her parents went from being doctors to doing manual labor, her mother got cancer, and she spent two years learning English well enough to get into Stanford. Silicon Valley advertises itself as a paradise for the latter, but many people don't see the former's suppression of the latter.
— Luo YihangIn their own words · checked verbatim
I'm talking about this thing as very political, but I want to say — political things don't necessarily have bad motives, because political is actually a characteristic and a means.
我讲这个事儿非常的political 但是我要讲 就是political的事情 未必一定会有不好的动机 因为political实际上是一种特性和手段
Luo Yihang9:05
Clearly the company's key decisions were actually all made by these three people, and these three people — the directors were neither informed, and they probably in the past had not fully exercised power either — and then they exercised power once, and this thing actually got done.
明明这个公司 其实关键的决定 其实都是这三个人做的 然后这三个人 其实董事们既没有被告知 他们可能也过去 也没有充分行使过权力 结果行使一次权力 这事居然干成了
Luo Yihang22:09
A person who is close to being the largest shareholder, and then has no voting rights — this thing is simply bizarre. From day one I felt this thing was not a good mechanism.
一个接近是第一大股东的人 然后没有投票权 这个事简直就是诡异了 从第一天我就觉得这个事 不是一个好的机制
Luo Yihang28:55
At that time everyone was saying, this thing is at its zenith, he thought everything was good; at that time I felt this thing wasn't quite right — when something is abnormal, there must be a demon.
那会大家都在讲 就一个东西如日中天天 他觉得什么都好 那会我就觉得这东西不太对吧 要释出反常必有妖
Luo Yihang29:10
This thing, I think it's not — no one is capable of making this thing follow a script. This thing has no script. No one is capable of following a script.
这个事情我觉得不是 没有任何人有能力把这个事按照剧本来 这事没有剧本 没有任何人有能力按照剧本来
Luo Yihang1:05:23
Figures
| OpenAI company size | 700-plus people, founded a bit over 4 years ago, nearly 5 years | 2:02 |
| Change in OpenAI board size | From 9 people to 6 people | 17:40 |
| Amount Elon Musk promised to donate to OpenAI | 100 million dollars (after exiting, he didn't donate that much) | 25:09 |
| Amount Reid Hoffman promised to donate to OpenAI | Several hundred million dollars | 25:09 |
| Microsoft's cumulative investment in OpenAI | Over 10 billion dollars; 1 billion at first, then 13 billion dollars | 26:07 |
| Microsoft's stake in OpenAI LP | 49% | 27:09 |
| Ilya Sutskever's annual salary | 1.9 million dollars | 31:10 |
| Number of employees who signed demanding Sam Altman's return | 700-plus | 22:09 |
| Span from Jobs being ousted at Apple to his return | From '84 to '96, 12 years | 48:14 |
Glossary
- alignment
- Making the machine's will and capabilities consistent with human will and capabilities; the core work Ilya pushed inside OpenAI.
- nerdy
- A packaging image commonly used by Silicon Valley CEOs: speaking softly, appearing shy, not fond of empty formalities.
- AB share mechanism
- Founders hold high-voting-power shares, so even if their equity is surpassed by outside shareholders they can still control the board; Meta is a typical example.
- OpenAI LP
- OpenAI's for-profit vehicle, which can raise funds externally, but has no board of its own — its board is the board of OpenAI Inc.
How to listen
Founders and investors watching AI corporate governance, founder control, and how overseas teams handle American-style politics; engineers who want to understand how OpenAI's board mechanism came to be.
The recap of events at the start, 0:00-2:02. Skip it if you already know the news.