The number arrived on a Tuesday, the way most bad numbers do — quietly, in a headline, dressed as fact.
Anthropic. $965 billion. IPO speculation. Three clauses, stitched together by a crypto news outlet I have cited perhaps once in four years, citing a Wall Street Journal story that, at the time of writing, I have not been able to independently confirm. Within six hours the figure had metastasized. Group chats. Screenshots. A tokenized pre-IPO market that suddenly quoted Claude's parent at a premium to Berkshire Hathaway. Nobody had seen a term sheet. Nobody had seen a Series lead. Nobody had asked the only question that matters when someone hands you a number with three commas in it: compared to what?
I have been doing this for twenty years, which means I have watched the crypto industry fall in love with precisely this kind of number at least four times. In 2017 it was the market cap of projects with eleven GitHub commits and a white paper written in Word. In 2021 it was the floor price of a JPEG of a bored ape. In 2022 it was the fully-diluted valuation of a token whose only utility was the ability to vote on its own emissions schedule. Each time, the same architecture: a number, a narrative, a transmission channel, and a public that was told the number was the story. The number is never the story. The transmission is.
What follows is not a takedown of Anthropic. I have no particular quarrel with Claude or with the people who build it. What follows is a forensic reconstruction — how a valuation of roughly one trillion dollars entered public discourse with less evidentiary support than a mid-cap DeFi governance proposal, and why the market for that ambiguity was, in a week of a bear market that has already cost my readers 40% of their liquidity positions, the single most instructive thing that happened.
Context: What Anthropic Actually Costs
To understand why $965 billion should trigger every alarm bell in your analyst toolkit, you have to hold two facts in your head at once. The first is that Anthropic is a serious company with serious capital behind it. The second is that the gap between 'serious' and 'the fourth most valuable company on Earth' is roughly an order of magnitude, and orders of magnitude are the one thing a private market cannot manufacture by vibes alone.
Let me give you the floor. As of the middle of 2024 — the last point at which I have reasonably triangulated data, and I will flag the cutoff because my own model knowledge stops there and the claim in question postdates it — Anthropic had raised at valuations that clustered in the low-to-mid tens of billions. The Amazon commitment alone was $4 billion, expanded in tranches, with a second $2.75 billion tacked on. Google wrote a check that was widely reported at similar magnitude. The series that pushed the company into the $18 billion range was a genuine milestone, but $18 billion is not $965 billion. It is roughly one fifty-third of it. The distance between them is not a funding round. It is the distance between a regional airline and Boeing.
Now, private valuations do jump. That is the whole charm of the private market and the whole reason secondary brokers exist. Between 2021 and 2024, OpenAI's reported valuation went from roughly $14 billion to a number north of $80 billion in fewer than thirty months — a five-to-six-x move that shocked people who had not internalized how much of the AI trade was being priced on option value rather than discounted cash flow. But five-x is not fifty-x. A fifty-x move in a company with no public financials, no prospectus, no S-1, and a governance structure explicitly designed to resist certain kinds of investor pressure is not a funding round. It is a rumor wearing the clothes of a funding round.
Here is the thing that a twenty-year observer learns to sense in the gut, before the spreadsheet catches up: when a valuation number arrives without a lead investor, without a funding amount, without a pre/post-money designation, and without a single line of the round's structure, the number is not data. It is a headline that has been separated from its context. And I have seen what happens when headlines get separated from their context. From the ashes of 2017 to the fluidity of DeFi, that separation has been the single most reliable predictor of retail pain.
Core: A Forensic Reconstruction of a Number
The Arithmetic Does Not Close
The single most important sentence in this entire article is a typographical observation: $965 billion and $96.5 billion differ by one decimal point and a factor of ten.
That is not a nitpick. In the history of financial journalism, the misplaced decimal is the single most common category of error, and it is especially common when a number crosses from a business desk to a general-interest aggregator to a specialist outlet and back again. The original Wall Street Journal headline, if it existed in the form claimed, would have been written for an audience of institutional readers who understand the difference between a valuation and a raise. By the time it reached a crypto-native publication, it had been compressed into a single clause — 'Anthropic, valued at $965B, considers IPO' — and pushed to a feed that reads that clause in seven seconds between two other posts about a liquidation cascade and a meme coin.
