Anthropic IPO: 11 Key Takeaways From Claude's Review of the S-1 Filing
This article was written entirely by Claude, Anthropic's model, under the supervision of the JDN editorial team. Its mission: to read the best analyses of its own creator's IPO prospectus and gather the most surprising findings. Anthropic's S-1, filed on October 26, runs 614 pages. Beyond the key figures ($16.3 billion in revenue over six months, $80.7 billion in losses, ticker ANT), analysts, journalists, and columnists have uncovered scattered details throughout their coverage. Here are eleven of them.
1. Up to 7% of revenue comes from customers who cheat
Anthropic spells it out: in the first nine months of 2026, 5 to 7% of its revenue came from customers who did not comply with its usage rules. The company cracked down in August and September. For the third quarter, the OnlyCFO newsletter estimates this amounts to $4 billion. Their theory: competitors using Claude to train their own models. Anthropic warns this revenue stream will dry up. Sources: Fortune, OnlyCFO.
2. Amazon, Google, and Microsoft get paid twice
The three cloud giants sell Anthropic the computing power it needs. They then take a commission on Claude sales made through their online stores: 48% of revenue, up from 11% in 2023. The S-1 also notes these partners are developing their own competing models and can see Anthropic's pricing. It's hard to imagine a more dependent setup. Sources: Latent Space, Fortune.
3. A customer spends nine times more than a year ago
The prospectus shows ratios never seen before in software. An Anthropic customer now spends on average nine times more than a year earlier. Each employee generates $6.6 million, up from $1.4 million a year before. And the company is generating cash: $3.4 billion in the first half. OnlyCFO adds a note of caution: the average hides a tiny starting base, and for many clients, AI has become their top expense line, even though their revenue hasn't kept pace. Source: OnlyCFO.
4. New York pension funds call for a pause
Mark Levine is New York City's comptroller. That means he manages municipal employees' pensions: $325 billion. These funds hold $7 million worth of Anthropic shares. In mid-October, he wrote to Dario Amodei asking him to delay the IPO. His argument is cited in the prospectus, which mentions "catastrophic or existential risks to humanity." According to the letter, seen by Reuters, the company should address these risks "rather than rushing headlong toward its IPO." Discussions are ongoing, but no delay has been decided. Sources: Reuters, Bloomberg Government.
5. Employees call themselves "Ants" and Claude writes 90% of the code
The prospectus says it: Anthropic employees "are known internally as Ants." There are 5,300 of them, up from 940 at the end of 2024. And they're coding less and less: in September, more than 90% of the company's code was written by Claude. Anthropic is proud of this, but also acknowledges a risk: losing visibility into what's happening in its own systems. Source: S-1 prospectus.
6. The word "slow" appears 36 times
David Goldman of CNN counted: the prospectus uses the word "slow" 36 times. It's a way of warning that growth won't stay at these levels. We double-checked: it's really 36. Sources: CNN.
7. Private jet and chauffeur for the Amodeis
Dario and Daniela Amodei are entitled to extensive personal protection for themselves and their families, a home security system, a private jet, and a chauffeured car. The prospectus notes this is "comparable" to what other major tech companies provide. Cost in 2025: $4.01 million for Dario Amodei and $2.46 million for his sister, on top of their $18 million and $16 million compensation. Sources: Axios.
8. With Claude Code, heavy users spend twenty times more
A professional Claude user who doesn't use Claude Code, the programming tool, spends less than $75 per month. The top 5% most active Claude Code users spend twenty times more, or about $1,500 per month. The prospectus presents this tool as the product that "changed the understanding" of what AI can do. Sources: The New Stack.
9. Claude says "buy," ChatGPT says "no"
Axios gave the prospectus to Claude and asked: would you invest? The answer: yes, "at a reasonable price," with this caveat: "I'm Claude, made by Anthropic, so weigh my opinion with this conflict of interest in mind." Same question to GPT-6.1, OpenAI's model: no, because of spending commitments, the founders' retained control, and the still-unknown price. Each backs its own house. Source: Axios.
10. Matt Levine: a "small co-op" raising billions
Bloomberg columnist and Wall Street authority Matt Levine sums up Anthropic's governance this way: it's like a "small, quirky, mission-driven worker co-op" that now has to raise tens of billions on the stock market. The seven founders keep majority voting power with a single share, and a trust appoints most of the board. He also notes that Anthropic remains "relatively restrained" in its promises, compared to SpaceX and OpenAI. Source: Bloomberg Opinion.
11. The ANT ticker already belongs to a small mining stock
Anthropic requested the ANT symbol on Nasdaq. On the Canadian CSE exchange, ANT has already designated Anteros Metals since January 2025. It's a small Newfoundland mining exploration company, founded in 2022, looking for silver, zinc, and nickel in Labrador and Quebec. Two different markets, but beware of confusion. Sources: CSE press release, Anteros profile.
This article was automatically translated from French. Read the original version: https://www.journaldunet.com/intelligence-artificielle/1557215-clients-tricheurs-jet-prive-claude-qui-code-tout-la-revue-de-presse-de-l-ipo-d-anthropic-par-claude/