跳到主内容
@wquguru
精选88SaaStr 博客(RSS)战略与拆解

Anthropic人效数据对B2B创始人的启示:从管人到管算力

Anthropic Will Get to $100B in Revenue with ~7% of the Headcount Google and Meta Needed

原文
发到 X
推荐理由

直接拆解了AI时代B2B公司的单位经济模型变化,给出了具体的财务指标替代方案(看毛利而非人均营收)和管理动作(算力月度审查),适合所有正在构建或转型AI产品的创始人参考。

Jon Ma posted Brad Gerstner’s Anthropic thesis from TomorrowX this morning. Google needed about 100,000 people to get to $100B in revenue. Anthropic will get there with about 5,000. The money Google spent on people, Anthropic spends on compute.

Jon Ma 今早转发了 Brad Gerstner 来自 TomorrowX 的 Anthropic 论点。Google 需要约 10 万人才能实现 1000 亿美元的收入。Anthropic 只需约 5000 人即可达到这一目标。Google 在人力上花费的资金,Anthropic 将其投入到了算力中。

Brad $ANT Thesis:

Brad $ANT 论点:

– Google had 105k people to get to $100B revenue

– Google 拥有 10.5 万名员工以实现 1000 亿美元收入

– Anthropic needs 5k people to get to $100B revenue

– Anthropic 需要 5000 名员工以实现 1000 亿美元收入

– the 100K people Google had, Anthropic can allocate the human cost to compute

– Google 拥有的 10 万人的成本,Anthropic 可以将其转化为算力成本

cc @altcap @AltimeterCap @TomorrowXSummit

— Jon Ma (@jonbma) November 17, 2026

— Jon Ma (@jonbma) 2026年11月17日

The exact figures in the post are a bit off. Anthropic’s S-1 supports the thesis anyway, and Meta’s numbers back it up too.

帖子中的具体数字略有出入。但 Anthropic 的 S-1 文件支持了这一论点,Meta 的数据也佐证了这一点。

Google Had 80,110 Employees at $100B. Meta Had 71,970.

Google 在达到 1000 亿美元收入时有 80,110 名员工。Meta 有 71,970 名。

Alphabet did $110.9B in revenue in 2017, the first year it passed $100B. It ended that year with 80,110 employees.

Alphabet 在 2017 年实现了 1109 亿美元的营收,这是其首次突破 1000 亿美元大关。该年底其员工总数为 80,110 人。

Meta passed $100B four years later. It did $117.9B in 2021 and ended the year with 71,970 employees.

四年后 Meta 也突破了 1000 亿美元。2021 年其营收为 1179 亿美元,年底员工数为 71,970 人。

Those were two of the most efficient large companies ever built, and each still needed 70,000 to 80,000 people to get to $100B.

这两家都是历史上最高效的大型企业之一,且各自仍需 7 万至 8 万人才能实现 1000 亿美元的收入。

Anthropic Has 5,300 Employees and an $81B Run-Rate

Anthropic 拥有 5,300 名员工,年化收入达 810 亿美元

Anthropic’s S-1/A shows about 5,300 employees as of September 30, 2026, and an $81B run-rate as of October 31. It added $10B of run-rate in October alone.

Anthropic 的 S-1/A 文件显示,截至 2026 年 9 月 30 日约有 5,300 名员工,截至 10 月 31 日的年化收入为 810 亿美元。仅在 10 月份,其年化收入就增加了 100 亿美元。

At that pace, Anthropic passes a $100B run-rate around the end of the year with fewer than 6,000 people. That’s about 7% of Google’s headcount and about 7% of Meta’s at the same milestone.

按照这个速度,Anthropic 将在年底前以不到 6000 人的规模实现 1000 亿美元的年化收入。这约占 Google 同期员工总数的 7%,也约占 Meta 在同一里程碑时的员工总数的 7%。

$15M in Revenue per Employee vs. $1.4M at Google and $1.6M at Meta

每名员工创收 1500 万美元,而 Google 为 140 万美元,Meta 为 160 万美元

Google did about $1.4M in revenue per employee in 2017. Meta did about $1.6M in 2021.

2017 年 Google 每名员工的营收约为 140 万美元。2021 年 Meta 约为 160 万美元。

Anthropic does about $15M per employee on a run-rate basis, roughly 10x either company at the same scale. On trailing twelve months, the S-1 says $6.6M per employee, up from $1.4M a year earlier. Over that year, revenue grew about 1,200% and headcount grew about 170%.

