On a Tuesday morning I opened a crypto brief that told me, in one sentence, that Anthropic had made its public debut and that Eric Schmidt was sitting on a transformative windfall. No exchange. No ticker. No price range. No raise size. No underwriters. I spent forty minutes on SEC EDGAR looking for an S-1 that does not exist, then cross-checked the company's own announcements, then went back and read the sentence again. It was the same sentence. The headline and the body were identical. That identity — headline restated as body, with zero verifiable elements — is not a reporting gap. It is a fingerprint.
Here is what the record actually shows. As of late 2025, Anthropic remains a private company. Its September 2025 Series F raised roughly $13 billion at a post-money valuation near $183 billion. That is private equity, not a public listing. There is no S-1 on file, no registration statement, no named bookrunner. A real IPO story cannot omit exchange, symbol, pricing range, proceeds, and syndicate — those five items are the load-bearing walls of the genre. When all five are missing, you are not reading a truncated report; you are reading something that was never built as one.
I can construct four explanations, and I hold them loosely. First, terminology drift: a machine or a rushed writer converted 'new mega-round' or 'IPO preparation chatter' into 'public debut.' Second, prediction-market transcription: crypto desks increasingly source stories from Polymarket and Kalshi odds, and a shift in an 'Anthropic IPO by 2026?' contract can be rewritten into a fact sentence by the second hop. Third, tokenized pre-IPO exposure: platforms do sell tokenized claims on private-company equity, and 'Anthropic now available' is a very different sentence from 'Anthropic now listed.' Fourth, the report is simply future-dated and I lack the timestamp.
Notice what all four have in common. None of them is IPO journalism. Each is a story about capital access, odds, or wrappers — dressed in IPO vocabulary.
The valuation curve is the only part of this I can actually compute, and it is more interesting than the rumor.
Roughly: pre-2023 at about $4 billion post-money; Series C in May 2023 near $5 billion; Series D in 2024 around $18.4 billion; Series E in March 2025 near $61.5 billion; Series F in September 2025 near $183 billion. An early-2022 entrant is sitting on something like 45x on paper. A Series E entrant, about 3x. Same company, same quarter, wildly different 'windfall.'
The word windfall is doing fraudulent work here, and it is doing it at the level of mechanics rather than honesty. An IPO is a liquidity event for the company's balance sheet and a liquidity path for shareholders. Standard lockups run 180 days, and early holders typically release in tranches. What an early investor actually receives is decided by the price six to twelve months after the bell — not by the first print. I have watched enough 2021-vintage high-multiple listings break issue price inside a lockup window to stop treating day-one pricing as an outcome.
Then there is the multiple itself. At $183 billion against a run-rate in the neighborhood of $7 billion, you are near 26x price-to-sales. At the rumored $350 billion, near 50x. Mature high-growth public software and cloud names have spent years in a 15-30x band. So the private mark has already pulled forward one to two years of growth that the public market has not yet agreed to pay for. That gap — not Schmidt's paper gains — is the actual story, and it is the one the brief omitted.
Gross margin is the second omission. Inference cost lands in cost of revenue. API-led revenue does not carry the 75-85% gross margins that pure software does. Every base-model company that walks into a public filing gets repriced for that, and any investor who has not modeled it is modeling a SaaS company that does not exist.
The same filings will also have to disclose long-term compute commitments to cloud providers — obligations that read as strategy in a pitch deck and as fixed cost on a balance sheet. Public markets have not yet settled on how much 'compute debt' they will tolerate, and that tolerance will set the discount rate for the entire category.
I have a specific reason to be dull about this. In 2017, during the ICO mania, I was the community liaison for MakerDAO's early team in Cape Town, and I personally vetted more than 200 community submissions, filtering scams out of the queue while trying to teach real participants what an unbacked stablecoin actually is. The pattern I learned then is the pattern I am looking at now: a single unverified sentence, repeated with confidence, moves more capital than a boring document ever will. In 2020, running SoulBound workshops for women in emerging markets, I taught people to read a prospectus rather than a headline — thirty live sessions, and the same lesson every time: if you cannot name the five load-bearing facts, you are not investing, you are joining.
On Schmidt specifically: the named major investor lists I can verify — Google, Spark, Menlo, Lightspeed, ICONIQ, Fidelity, General Catalyst, Jane Street, Baillie Gifford, MGX, plus strategic money from Microsoft and Nvidia — do not include him or a family office under his name. A sub-5% position held through an SPV or bought on the secondary would never surface in a major-holder table. So 'early investor' is neither provable nor disprovable from the public record. That is precisely the shape of a claim that travels well and informs nothing.
Here is where I will disagree with almost everyone who covered this.
The debate has been about whether the story is true. That is the wrong question, and it is the question the story wants you to ask. The fact that a one-sentence, source-free item about a private AI lab could travel through crypto channels at all is the data. Crypto desks now cover AI capital events because AI is where attention and allocator interest have migrated — and because tokenized pre-IPO wrappers, prediction markets, and secondary share desks give that coverage a commercial spine. That grey zone was entirely absent from the piece. Solidarity over speculation means naming it: when your exit route is a wrapper product, your news incentive is not accuracy, it is urgency.
The overlooked casualty is not Schmidt. It is every unlisted AI founder and employee whose paper equity is quietly benchmarked to whatever multiple a first base-model listing prints. If that print lands at 15x rather than 50x, the markdown propagates backward through the entire private stack — and nobody writes a one-line brief about that.
And the piece never asked who is selling. An IPO's immediate beneficiaries are early funds and employees gaining liquidity, not the company. Framing it as an investor windfall is, whether intended or not, an advertisement for an exit.
Watch the S-1, not the sentence. The first public multiple assigned to a foundation-model company will become the reference price for an entire asset class — and that number, not any founder's paper gain, is what will reprice the next two years of AI private markets. Culture on-chain, heart on-screen: our job is to keep the two honest with each other. Code is law, but ethics is conscience.