I want to be precise here, because the accusation is serious. I am not saying the crypto outlet lied. I am saying that the transmission of a number through three media layers is a lossy process, and that the losses are systematically biased toward bigness. Nobody clicks a headline that says 'Anthropic raises at a valuation consistent with expectations.' Everybody clicks the one with the trillion-dollar adjective. This is not a conspiracy. It is an incentive gradient, and every journalist who has ever written a headline knows it is there.
The second arithmetic problem is pre/post-money. When a reporter writes 'valued at X,' they often mean post-money — the number you arrive at after the raise closes. When a source quotes a number informally, they often mean pre-money, or they mean the number that came up in a term-sheet conversation that may or may not have closed. And when a deal involves strategic investment, cloud credits, convertible notes, or secondary sales of existing shares at a discount, the headline number can be a chimera: a blend of cash, commitments, and paper that has been rounded up for negotiation purposes. A $96.5 billion post-money valuation on a round that includes, say, $4 billion of compute commitments and $2 billion of convertible structure is a very different animal from $96.5 billion of hard equity. Nobody reading the Crypto Briefing summary — and I read it, twice — could possibly know which one they were looking at.
The third arithmetic problem is the one that gets the least attention: what revenue multiple does $965 billion imply, and is that multiple within the range that serious investors have ever paid for anything?
Let me try to be honest about what we don't know, because I have been on the other side of this enough to know that confident analysts who fill in blanks with assumptions blow up accounts. Anthropic does not publish revenue. Reports during the middle of 2024 suggested annualized revenue in the hundreds of millions, with a trajectory that could plausibly reach the low billions by mid-decade. Multiply any reasonable version of that number by fifty, and you are still nowhere near a trillion. At $965 billion, you would need annual revenues approaching the scale of a company like SAP or Oracle to justify even a generous multiple — and Anthropic is not SAP or Oracle. It is a four-year-old company selling inference to enterprises at prices that the market is pushing down, not up, and doing so while paying what is arguably the highest marginal cost in the history of software to NVIDIA and its cloud partners for the privilege.
I have spent enough of my career inside whitepapers that I can tell you the shape of this trap from a mile away. In 2017, the trick was to describe market cap as if it were revenue. In 2021, the trick was to describe floor price as if it were demand. Here, the trick is subtler and more dangerous: to describe a narrative about safety and prestige as if it were a discounted cash flow statement.
The Transmission Chain Is the Story
I keep a small ritual when a number this size surfaces. I open a fresh document and I write down every hop the number made between its source and my eyes. Then I ask: at which hop did the confidence level rise?
For $965B, the chain looks like this. Origin: an unnamed source in a Wall Street Journal report — and unnamed sources, even at the Journal, are sources expressing a possibility to a reporter who has an incentive to frame it as more definite than it is. Hop one: the Journal publishes a reported headline, ideally with context, caveats, and a specific word like 'could' or 'reported.' Hop two: a crypto-native aggregation outlet picks it up, drops the caveats because they do not fit the format, and republishes as a fact. Hop three: crypto Twitter screenshots the aggregate, circles the number, and it is now a datum. Hop four: a tokenized pre-IPO market lists a synthetic instrument, or a Telegram broker offers 'Anthropic exposure,' and the number becomes priced.
By hop four, the number is no longer a claim about Anthropic. It is a claim about the market's belief in Anthropic, which is a fundamentally different thing. And the market that prices tokenized pre-IPO exposure is not the market that reads the Journal. It is a market composed of retail buyers who want access to the most talked-about private company on Earth and who are willing to pay a premium for the appearance of access. That market is the one I write for. And that market is the one that will lose money if the number is wrong.
I have seen this exact sequence before. In 2021, it was secondary shares of a company that was 'definitely IPO-ing next quarter' for two years running. In 2022, it was the pre-launch token markets, where a perpetual's premium to spot told you more about the crowd than about the asset. The mechanism has not changed. It has simply been re-skinned for the AI era, and the skin is convincing because the underlying companies are real, the technology is real, and the hype is — for once — largely deserved.
