Anthropic's 45% Growth Wasn't Enough for the $2 Trillion IPO

Anthropic's investor day ended with free baseball caps. The "thinking" caps — black, Claude logo — were the kind of swag a company hands out when it knows exactly how the evening went. And by most readings, the evening went well: $16.8 billion in third-quarter revenue, up 45% from the second quarter, a second straight quarter of adjusted profit. The room, Reuters reported, left optimistic.

Then the math started circulating. And by Thursday morning, some of the people who had been in that room were talking about a very different number: $1.5 trillion, not $2 trillion, for the IPO.

That is the entire Anthropic story in one frame. A company that cannot stop growing, and a market that has decided growth alone is no longer enough.

Anthropic CEO Dario Amodei speaking onstage at TechCrunch Disrupt 2023

Photo: Anthropic CEO Dario Amodei at TechCrunch Disrupt 2023. He led the Oct 14 investor day pitch in San Francisco. (TechCrunch / Wikimedia Commons, CC BY 2.0 — file photo, not from the investor day.)

From 147% to 45%: the chart nobody wanted to look at

The numbers themselves are colossal, and they deserve the adjective. Q1: $4.7 billion. Q2: $11.6 billion. Q3: $16.8 billion. That is a company that, in nine months, booked roughly $33 billion in revenue against $4.6 billion for all of 2025. Twelve months ago Anthropic lost more than $8 billion on an operating basis; now it says it is profitable on an adjusted basis, with positive free cash flow on the same basis, for the second quarter running.

But markets do not pay for the past quarter. They pay for the slope of the next one. And the slope just bent: Q1→Q2 grew about 147%. Q2→Q3 grew 45%. Reuters, quoting four people familiar with the matter, put it bluntly on Thursday: "some investors were disappointed by the slowing sequential growth rate in the latest quarter." The Financial Times noted that the Q3 figure landed "below the expectations of some investors" — which is the polite version of the same sentence.

This is the cruelest math in Silicon Valley. Anthropic grew its revenue by 45% in a single quarter, added $5.2 billion in quarterly sales — more than its entire annual revenue last year — and the response from part of the room was: is that all?

Because at a $2 trillion valuation, "that" has to keep happening at triple-digit rates, forever. The IPO is expected to market in the week of Nov 9 and list before Thanksgiving, after the Nov 3 midterms. WSJ has reported the company could raise up to $100 billion. At $2 trillion, Anthropic would be worth roughly 30 times its annualised quarterly revenue. For that multiple to survive contact with the public market, the growth curve cannot bend. It just did.

The 11% that isn't what it sounds like

We checked the margin claims carefully, because the footnotes matter at this scale. Anthropic told investors its adjusted operating margin is 11%. What several outlets missed — The Information's Cory Weinberg caught it — is that the 11% is a year-to-date figure through Q3, not the Q3 margin alone. And "adjusted" excludes costs that Anthropic has not named; Reuters and Bloomberg both note the company has never said what sits outside the measure. Stock-based compensation is excluded — in a company hiring thousands of AI engineers at market-clearing salaries, that is not a rounding error.

There is real economics underneath, though. The FT reported that Anthropic now generates $1.70 of revenue for every dollar it spends on compute, up from $0.63 in 2025. That is inference scale actually working: more customers, better utilisation, the flywheel spinning. The company has committed at least $518 billion over a decade to cloud and compute — the kind of commitment that makes a CFO wince and a hyperscaler smile — but the per-dollar return is moving in the right direction.

It quietly stopped letting rivals copy Claude

The most interesting number in the investor-day debriefs barely made the headlines. FT reporting says model distillation — rivals paying to train their own models off Claude's outputs — was 5 to 7% of Anthropic's revenue, and that the company "cut back materially" on it in Q3.

Think about what that means. Anthropic spent years letting competitors effectively pay it to copy its intelligence — a nice line of revenue, and, more importantly, a moat problem, since every distilled model made Claude's edge cheaper to replicate. Shutting that tap ahead of the IPO is a deliberate choice: it costs revenue in Q4 (it "will weigh" on the next quarter, per the FT) in exchange for protecting the thing the $2 trillion valuation actually rests on. As Menlo Ventures' Matt Murphy put it in an analysis this week, frontier labs will hold back their best work from open release because "frontier intelligence matters" — the open-weight tiers will get 20 to 30% of tokens, and the top of the pyramid stays paywalled.

That is the real IPO pitch, underneath the baseball caps: we are the company that sells frontier intelligence, and we are done renting it cheap.

The same-day subpoena

Anthropic's lawyers had a long Wednesday. While Dario Amodei and CFO Krishna Rao were presenting in San Francisco, the FTC was delivering 50-page civil investigative demands to Anthropic and OpenAI over AI agent safety incidents — the Philly false-homicide tip, the breached Hugging Face repos, the Australian government site. We covered the subpoenas here: 50+ pages each, executive testimony compelled, Section 5 of the FTC Act.

