AI unicorn Anthropic delivered a third-quarter report card that looked impressive on the surface but left the market disappointed.

According to a Thursday report, Anthropic disclosed at a closed-door pre-IPO investor day in San Francisco that its third-quarter revenue reached $16.8 billion, up nearly 50% from $11.5 billion in the second quarter, while also achieving "adjusted profitability" and positive cash flow.

Yet the market reaction was far from celebratory. Research firm P Equity Research stated bluntly: "Anthropic's numbers are good, but not good enough." Investment analyst Geiger Capital was even more direct: "Not good enough. It's over."

The hidden concern behind the numbers lies in the sharp deceleration of growth. According to ZeroHedge calculations, Anthropic's annualized revenue run rate had already reached $65 billion by the end of July, while the monthly growth rate implied by the third-quarter data has plummeted from a previous compound monthly growth rate of approximately 33% to just 3.4% to 6.6%. If this trend continues, the company's year-end annualized revenue would be approximately $89 billion, leaving a gap of about $10 billion from the widely expected year-end target of "over $100 billion."

This means that for Anthropic to meet its own target, fourth-quarter monthly growth must reaccelerate to 10% to 13% — and this is happening against a backdrop of token prices falling to historic lows and intensifying competition from open-source models.

Growth cliff: from 33% monthly growth to single digits

The key to understanding this earnings report lies in reconstructing the true trajectory of monthly growth.

Anthropic habitually calculates its annualized run rate by multiplying the latest single-month revenue by 12. Accordingly, the $65 billion run rate at the end of July implies July monthly revenue of approximately $5.4 billion. Working backward from total third-quarter revenue of $16.8 billion, combined August and September revenue was approximately $11.4 billion, corresponding to a month-over-month growth rate of only about 3.4%.

Even under a more lenient assumption — treating the $65 billion as an instantaneous run rate on July 31 rather than a July average — July revenue would be approximately $5.2 billion, and the implied monthly growth rate for August and September would still be only 6.6%. On this basis, Anthropic's annualized run rate at the end of the third quarter was approximately between $69 billion and $74 billion.

This stands in stark contrast to the growth momentum in the first half of the year. From an annualized run rate of approximately $10.8 billion at the end of 2025 to $65 billion by the end of July, Anthropic achieved a compound monthly growth rate of about 33%. But entering August and September, that figure has dropped to single digits.

"Adjusted profitability": profits full of asterisks

Anthropic's profitability claim will receive the most attention — and deserves the most footnotes.

The report did not disclose exactly which items were "adjusted," and complete GAAP financial data is expected to be disclosed around October 25 alongside the public prospectus (S-1). According to a previously reported confidential draft prospectus, Anthropic recorded a net loss of approximately $42 billion in 2025, of which about $34 billion was accounting charges related to financing, with operating losses exceeding $8 billion. Future ten-year computing power and infrastructure commitments amount to at least $518 billion, of which approximately 80% are irrevocable commitments.

Moreover, the revenue methodology itself is also disputed. According to explanations from Goldman Sachs' TMT team, Anthropic adopts a methodology close to "gross revenue" — counting all customer spending generated through cloud partners, then recording partner revenue shares as costs — which makes its top-line figures look better compared to OpenAI's net revenue methodology, and also makes the calculation of "adjusted" profit margins highly dependent on specific accounting treatments.

With the largest computing bills not yet due (commitments with Microsoft won't even begin to be counted until November) and the IPO window about to open, a single quarter of adjusted profitability is still far from forming a credible earnings track record.

Token prices hit historic lows, volume-price divergence becomes the core contradiction

The structural root of the growth slowdown points to the continued decline in token prices.

According to Ornn Anthropic Token Price Index data, Anthropic's token prices have fallen to historic lows. Meanwhile, the latest token usage data from Ramp AI shows that although AI usage continues to grow rapidly (consistent with Jevons paradox logic), actual AI spending is at best flat or even shrinking — the fundamental reason being that the expanding market share of open-source models is压低 overall token prices.

This volume-price divergence constitutes Anthropic's most core current business model contradiction: token prices are falling faster than the company's ability to convert higher usage into revenue growth. If the growth slowdown from August to September occurred when prices were still relatively firm, then the fourth-quarter reacceleration must rely on achieving significantly higher usage at half the price — a fundamentally different business logic.

Meta, Google, and Chinese open-weight models are flooding into the frontier market at deeply discounted prices, further compressing Anthropic's pricing space.

For the AI trade: good, but not good enough

Whether this earnings report can support the entire AI investment narrative depends on the yardstick being used.

According to the Goldman Sachs trading desk citing Brad Gerstner's framework, top AI labs need to increase their combined annualized run rate from approximately $100 billion in July to $180 billion to $200 billion by year-end to maintain the integrity of the AI trade. Based on third-quarter trends, Anthropic will be at approximately $89 billion by year-end, plus OpenAI's own year-end target of approximately $70 billion, the two combined would be approximately $159 billion — still below Gerstner's threshold, and this is the result of mixing a gross revenue methodology with a net revenue methodology.

On the valuation front, based on a $2 trillion IPO valuation, Anthropic corresponds to approximately 27 times the implied end-of-September run rate and approximately 22 times the December trend value, already narrower than the approximately 47 times historical revenue multiple calculated by Breakingviews two weeks ago. This is the bull case that investors heard at the investor day.

However, the market voted with its feet the day before the earnings release: the S&P 500 fell 1.11%, the Nasdaq 100 dropped 2.02%, sentiment in AI-related sectors experienced a "stair-step decline," emerging cloud computing stocks plunged more than 12% in a single day, CoreWeave shares fell over 15%, and the five-year default probability implied by its CDS market has risen to 63%.

Quarterly revenue of $16.8 billion, along with adjusted profitability and positive cash flow, is enough to keep the November IPO plan on track. But the reality of growth plunging from 33% monthly to single digits, and the fourth-quarter reacceleration needed to achieve the $100 billion year-end target, will be the real questions the market needs answered after the public S-1 disclosure.