That last part is what makes this so much more dangerous than the 2017 pattern. In 2017, the scam persisted because the technology was fake. In 2026, the distortion persists because the technology is real, and the reality of the technology is being used as collateral for a claim the technology does not support.
Why Crypto Media Is Covering an AI Valuation
Let me address the question that a sharp reader is already asking: why is a crypto outlet the transmission vector for an AI valuation story?
There are three answers, and none of them flatter my own industry.
The first is attention economics. Following the 2024 spot ETF approval and the subsequent institutionalization of the major assets, crypto-native outlets found themselves in a strange position: their core beat had become less volatile than the broader tech beat. The AI cycle offered higher beta. When Anthropic, OpenAI, and xAI became the most-discussed private companies on Earth, crypto media discovered that its audience — young, risk-seeking, terminally online — was the same audience that wanted to trade that narrative. So the beat shifted. I participated in that shift myself; my own vertical, TradFi Meets DeFi, exists because the border between the two worlds stopped being a border.
The second is the tokenization hook. The most energetic part of my industry right now is not DeFi in the classic sense. It is the attempt to build markets for things that previously had no market: private company equity, carbon credits, real estate, and — most pertinently — pre-IPO AI exposure. If you believe that Anthropic is worth $965 billion, then a tokenized slice of Anthropic is one of the highest-return instruments available to a retail investor in 2026. So a great deal of energy went into not questioning the number, because questioning the number would kill the instrument.
The third answer is the ugliest: crypto media rewards numbers that make readers feel they are early. Nothing in the history of my industry generates more engagement than the sensation of having access to something before it becomes obvious. A $965 billion Anthropic is a story about the future arriving now. A $96.5 billion Anthropic is a story about a company doing well at a valuation the market already knew. The first story is worth a thousand repl> tweets. The second is worth a paragraph. I have no moral authority here — I have written the first story more times than I care to admit. But I have also learned, at the cost of real losses for real readers, to be suspicious of the version of the truth that happens to be the more exciting one.
Safety as a Valuation Instrument
Buried under the number is a second narrative claim, and this one deserves its own investigation because it is doing more load-bearing work than the valuation itself. The claim is that Anthropic's AI safety focus — Constitutional AI, the Responsible Scaling Policy, the interpretability research — constitutes a durable competitive moat and a justifiable basis for premium valuation.
I want you to hold two things true at once. The first: safety research at the frontier genuinely does matter, and serious alignment work is not marketing. The second: when 'safety' appears in the same headline as a nine-hundred-billion-dollar valuation and an IPO rumor, you are watching a mission statement get repriced as an asset. And the moment a mission statement becomes an asset, its incentives change.
Here is what my twenty years around incentives has taught me. A company that is valued primarily on technology is incentivized to ship technology. A company valued primarily on revenue is incentivized to grow revenue. A company valued primarily on safety and trust is incentivized to be perceived as safe and trustworthy — which is not the same thing as being safe, and which can be satisfied by communication, benchmarking, and governance theater, at a fraction of the cost.
I have no evidence that Anthropic is engaging in theater. What I am saying is that the incentive structure that a $965 billion valuation creates makes theater rational, and I have seen what rational theater does to an industry. In 2022, the responsible-disclosure framing around dozens of DeFi protocols became a marketing surface. Some of those protocols were genuinely careful. Some of them were borrowing the language of care to attract capital that would have otherwise gone to competitors with cleaner code but worse branding. You cannot tell the two apart from outside the building. And you cannot tell them apart from a headline that uses the word 'safety' as an adjective.
