Anthropic's Q3 dollar gain was 21% to 24% smaller than Q2's, and a $2 trillion valuation is 27x a $74bn run rate but 18x $110bn

The dollar adds fell first

Start with dollars, because percentage growth falls mechanically as the base gets bigger. On PYMNTS.com's quarterly figures of $4.8bn, $11.6bn and $16.8bn, Anthropic added $6.8bn of quarterly revenue between the first and second quarters and $5.2bn between the second and third ^[2][32]. The Economic Times, citing an investor who attended the meeting, puts the first quarter at $4.7bn ^[3]. That would make the earlier add $6.9bn, and the direction is the same ^[32]. The Financial Times, as relayed by simonwillison.net, puts Q2 at $11.5bn rather than $11.6bn ^[31]. On those figures the adds are $6.7bn and $5.3bn, and the Q3 add is about 21% smaller ^[40]. On PYMNTS.com's base, the Q3 add is 24% smaller than the Q2 add ^[32]. Depending on which Q2 figure is used, the Q3 add is 21% to 24% smaller, and the direction holds on every source's figures ^[40][32].

The growth rates say it in a form that is easier to wave away: exponentialview.co puts growth at 147% from the first to the second quarter and 45% from the second to the third ^[4].

Why the add shrank is not settled, and the candidate causes point in different directions. The Decoder credits the Q2 surge to demand for Claude Code and calls a summer slowdown a likely factor in Q3, since people work less in those months and burn fewer tokens ^[5]. It adds that OpenAI launched a strong competitor, Astra, in September and that Anthropic only countered it successfully in late September, with Opus 5.5 ^[6]. SemiAnalysis analyst Joey Brookhart said investors had lowered their forecasts over the past three months as major customers, Meta among them, increasingly relied on their own AI tools, according to the Financial Times report that PYMNTS.com relays ^[8]. exponentialview.co asks whether Q1 and Q2 pulled revenue forward from later in the year and writes: "It's hard to imagine a world where only one of those reasons is the whole story; more likely each has played a role." ^[7]

Each cause implies a different Q4. A seasonal dip reverses by itself, a pull-forward leaves Q4 short, and competition plus in-house tools keep the slope falling. The Q3 figures alone cannot separate them.

One cause was a choice. The FT reported, as summarized by simonwillison.net, that Anthropic told investors distillation accounted for 5% to 7% of revenue and that it cut the practice back "materially" in Q3 ^[27]. Anthropic also told investors the revenue will likely drop and weigh on the current quarter's numbers, The Decoder reported ^[28].

One quarter, three run-rates

Annualize the quarter at face value and $16.8bn times four is $67.2bn ^[33]. That understates the September exit if revenue grew through the quarter. exponentialview.co puts the Q3 exit at $76bn "based on the quarterly growth rate" ^[9], and SemiAnalysis estimated roughly $74bn, per the FT ^[10]. Anthropic did not discuss its September annualized rate or a full-year outlook, The Decoder reported ^[12]. Both figures are therefore outside estimates.

The $76bn can be reproduced. If revenue compounds evenly each month at the rate that produces 45% a quarter, about 13% a month, the third month annualizes to $75.7bn ^[33]. It is a model of steady growth, and one disclosed month would replace it.

The year-end figure works the same way. The Wall Street Journal reported that investors at the meeting projected the company could reach $110bn of annualized revenue by year-end ^[11]. One more quarter at Q3's 45% pace takes the same model to $109.7bn ^[34]. The $110bn projection is therefore a bet that the slope stops falling, after a quarter in which it fell from 147% to 45% ^[4]. Every ten points taken off Q4 growth removes $1.7bn from Q4 revenue and $7.6bn from the December exit rate ^[34].

The labels are not clean. PYMNTS.com's summary line says investors have cut year-end forecasts to about $74bn ^[13], while the body of the same article attributes $74bn to SemiAnalysis's estimate for the third quarter ^[10]. Because exponentialview.co's Q3 exit of $76bn is already above $74bn ^[9], we read $74bn as a Q3 estimate.

On the full year, exponentialview.co calculates that $60bn needs $26.9bn in Q4, or 60% sequential growth, and that Q4 growing as fast as Q3 gives $57.5bn ^[14]. It frames the next question this way: "The question becomes: at what rate will that growth taper over the next 2 to 3 years? The slower the taper, the better the business." ^[15]

What the margin shows

In 2025 Anthropic spent $7.33bn on computing and infrastructure against nearly $4.6bn of revenue, according to the Reuters report PYMNTS.com cites ^[18]. That is $0.63 of revenue per compute dollar ^[35]. For Q3 the FT-reported figure is $1.70, alongside an 11% adjusted operating margin ^[16]. At that ratio the quarter's compute bill was about $9.9bn, more than the whole of 2025, and about $39.5bn annualized ^[36].

exponentialview.co reports an 85% adjusted gross margin and 44% of revenue going into compute for research and development ^[19]. A $1.70 ratio means total compute is 58.8% of revenue. Take away the 44% and 14.8% is left, close to the 15% that an 85% gross margin leaves for cost of revenue ^[37]. If the three figures share a definition, which we cannot confirm, about three-quarters of Anthropic's compute is research, roughly $7.4bn in Q3 ^[37], against $5.2bn to $5.3bn of revenue added in the quarter ^[32][40].