Anthropic 按年化收入计算,每名员工创收约 1500 万美元,在同等规模下约为这两家公司的 10 倍。根据过去十二个月的数据,S-1 文件显示每名员工创收 660 万美元,高于一年前的 140 万美元。在这一年中,营收增长了约 1200%,员工人数增长了约 170%。

Q3 2026: Compute Was 58% of Revenue, People Were 10%

2026 年第三季度:算力占收入的 58%,人力占 10%

Anthropic’s Q3 2026 numbers:

Anthropic 2026 年第三季度的数据:

  • Revenue: $16.96B
  • Compute and infrastructure: $9.86B, or 58% of revenue, up from 48% in Q2
  • Personnel: $1.71B, about 10% of revenue
  • 营收:169.6 亿美元
  • 算力和基础设施:98.6 亿美元,占收入的 58%,高于第二季度的 48%
  • 人员费用:17.1 亿美元,约占收入的 10%

Anthropic spent almost 6x as much on compute as on people. Year over year in Q3, personnel costs grew about 3.5x, from $489M to $1.71B. Quarterly revenue grew more than 12x.

Anthropic 在算力上的支出几乎是人力支出的 6 倍。同比去年第三季度,人员成本增长了约 3.5 倍,从 4.89 亿美元增至 17.1 亿美元。季度营收增长了 12 倍以上。

Google Had a 23.5% Operating Margin and Meta Had 40%. Anthropic Lost $567M in Q3.

谷歌的营业利润率为23.5%,Meta为40%。Anthropic在第三季度亏损5.67亿美元。

Google earned $26.1B in operating income in 2017. Meta earned $46.8B in 2021. Anthropic posted a $567M operating loss in Q3 2026.

谷歌在2017年的营业利润为261亿美元。Meta在2021年的营业利润为468亿美元。Anthropic在2026年第三季度录得5.67亿美元的营业亏损。

Anthropic didn’t eliminate the cost of 75,000 people. That money now goes to Google, Amazon, Broadcom and the other compute providers. On Q3’s numbers, it costs more than the people did. That could change as inference gets cheaper per token and pricing matures. For now, the small headcount is driving growth, not profit.

Anthropic并没有裁掉7.5万人的成本。这笔钱现在流向了谷歌、亚马逊、博通和其他计算服务提供商。根据第三季度的数据,其成本比人员成本更高。随着每个token的推理成本降低和定价趋于成熟,这种情况可能会改变。目前,精简的人员编制正在推动增长,而非利润。

What Changes for B2B Founders

对B2B创始人的影响

The operating plan used to be built around headcount. Revenue per rep, ramp time and quota capacity set how fast you could grow, and the hiring plan sat in the middle of the board deck.

运营计划过去是以人员数量为核心构建的。每位销售代表的收入、上手时间和配额容量决定了你的增长速度,而招聘计划则位于董事会演示文稿的中心位置。

Now the plan gets built around how much work your agents can do and what that work costs. Three things follow:

现在,计划的构建围绕你的智能体能完成多少工作以及这些工作的成本展开。随之而来的是三点变化:

  • Revenue per employee tells you less than it used to. Anthropic does $15M per employee and still lost money in Q3. Gross margin after inference and cost per unit of output tell you more.
  • Compute spend grows with usage. A hiring freeze does nothing to it, so it needs the same monthly review headcount used to get.
  • Customers are moving payroll money into AI spend. B2B AI companies that charge 2x what pre-AI competitors charge are being paid out of salary budgets.
  • 人均收入所反映的信息不如以往重要。Anthropic的人均收入为1500万美元,但在第三季度仍然亏损。推理后的毛利率和单位产出的成本能提供更准确的信息。
  • 计算支出随使用量增长。招聘冻结对此毫无作用,因此它需要像以前的人员数量那样进行月度审查。
  • 客户正将薪资预算转向AI支出。向客户收取比AI前竞争对手高出两倍费用的B2B AI公司,其收入来自薪资预算。

SaaStr Runs on 3 Humans and 20+ Agents

SaaStr由3名人类员工和20多个智能体驱动

We see the same thing at SaaStr on a much smaller scale. We run the company with 3 humans and 20+ AI agents in production, and our fastest-growing cost is the AI tools those agents run on.

我们在SaaStr以更小的规模看到了同样的现象。我们仅用3名人类员工和20多个生产环境中的AI智能体来运营公司,而我们增长最快的成本是这些智能体所运行的AI工具。

Google needed 80,000 people to get to $100B, and Meta needed 72,000. Anthropic will get there with about 5,300 people, and in Q3 it spent $9.86B on compute.

谷歌需要8万人才能达到1000亿美元的收入,Meta需要7.2万人。Anthropic只需约5300人就能达到这一目标,而在第三季度,它在计算方面的支出为98.6亿美元。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件