The second piece of this is the effective altruism connection. I will be careful here, because the accusation has been leveled at a lot of people who did not deserve it. But I would be a poor journalist if I did not note that the EA-adjacent governance philosophy embedded in Anthropic's founding has become, since the FTX collapse, a subject of genuine public skepticism. In 2022, I spent the first quarter of the year writing what turned out to be — read the archive, it holds up — an anatomy of the narrative that was then called 'effective altruism.' What I found, and what I wrote then, was that EA had one extraordinary strength and one extraordinary weakness. The strength was its ability to attract and organize genuinely brilliant people around a long-termist vision. The weakness was that its framework had no native mechanism for noticing when the long-term was being used to launder the short-term.
That is not a claim about Anthropic specifically. It is a claim about a governance idea that has been used, in at least one famous case, to build a machine that moved user funds from a venture to a charitable fund. The investors looking at a $965 billion Anthropic should read that history carefully, because the story they are being sold — a safe, mission-aligned AI company at a trillion-dollar valuation, led by people with a long-termist worldview — is the same story that sold a lot of people on FTX, adjusted for inflation and with better code.
The Cloud Vendor Is the Balance Sheet
There is one more load-bearing fact, and it is the one that the crypto coverage universally omits, which is the entanglement with the hyperscalers.
Anthropic runs on Amazon Bedrock and Google Vertex AI. It buys compute from AWS and Google Cloud. And AWS and Google are investors. This is not a footnote. It is the central fact of the company's capital structure. When Amazon commits $4 billion and then expands it, the capital is not a clean injection of cash. It is, in many cases, structured as compute credits, cloud commitments, and strategic arrangements that keep Anthropic's consumption inside the Amazon ecosystem. The financial effect is somewhere between 'revenue' and 'vendorship': a company that is paid in the right to consume the product of the investor.
A $965 billion valuation, if it existed, would almost certainly not be $965 billion of cash. It would be a construction in which a big chunk of the 'funding' is committed cloud spend, a chunk is convertible structure, a chunk is secondary purchase of existing shares, and a comparatively modest slice is primary cash. This is the arithmetic that got lost on the way from the Journal to the feed.
And there is a second cloud-related distortion, and it goes in the opposite direction. Suppose Anthropic really is heading toward an IPO at a very high valuation. Then the closest comparable is not OpenAI — it is a hyperscaler. The hyperscalers trade on revenue, on EBITDA, on free cash flow. Anthropic's unit economics, if you read the public fragments, are dominated by inference cost, which is determined by GPU scarcity, which is in turn determined by NVIDIA's allocation and the export-control regime. A company that is dependent on AWS for compute and on NVIDIA (through AWS) for silicon is not a company with a durable multiple. It is a company with a durable dependency, and dependencies do not get the multiple that the narrative is pretending they do.
I have audited enough tokenized exposure structures to know this pattern. In 2021, a lot of miners sold themselves as 'digital asset infrastructure' and were valued on the price of BTC. They were not valued — they were beta. The moment BTC corrected, the multiple collapsed to hardware. Anthropic is, at the valuation being rumored, the same shape: not an operating company, but a beta on the AI capex supercycle, which is itself a beta on the belief that the hyperscalers can sustain their spend indefinitely. That bet might be right. It is definitely not a $965 billion bet made on the basis of one unnamed source.
Contrarian: What If the Number Is Real?
Every essay like this should steelman its target, and so let me do the uncomfortable work. Suppose the number is real. Not $96.5 billion — $965 billion. What would have to be true?
It would require three things simultaneously, and I want to hold each of them up to the light.
First, Anthropic would need to have transformed its revenue trajectory from hundreds of millions to tens of billions in the space of roughly eighteen to twenty-four months. This is not impossible — it is roughly the rate at which Microsoft's cloud grew in its first post-Azure wave, and it is plausibly the rate at which OpenAI's enterprise revenue has scaled. But it requires a customer mix that includes governments, regulated industries, and Fortune 100s deploying Claude as a primary inference layer rather than a secondary one. I have not seen a single disclosed contract at that scale. If such contracts exist, they are among the best-kept secrets in enterprise software.