Inference looks cheap and profitable on that reading. The expensive line is the one the CFO has described as having a floor. Krishna Rao has said "There is a level of compute for model development that we will not go below", and exponentialview.co notes the compute mix has shifted toward training and R&D ^[20]. With a floor under that line, the margin improves from here only if revenue grows over a base that does not shrink.

The 11% needs two caveats. The Information reported it as an adjusted operating margin for the year through Q3, and the adjustment leaves out costs such as stock-based compensation ^[17]. PYMNTS.com presents it as the quarter's margin ^[16]. The sources do not agree on whether it covers one quarter or nine months. In Q3, then, Anthropic chose to cut back a revenue line worth 5% to 7% of sales ^[27] and to keep 44% of revenue going into R&D compute ^[19], and the margin is what remained.

A decade of commitments against a quarterly slope

Anthropic has committed to at least $518bn over a decade for cloud services, computing and infrastructure, the Economic Times reported ^[21]. Straight-lined, that is $51.8bn a year, 31% above the $39.5bn at which Q3 compute annualizes ^[38][36]. Holding the $1.70 ratio against $51.8bn takes $88.1bn of revenue ^[38]. That sits between SemiAnalysis's $74bn ^[10] and the investors' $110bn ^[11]. In the steady-growth model, Q4 growth of about 16% puts the December exit at that $88bn ^[38].

The test is gentle by construction, because it assumes payments are spread evenly. A back-loaded schedule would raise the bar and a front-loaded one would lower it. The pressure comes from how fast growth tapers over the 2 to 3 years that exponentialview.co asks about ^[15], measured against a bill that runs ten.

Lenders are in the same position. Days before the meeting, bankers were syndicating $60bn of AI chip financing for Anthropic and looking to fund another $50bn for OpenAI, bloomberg.com reported ^[25]. Semafor reported yields touching 15% on the bonds of one AI cloud provider ^[26].

How the slope becomes a price

Anthropic is seeking a valuation above $2 trillion in a listing expected after the Nov. 3 midterm elections, per Reuters as relayed by PYMNTS.com ^[22]. Its private valuation jumped to $965bn in May, and The Decoder reports that several investors now expect its bankers to lower expectations to possibly under $1.5 trillion ^[23]. Against the estimates above, $2 trillion is 27.0 times $74bn of annualized revenue and 18.2 times $110bn, while $1.5 trillion is 20.3 times $74bn ^[39]. The 8.8-turn spread between 27.0 and 18.2 is wider than the 6.8-turn drop from 27.0 to 20.3 that a cut to $1.5 trillion would bring ^[39]. (One figure is a current estimate and the other a year-end projection, so part of that spread is only timing; the part that is not timing is the slope.)

The sell-off ran well beyond one company. After the news, Kospi fell by as much as 9.3% and the Nikkei 225 by as much as 5.4%, bloomberg.com reported ^[24]. Semafor quoted an analyst, unnamed in its piece, saying traders were "increasingly concerned about the side effects of sustaining that growth" ^[26].

There are three ways the next quarter can go. The slope holds near 45%, the December exit lands near $110bn, and Q3 looks like one decelerating print ^[34]. Or Q2 was the outlier and Q3 is the baseline, with growth sliding toward the mid-teens; at about 16% the December exit is $88bn, where a straight-lined commitment is just covered at $1.70 ^[38]. Or price pressure and customers' in-house tools pull the slope below that.

We think the second or third is likelier than the first. Of the explanations on the table, only the summer slowdown reverses by itself ^[5], and Anthropic told investors that a revenue line worth 5% to 7% of sales is being cut back ^[27][28]. The counter-thesis has force. The investors in the room already projected $110bn, which is only the constant-slope case ^[11][34]. The Decoder lists high inflation and rising borrowing costs among the headwinds ^[29], so part of the sell-off may be the macro backdrop with Anthropic as the trigger. And Q2, not Q3, may be the quarter that needs explaining ^[7].

We would be wrong if Anthropic disclosed a September or December run-rate above $76bn, or if Q4 growth came in near 45%. The first place to look is the S-1: exponentialview.co says Anthropic will have an opportunity to signal its trajectory when it goes public ^[30].