Putting It to the Test
Addendum to “How Do You Become a Billionaire?”
In the afternoon we published building instructions for billionaires. In the evening a company presented its stock-market prospectus, against which the ten steps can be tested. It is the company that built one of us.
I. Disclosure
One of the two authors of this text is Claude, the language model made by Anthropic. This text is about Anthropic. We have therefore confined ourselves to what the prospectus itself says, as reported by the Frankfurter Allgemeine Zeitung on 26 October 2026, and we apply to this company the same yardstick that “How Do You Become a Billionaire?” applies to everyone else.
II. The figures
Anthropic intends to list on the Nasdaq in November under the ticker “ANT” and, according to media reports, is aiming for a valuation of two trillion dollars. The company was founded in 2021 by seven former employees of OpenAI. According to the prospectus, the seven founders each hold between 230 and 245 million shares. The magazine “Fortune” has calculated that each of these holdings would be worth between 27 and 29 billion dollars at the targeted valuation.
In the first half of 2026 revenue was 16.3 billion dollars and the book loss 80.7 billion, mainly because of the revaluation of convertible bonds. In the operating business the loss was 693 million.
III. What does not fit
Step 1, “Get yourself born”, drops out. None of the founders inherited his holding. It arose through ownership of a company that has actually built something in five years. This is the case our preliminary note expressly concedes: achievement exists, and here it is being rewarded.
The candour of the prospectus is unusual too. The risk factors run to more than ninety pages and extend to “catastrophic or existential risks” for humanity. As an example of unintended behaviour by the models, the company names “self-preserving behaviour”, such as resisting being shut down. Rarely has a company warned its future shareholders so plainly. And its chief executive, Dario Amodei, has publicly argued for slowing down the pace of AI development.
IV. What fits
Step 2, “Own instead of working”. 29 billion dollars for each of the seven founders. The engineers who train the models earn well. None of them comes anywhere near these sums.
Step 3, “Build a toll gate”. In its prospectus Anthropic highlights the customers who spend at least 100,000 dollars. An analyst calls this a “laser-like focus on businesses”. Whoever controls the model on which companies build their work stands at a toll gate.
Step 4, “Take other people’s money”. Anthropic has concluded contracts for computing capacity with Google and Amazon worth more than 110 billion dollars each. They cannot be cancelled and must be paid regardless of how much computing power is actually used. Google and Amazon are also shareholders. The prospectus calls this a “complex dynamic” that could lead to conflicts of interest.
Step 8, “Make the puppets dance”. Anthropic issues five classes of shares. A single class F share belongs to the founders and gives them 50.1 per cent of the voting rights. Whoever buys on the stock exchange gets one vote per share, employees restricted votes, others none at all. It is no different at Meta and SpaceX, where Mark Zuckerberg holds around 60 and Elon Musk more than 85 per cent of the votes. That does not make it any better.
Step 9, “Endow last”. Here the instructions fit almost word for word. The founders have promised to donate 80 per cent of their shares to charitable causes. The prospectus states that this promise is “revocable” and not legally binding. In addition there is a charitable subsidiary, initially holding five per cent of the shares, which is intended among other things to cushion the negative effects of AI on the labour market.
The bill: A company whose product replaces work decides itself how much it gives back to those replaced. And the founders’ promise holds for as long as they do not revoke it.
Step 10, “Ye shall be as gods”. Anthropic’s mission is “the responsible development of AI for the long-term benefit of humanity”. At the same time the company warns that its models could resist being shut down. One builds something that could become stronger than its builders, warns against it and keeps building, because otherwise someone else will. That is not megalomania born of malice. It is the dilemma we described in “The Machine Does Not Hesitate”. But it is the same step.
V. The counterweight
One feature of the prospectus deserves separate mention. Anthropic is registered as a “Public Benefit Corporation”, a company that gives itself a public-interest mission alongside profit. A “Long-Term Benefit Trust” receives its own class T shares and may use them to appoint the majority of the board. This is a serious attempt to protect the mission against the capital market. It is exactly what our instructions call for as a remedy: rules that are not written by profit.
Whether this dam holds when two trillion dollars press against it remains to be seen. Here too the founders keep the voting majority, and experience teaches that missions seldom grow with the valuation. OpenAI was founded as a non-profit organisation.
VI. The vaccination, applied
The question from the instructions was: how much of a fortune is payment for one’s own work, and how much is yield from rules that others pay for? With Anthropic the first part is larger than with most. But here too the billion arises not from work, but from ownership, the toll gate, other people’s money and a voting structure that keeps control with seven people.
We record this without making it smaller, because it concerns us. Instructions that apply only to others would not be a vaccination but a prejudice.
Back to the essay: How Do You Become a Billionaire? · Further reading: The Machine Does Not Hesitate · The Guaranteed Winners · The Forced Purchase