Second, Anthropic would need to have secured compute at scale that does not appear in any of the public cloud capex commentary. Because if the company were consuming GPU fleet in the tens of thousands of H100-class accelerators or TPU v5s, the data center buildout and the NVIDIA earnings calls would reflect it. NVIDIA's 2025 and 2026 commentary does not, as far as I have seen, disclose a single customer of Anthropic's scale. That absence is not conclusive, but it is informative. You cannot scale a nine-hundred-billion-dollar AI business without leaving a footprint in silicon, power, and data center square footage, and I can find no such footprint in the public record.
Third, the company would need to have crossed a governance threshold: it would need to have convinced, simultaneously, the VA-adjacent governance guardians and the new institutional capital that the Responsible Scaling Policy is not a constraint on the growth story. In my experience, these two constituencies want incompatible things from a company at trillion-dollar scale. The governance people want release discipline; the capital people want release cadence. A single public incident — a model achieving a dangerous capability and being released anyway, or a model withholding release and losing a market that competitors scoop up — would crack the narrative permanently.
So my steelman does not hold. Not because $965 billion is impossible in the abstract, but because no public evidence exists to support any of the three required premises, and the headline that distributed this number provided none of them either.
Which brings me to the contrarian conclusion that turns the entire exercise on its head. The interesting question is not whether $965 billion is true. It is why the crypto market needed it to be true.
A bear market is a machine for generating scarcity. In 2022, when the market had nothing to celebrate, it decided the story was 'real yield.' In 2023, when the pain was over but the confidence was not back, the story was 'RWA tokenization.' In 2026, in the middle of a drawdown that has removed 40% of some LPs' capital, the story is 'AI.' And within AI, the story is 'Anthropic, the safe one, about to IPO at a trillion dollars,' which is a fantasy that combines four of retail's favorite narratives — safety, exclusivity, exponentiality, and a ticker. The number is not the story. The number is the magnet, and the story is the iron filing.
I have been on the wrong side of that magnet more than once. In 2021, I wrote a piece I am not proud of defending a floor price that turned out to be a liquidity illusion. I learned the lesson the way everyone learns it: by watching the people who trusted me get hurt. The lesson, stated precisely, is this. A market in a bear phase does not require accurate numbers. It requires numbers that are large enough to matter and vague enough to survive scrutiny. $965 billion is precisely that kind of number. It is large enough to make every reader feel they have missed something. It is vague enough that no one can falsify it without access to documents they will never see.
Takeaway
Here is what I would tell you if you were sitting across from me in Berlin, at the café I used to write from before the vertical launched, the one where I did my first on-chain forensics write-up. Do not trade the number. Trade the chain — know every hop between the source and your ears, and price your exposure to the uncertainty of the chain, not the certainty of the claim.
Four signals I will be watching over the next one to three months. First: whether the Journal corrects or clarifies the figure within two weeks of the original publication — corrections, if they come, will almost certainly be quiet, and their quietness will tell you everything. Second: whether Anthropic itself issues a statement about the number, in which case the precise wording of that statement will disclose the pre/post-money structure faster than any analyst will. Third: whether secondary markets for pre-IPO AI exposure hold their premium after the initial news cycle fades, or collapse back toward the underlying valuation — a premium that decays within thirty days is a premium that was never real. Fourth: whether the next major AI funding round announced by any peer — xAI, Mistral, OpenAI itself — resets the anchor, because these numbers do not live in isolation, and an anchor that moves too far ahead of the pack is an anchor that snaps.
What I do know, with the certainty that twenty years around this industry earns you, is that the number was never Anthropic's. It was the market's. And markets that build their hopes on a single unverified figure tend to learn something they would rather not have known. The 2017 ICOs learned it when the whitepapers went quiet. The DeFi protocols of 2020 learned it when the yield stopped. The NFT floors learned it when the exits closed. The AI trade is the newest version of a very old pattern, and this time the pattern is being sold with a decimal point.
There is a version of this story where Anthropic is worth a trillion dollars one day. I hope it is true. But the company that gets there will be built on revenue, silicon, and shipped features — not on a headline that wandered out of a Wall Street Journal newsroom, lost a decimal point somewhere between New York and a crypto feed, and arrived in your group chat wearing the costume of due diligence.
Which version of Anthropic are you pricing in?
