Anthropic昨天与投资者分享了高层次的Q3金融，报告NYT的安德鲁·罗斯·索金(Andrew Ross Sorkin)，劳伦·赫希(Lauren Hirsch)和斯里·穆皮迪(Sri Muppidi).
据报告，第三季度的收入为168亿美元。
乍一看，这个数字似乎令人吃惊。 而从一种意义上说，它是。 也是历史上增长最快的公司
但是，将趋势细分为隐含的月增长，而画面看起来也有所不同。
具体而言，Anthropic似乎从7月到9月几乎没有增长。
此外，Anthropic在Q3的连续收入增长，以绝对美元计算，比其在Q2的(gobsmaking)增长要小。
因此，由于几个可能的原因，Anthropic的增长已经有意义的放缓。
这很可能是像CoreWeave这样的“无名”供应商的库存正在崩溃的另一个原因。 CoreWeave等人借了数百亿美元，其假设是，多年来对AI计算的需求将继续以远远超过共识预期的速度增长。
现在，不仅利率飞涨——提高这一债务的承担成本——大赦国际服务的近期收入增长可能正在趋于平缓。
分析Anthropic的成长。.....
Anthropic今年收入增长史无前例：
问题1:48亿美元
问题2:116亿美元(+6.8B美元和+141%，顺序)
问题3:168亿美元(+5.2B和+45%，顺序)
这些数字再次令人吃惊。
但是AI的估价和期望也一样。
注意Anthropic在Q2中增加的收入比在Q3中多。
也请注意，Anthropic IPO 的估价最为重要，很可能是AI泡沫的未来轨迹
关键的问题是，“这些数字对Anthropic从这里增长的速度有何影响？”
这些数字表明，市场至少应暂时降低预期。
如果我们将上面的季度数字细分为每月估计数，并增加一个报告的数据点——即Anthropic在7月份有650亿美元的年化收入运行率——我们看到Anthropic的增长比季度数字显示的更平缓。
以下是2026年Anthropic月收入的背信封估计数。 这些收入是根据报告的季度收入数字以及7月的650亿美元年收入(相当于7月的54亿美元收入)计算的。
甚至季度数字也表明，Anthropic的月平均增长从7亿6千万的Q2减慢到每月5.8亿的Q3。
但“7月650亿美元的年化收入”的数据表明，减速更为明显。
如果Anthropic在7月的收入为54亿美元，在Q3中的收入为168亿美元，那么很可能只在8月和9月略有增长。
根据Deal Book，在昨天与投资者的会晤中，Anthropic没有根据9月份的收入分享年化收入号。 这向这位分析家表明，Anthropic认为9月份的年化收入数字不会显示比7月份增长得惊人。
事实上，当我们推断每月数字时，Anthropic似乎在8月和9月其增长轨迹发生了急剧变化。
这是什么意思？
这意味着，该部门领先公司的增长在过去两个月中似乎急剧放缓。
减速的部分原因是季节性。 八月，至少，往往是一个缓慢的月份。 但是在产品周期的早期，9月的增长应该有所回升。
这也可能意味着Anthropic根本无法跟上需求的增长。 如果这是减速的原因，那就是一个高质量的问题，它不会对AI总体支出增长轨迹(或泡沫)产生长期影响。
但这也可能意味着AI支出的近期增长率正在缓和——要么是因为需求增长在缓和，要么是因为公司正在使用中国更便宜的模型来满足部分AI需求。 (用户数据表明后者实际上正在发生).
如果支出增长放缓是原因，我们应当期望投资者对所有事物的热情以及支持增长的巨大基础设施支出也开始缓和。
当供应开始赶上需求时，所有泡沫的终结就会到来。
Anthropic明显放缓
另见：AI泡泡什么时候爆炸？为什么？
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Anthropic shared high-level Q3 financials with investors yesterday, report the NYT’s Andrew Ross Sorkin, Lauren Hirsch, and Sri Muppidi.