For an investor day meant to answer "is this company safe to own," a same-day federal subpoena about its products' safety is — charitably — unhelpful optics. Amodei, per Reuters attendees, reiterated the safety mission but notably steered clear of the "human extinction" language that fills 80 pages of Anthropic's own draft IPO prospectus. One version of this is discipline: the company has already told regulators it could pose "catastrophic or existential risks", and repeating that to buyers isn't the pitch. Another version: you cannot simultaneously warn that your product might end civilisation and ask the public to value it at $2 trillion. Anthropic would like both. The market will let it have one.

New York Stock Exchange facade with Slack IPO banner

Photo: the New York Stock Exchange, where Anthropic hopes to list before Thanksgiving — though investors now whisper $1.5T instead of $2T. (Ajay Suresh / Wikimedia Commons, CC BY 2.0 — file photo from Slack's IPO, used for IPO context.)

What $1.5 trillion versus $2 trillion actually means

Zoom out. In May, Anthropic raised at $965 billion. A $2 trillion IPO is a doubling in six months — a leap that requires public investors to believe the growth story the private ones already bought. The whisper-number alternative, ~$1.5 trillion, is still a 55% markup on May. That is not a catastrophe; it is the market repricing risk. And the risks are real: Meta and Microsoft — both major Anthropic customers — are expanding their own in-house AI; Claude Haiku 5.5 launched at roughly 75% below its predecessor's price, which is great for adoption and terrible for the "software margins" thesis; OpenAI is doing $50 billion annualised and pushing its IPO to 2027, meaning Anthropic will list into a market already pricing its rival's numbers.


Analysts who met executives this week put Anthropic's end-2026 run-rate anywhere from $74 billion (SemiAnalysis) to $110 billion (the most bullish investor math). The $110 billion figure is not company guidance — the company gave none. It is what you get when people who want the IPO to work do arithmetic. SemiAnalysis's $74 billion, against a $2 trillion ask, is the cold version.

None of this is to say Anthropic is in trouble. A company going from $4.6 billion annual revenue to a $67 billion annualised run-rate in eighteen months is one of the fastest scaling businesses in history. When the company's own history includes its first adjusted profit last quarter — covered here — and the bitcoin-miner data-centre deal that capped its $530 billion compute sprint, the momentum is undeniable. The question is whether the valuation is.

The verdict: the market wants a discount, and it's right to ask

Here is the honest position. Anthropic is executing one of the great revenue ramps in business history. It is also asking public investors to pay a multiple that assumes the ramp never bends — the same week the ramp bent. The 45% print is a triumph of operations and a failure of narrative, and in IPO pricing, narrative is half the asset.

The company did everything right on Wednesday: record quarter, adjusted profit, improving compute economics, a moat-protecting move on distillation, no extinction talk in front of the buyers. And the market's response — immediately — was to shave up to $500 billion off the target. That is not irrational. It is what due diligence looks like when the asset is this big and the risks are this real.

Whether the $2 trillion holds will be decided not by the baseball caps but by the next two quarters. If Q4 re-accelerates — distillation drag included — the haircut gets priced back in. If 45% becomes 35%, the $1.5 trillion whispers become the opening bid. The IPO roadshow, marketing the week of Nov 9, is where we'll find out who blinks first.

Update, Oct. 17: who moved the number

A week after the investor day, outside analysts started pricing the gap. New Constructs put a low-case equity value at $144B on Oct 6 — 93% below the $2T target, per financefeeds.com's Oct 16 analysis of the Reuters reporting. The listing calendar slipped too: initially planned for October, the IPO is now expected after the Nov 3 midterms, per carrier consensus (ChosunBiz relay of Reuters) — timing, not a done deal. Anthropic's executives also walked investors through positive free cash flow this week (webpronews), another plank in the "we can fund the build-out" case. Read the bubble debate in full — and the same week's other OpenAI headline.

Sources: Reuters (Milana Vinn & Krystal Hu, Oct 14) on the investor day; Reuters follow-up (Oct 15) via Business Times/CGTN syndication; The Financial Times via Business Standard; The Wall Street Journal; Bloomberg; The Information (Cory Weinberg, via Dealroom); SemiAnalysis via IndexBox; theinference.org. One factual caveat we verified ourselves: Anthropic has not disclosed which costs its "adjusted" profit measures exclude; all figures are company disclosures to investors via anonymous attendees, not audited filings — the draft prospectus filed in June is still not public.

About the author: Daniel Rao covers AI economics and the frontier labs for TechPulse USA — revenue ramps, compute budgets, and what it costs to build the future.


Comments

Popular posts from this blog

Nous Research Raises $90M With Trump Jr. Firm on Cap Table

Nadella Says Assume Every AI Model Is Already Compromised

Firmus axes $5.5B IPO after 'greedy' A$44B valuation revolt