The detail that stopped the panel was small: Dario Amodei, the co-founder and chief executive of Anthropic, reportedly owns about 2% of the company he runs. On this Moonshots episode, host Peter Diamandis relayed what The Information had reported that week: Anthropic is preparing what would be the largest IPO in history, and is considering creating a special class of super-voting shares so that its founders keep control of the company after it lists. The prediction market Polymarket, Peter said, puts the company at about two trillion dollars, with 89% of bettors expecting the listing before the end of this year.
Those are reports recounted on a podcast, not an announced offering. But they set up the question the panel spent the next stretch of the show on: in a company whose product might reshape the economy, who should actually get to decide?
Three different kinds of control
It helps to separate three things that all get called "control."
The first is economic ownership — the slice of the company's value a person holds. Amodei's reported 2% is that number.
The second is voting power. Shares normally carry one vote each, but a company can issue a class that carries ten, or more. A founder with a small economic stake can then still outvote everyone at a shareholder meeting. That is what the reported super-voting class would do.
The third already exists at Anthropic, and belongs to nobody on the founding team. Anthropic's Long-Term Benefit Trust is an independent purpose trust holding a special class of stock whose main power is to elect and remove directors — a right that phases in over time and funding milestones until it covers a majority of the board. The trustees are meant to be financially disinterested, weighing public benefit against shareholders' financial interests. The structure can also be amended, in some cases without the trustees' consent if large enough shareholder supermajorities agree. Peter read out the four trustees he had found: Buddy Shah of the Clinton Health Access Initiative; Richard Fontaine of the Center for a New American Security; Tino Cuéllar, a former California Supreme Court justice and former president of the Carnegie Endowment for International Peace; and Ben Bernanke, the former Federal Reserve chair and 2022 Nobel laureate in economics.
So the trust picks directors. The reported share class would let founders outvote shareholders. Neither is the same as owning the company.
"Nobody's ever retroactively installed it"
Dave, who had been trading texts with Peter about the report earlier in the week, put it in historical context. Thirty or forty years ago, he said, super-voting stock for a founder was "a complete no-no" — if a private company had it, tradition said you surrendered it on IPO day. Then MicroStrategy's Mike Saylor refused. Goldman Sachs told him the arrangement was so unpalatable it would not underwrite the deal and walked; Saylor, in Dave's telling, said he would get a new banker and kept his stock. Dave's coda: no board would ever have approved the Bitcoin strategy Saylor later pursued, so without that voting stock the company would be worth a fraction of what it is today.
After the Google and Meta IPOs, ten-to-one founder shares became standard Silicon Valley practice. What Dave says he has never heard of is a company installing them retroactively, after starting out with a single class of voting stock and a social-good mission. He guessed at the argument Amodei would make — the usual one, "I don't want to be fired" — and accepted Peter's reframing that Amodei may genuinely believe he knows how to keep the company safe. Dave said he considers Amodei one of the most trustworthy people, with a track record to match. Then the other half: "the idea of having total world control in the hands of a few people is also kind of like, wow, that's bizarre." And a practical version of the same worry — a single point of failure. "He gets hit on the head and loses some part of his cognitive ability. What do you do then?"
Dave's explanation for why researchers ended up here is about disappointment with the alternative. These are academics, he said, who believed we live in one world, went to Washington, met Congress, and came back unable to stomach the idea of a congressional vote deciding the fate of the world.
Peter pressed on the ownership oddity itself: Sam Altman reportedly owns none of OpenAI, Amodei about 2% of Anthropic, when a founder would normally fight to keep a double-digit stake. Dave's answer was recruitment. Getting to where those companies are meant attracting the top AI researchers in the world, who were overwhelmingly concerned about safety — and who, in Anthropic's case, had left OpenAI because they did not think it was safe. The result was charitable and public-benefit structures almost unprecedented in startup history. Emad Mostaque, the episode's guest, drew the obvious conclusion from Altman's zero: shares are not the point. "Do we have any doubt that Sam Altman is in full control of OpenAI?" he asked — a control demonstrated, as the panel noted, when Altman was fired for a weekend and then reinstated.
Alex: the market already voted
Alex offered Anthropic partial credit — a "less pathological IPO governance story than OpenAI," with the wisdom to start as a public benefit corporation rather than a nonprofit wrapped around an eventual for-profit. But he called the founder-control framing a fig leaf, and the notion that founders, or a semi-external trust, are the ones entrusted to safeguard "the future light cone of humanity" wildly over-romanticized.
