Anthropic’s $16.8 Billion Quarter: The Arithmetic Behind a $110 Billion Run Rate

TL;DR

Anthropic told prospective investors its third-quarter revenue reached $16.8bn, up 45% sequentially from $11.6bn, and that it was profitable on an adjusted basis and cash-flow positive (Source : Financial Times — Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn).
The same briefing produced the first hard unit-economics figure from a frontier lab: $1.70 of revenue per $1 of compute, against $0.63 in 2025 (Source : Financial Times).
Analysts at the meeting estimate a $110bn annual run rate by year-end — arithmetic that requires a Q4 of roughly $27.5bn, or 64% sequential growth, faster than the 45% just reported (Source : Reuters — Anthropic’s Q3 revenue surges 45% to nearly $17 billion).
A listing is expected in November, after the US midterms, at a valuation approaching $2tn, against at least $518bn of decade-long cloud and compute commitments (Source : Reuters).

Introduction

Six days after OpenAI’s investor update opened a $20bn definitional hole, Anthropic’s Wednesday investor day at its San Francisco headquarters ran the other way: numbers moving up, and — for the first time — numbers that describe economics rather than scale. CEO Dario Amodei and CFO Krishna Rao briefed attendees two to three weeks before a public debut that Reuters reports could slip past the November midterms, and which would make Anthropic the first frontier lab to test public-market appetite for the agent economy.

Three quarters of hyper-growth, then a slowdown

The revenue series is the cleanest part of the disclosure: $4.7bn in Q1, $11.6bn in Q2 (+147% sequentially), $16.8bn in Q3 (+45%) (Source : Business Insider via Mint — Anthropic says it is profitable). The deceleration is arithmetic as much as it is demand: a 147% jump from a $4.7bn base adds less absolute revenue than a 45% jump from $11.6bn. What matters for the IPO is that the absolute add — $5.2bn of new quarterly revenue — is the largest single-quarter increase the company has reported.

For scale, Anthropic’s full-year 2025 revenue grew twelvefold to roughly $4.6bn while it posted an operating loss above $8bn, excluding write-downs tied to earlier fundraising (Source : Reuters via Mint). It is now growing at a slower percentage rate off a base nearly four times larger.

Investors who attended left optimistic, but PitchBook’s Harrison Rolfes framed the trend less flatteringly: “They’re still growing, but not growing as fast. So they’re reaching this inflection point.” (Source : Business Insider via Mint)

The number that actually matters: $1.70 per compute dollar

Buried under the headline revenue is the metric that will decide the multiple. Anthropic told investors it generated $1.70 of revenue for every $1 spent on computing power in Q3, up from $0.63 in 2025 — a 2.7× improvement in roughly a year (Source : Financial Times).

That ratio is a gross margin in disguise. Below 1:1, every unit of inference sold costs more in compute than it earns; above 1:1, each additional token contributes gross profit. Anthropic crossed the line and kept going, which is what makes an “adjusted profitable” claim credible rather than cosmetic. The likely driver is mix: long-horizon agentic work — Claude Code above all — bills far more tokens per task than chat, while prompt caching and batch-eligible workloads cut the compute consumed per billable token. Efficiency, not price, is the lever.

$110bn: a forecast, not an extrapolation

A $16.8bn quarter annualizes to about $67bn. So the analysts’ year-end estimate of $110bn is not a snapshot of the current pace — it implies fourth-quarter revenue of roughly $27.5bn, a 64% sequential increase that requires growth to re-accelerate from 45%. Two things could deliver it: a year-end enterprise contracting cycle (Anthropic’s book is enterprise-heavy), or an expanding accounting base if gross partner sales rise as a share of the mix.

The compare-and-contrast with OpenAI is instructive. OpenAI told investors its annualized revenue was approaching $50bn at the end of September, against a $70bn year-end target — roughly 40% growth in one quarter, off a base three times larger than Anthropic’s implied Q4 (Source : Bloomberg via Yahoo Finance — OpenAI projects $70 billion annualized revenue by end of 2026). Both numbers are paces, not tallies.

The compute ratio then disciplines the pitch: at $1.70 per compute dollar, a $110bn run rate implies roughly $65bn of annual compute spend, against the ~$52bn/year average implied by Anthropic’s $518bn decade-long commitment. Either the ratio keeps improving or the company draws faster on commitments it has already signed.

Finally, the multiple. $2tn on a $110bn run rate is about 18×; $2tn on the reported quarter annualized is about 30× — a 1.6× spread on identical news, decided entirely by which figure the underwriters put on the cover.

What is still undefined

Three gaps remain. Anthropic has not said whether $16.8bn is gross or net revenue — it has historically counted gross sales through cloud partners such as AWS and Google Cloud, including amounts it never receives. It has not disclosed how “adjusted profit” is calculated or which costs are excluded. And it has not published gross margin, which is why the $1.70 ratio is presented per compute dollar rather than as a percentage. If the denominator excludes compute bought through partner channels, the ratio flatters the economics further.

The audited prospectus will settle all three. Until then, the deck is a narrative and the ratio is a claim.

FAQ

Is Anthropic actually profitable? On an adjusted basis for Q3, and cash-flow positive, per the investor day. Neither is a GAAP operating profit, and the adjustment methodology has not been published.

What is an “annual revenue run rate”? An extrapolation: a recent slice of sales multiplied out to a year as if conditions persist. It is not revenue booked, not audited, and highly sensitive to which month or quarter you pick.

Why does $1.70 per compute dollar matter more than $16.8bn? Revenue tells you how big the business is; the ratio tells you whether the business works. It is the first public signal that agentic inference can be sold above its compute cost.

What could derail the November listing? Timing after the midterms, the growth deceleration PitchBook flagged, unresolved revenue definitions, and the $518bn of commitments that need funding. Anthropic’s own prospectus warns of “catastrophic or existential risks to humanity” from advanced AI — a risk factor no previous mega-IPO has carried.

Anthropic's $16.8 Billion Quarter: The Arithmetic Behind a $110 Billion Run Rate