Revenue for Q3 was reportedly $16.8 billion.
At first glance, this number seems astonishing. And, in one sense, it is. It’s another jaw-dropping sequential gain for what is likely the fastest-growing company in history.
But, break the trend down into implied monthly growth, and the picture looks different.
Specifically, it seems Anthropic barely grew from July to September.
Anthropic’s sequential revenue growth in Q3, moreover, was smaller in absolute dollars than its (gobsmacking) growth in Q2.
So, for several possible reasons, Anthropic’s growth has meaningfully slowed.
This is likely another reason why the stocks of “neocloud” providers like CoreWeave are collapsing. CoreWeave, et al, have borrowed tens of billions of dollars on the assumption that demand for AI compute will continue to grow at rates way beyond consensus expectations for years.
Now, not only are interest rates soaring — raising the carrying costs of this debt — the near-term revenue growth for AI services may be flattening.
Analyzing Anthropic’s growth…
Anthropic’s reported revenue growth this year has been unprecedented:
Q1: $4.8 billion
Q2: $11.6 billion (+$6.8B and +141%, sequentially)
Q3: $16.8 billion (+$5.2B and +45%, sequentially)
Again, these numbers are astonishing.
But so are AI valuations and expectations.
Note that Anthropic added more revenue in Q2 than it did in Q3.
Note, too, that what matters most for Anthropic’s IPO valuation — and, likely, for the future trajectory of the AI bubble — is not what has happened but what will happen.
The key question is, “What do these numbers suggest about how fast Anthropic will grow from here?”
These figures suggest that the market should temper its expectations, at least temporarily.
If we break down the quarterly figures above into monthly estimates and add one more reported data point — that Anthropic had a $65 billion annualized revenue run-rate in July — we see that Anthropic’s growth has flattened even more than the quarterly figures suggest.
Below are back-of-the-envelope estimates for Anthropic’s monthly revenue in 2026. They are based on the reported quarterly revenue figures as well as the $65 billion annualized revenue in July (which equates to ~$5.4 billion of July revenue).
Even the quarterly numbers suggest that Anthropic’s average monthly growth slowed from ~$760 million in Q2 to (“only”) $580 million a month in Q3.
But the “$65 billion annualized revenue in July” data point suggests the slowdown is even sharper.
If Anthropic did $5.4 billion in revenue in July and $16.8 billion in Q3, it likely only grew modestly in August and September.
According to DealBook, at its meeting with investors yesterday, Anthropic did not share an annualized revenue number based on September’s revenue. This suggests — to this analyst, anyway — that Anthropic did not think the September annualized revenue figure would show impressive-enough growth versus July.
Indeed, when we extrapolate the monthly figures, it seems Anthropic saw a sharp change in its growth trajectory in August and September.
What does this mean?
It means that the growth of the leading company in the sector appears to have slowed sharply in the past two months.
The slowdown might at least in part be the result of seasonality. August, at least, tends to be a slow month. But it’s early in a product cycle to see such sharp seasonality. And September’s growth should have sped back up.
It might also mean that Anthropic is simply not able to keep up with the growth of demand. If this is the reason for the slowdown, it’s a high-quality problem, and it won’t have a long-term impact on the overall AI spending growth trajectory (or bubble).
But it may also mean that the near-term growth rate of AI spending is moderating — either because demand growth is moderating or because companies are using cheaper models from China for some of their AI needs. (Usage data suggests the latter is in fact happening).
If a slowdown in spending growth is the explanation, we should expect the wild investor enthusiasm for all things AI — and the colossal infrastructure spending growth supporting it — to also begin to moderate.
The end of all bubbles comes when supply begins to catch up with demand.
Anthropic’s apparent slowdown suggests we may be closer to reaching that moment.
See Also: When will the AI Bubble burst? And why?
Auto-fetched and translated by AI Atlas. All rights belong to the original author.