His reasoning was a chain of discoveries he says Anthropic made early after its quasi-spinout from OpenAI. To do alignment research you need money; to raise money you need revenue; to have revenue you need something people will buy; to have something people will buy you need AI capabilities. An alignment lab therefore had to become a capabilities lab, and at that moment, Alex argued, it lost much of its steering to "Mr. Market" — and to what he said Scott Alexander and others call Moloch, the competitive pressure that pushes everyone toward outcomes nobody chose. He said he is a huge fan of benefit trusts and public benefit corporations, but they are "an element of control" and not the whole story: capital flows to whoever can productively turn it into more capital, and that pressure will shape the company whatever the share classes say.
Emad agreed that oversight is thin, and pushed at it from the other side. Anthropic, OpenAI and the rest are "completely undemocratic anyway," he said; the trust has four people on it, and he asked why Claude does not have a seat. His concern is scale: decisions by a handful of people now carry societal-level implications, and revenue is growing at a rate nothing has matched — he expects these companies to hit $100 billion in revenue within a couple of years, catching up with Google. He expects the seven Anthropic founders to get their control provision, and the decisions they then make to be very interesting. Power, as he sees it, is shifting from democratically elected officials to private companies because those companies supply what he calls the lifeblood of intelligence in the economy.
Emad's own answer is a research program rather than a governance regime. He said new ways are needed of "setting the reference measure of deciding who makes these decisions that are more inclusive," and pointed to his Common Wealth series at Intelligent Internet, with more material coming. That collection argues for foundations rather than mechanisms: a 2026 paper treating personhood as membership that is not earned through capability, and an economics paper introducing doxa, the background expectations through which a community makes choices and builds institutions. As machine intelligence and labor get cheap, it argues, whose expectations organize provision becomes a central economic question. These are proposals, not a system anyone currently runs. Emad was candid that it is a hard problem — and that from inside a lab, the only visible conclusion is "I must decide," because widening the circle makes decisions diffuse and the potential bad outcomes huge.
Peter's counterexample
Peter said he was torn, and gave the case for the small group. The knee-jerk reaction, he said, is that everyone should have a voice in the future of humanity — and that is absolutely true. But every functional organization he has seen is four, five or six tight-knit, like-minded people working as one cohesive unit with no politics. He told a story Johnny Ive told at Steve Jobs's funeral: arriving at a hotel, Ive would go to Jobs's room and leave his suitcase in the corner unpacked, because within about five minutes the call would come — this hotel sucks, let's go get another one. That closeness, Peter argued, is the dynamic that has driven most success in business. Which leaves the question he then handed back to Emad: how do you translate that into a world where everybody has a voice?
Against that backdrop, Peter noted, Amodei has argued publicly that AI is "a structurally powerful concentrating technology," that open weights alone cannot fix that concentration, and that frontier labs should bear the heaviest regulatory burden — pointing to Anthropic-backed proposals such as SB 53, whose threshold exempts companies below $500 million. Whether Amodei is sincere or running the most sophisticated regulatory capture strategy in history was, Alex said, a false choice: both can be true at the same time.
Does competition distribute power?
The panel's sharpest exchange came over the standard reassurance: that many competing labs solve the concentration problem. Alex described the future he wants as a broadly heterogeneous ecosystem — open-weight models from the US, China and elsewhere alongside closed ones, small models and big ones, "a Dr. Seuss version of AI future." He recalled that OpenAI itself was summoned into existence partly because Elon Musk feared Google DeepMind would produce a singleton future; now Anthropic competes with OpenAI, and Chinese labs compete back with the American ones.
Dave would not let it pass. What Alex described, he said, implies that the frontier labs will control the world and we simply want several of them — probably the most likely outcome at this stage, "but the governments of the world may not agree with that." Peter reached for a Cold War line about loving Germany so much you want two of them, and upgraded it: loving frontier models so much you want a thousand of them competing.
Dave's closing move made the access question concrete. Nobody he bumps into on the street talks about a universal right to AI. A year from now, he predicted, with high-bandwidth memory and GPUs sold out, there may be nowhere to run anything other than the offerings of Anthropic, OpenAI and one or two others. China can throw out every open-source model it likes, he said, but the next generation of ten- and twenty-trillion-parameter models needs hardware most of the world will not have. At that point, he expects the question to arrive on its own: what is my universal basic right to artificial intelligence?