💡 一句话总结：Claude 开发商 Anthropic 于 10 月 26 日公开递交 S-1，拟以代码 ANT 登陆纳斯达克——半年营收 163 亿美元、增速 1300%，账面净亏 807 亿（大头是会计计提而非现金），传闻估值 1.5-2 万亿美元，有望成为史上最大 IPO 之一；计划 11 月中挂牌。
看 AI 行业新闻的人，这半年多少有个预感：Anthropic 迟早要上市的。6 月它确认了机密递交，此后估值传闻一路从"万亿美元俱乐部"涨到"2 万亿、超越 SpaceX"。10 月 26 日，靴子落地——S-1 正式公开，代码就一个词：ANT。
这份 300 多页的文件是外界第一次能逐页看清这家公司的账本。读下来的感受，用 Fortune 的形容挺贴切：surreal——一半是"奇迹时代"的宣言，一半是"生存风险"的警示，中间夹着一份增长凶猛、亏损惊人的财务报表。
想核对原始数字？
账本上最扎眼的两组数字
第一组是营收：2023 年 4100 万 → 2024 年 3.86 亿 → 2025 年 45.85 亿 → 2026 上半年 162.86 亿美元。半年干了去年全年 3.5 倍的量，同比增速 1300%。这个增速放在整个科技 IPO 史里都找不出几个对手——作为对比，Anthropic 是一家 2021 年才成立的公司。
客户结构也在往好的方向走：Q3 约 79% 的营收来自前五大客户以外的客户。对一家常被质疑"靠几个大客户撑营收"的 AI 公司来说，这是招股书里少见的加分项。
第二组是亏损：上半年净亏损 807 亿美元。先别急着倒吸凉气，这里有个绝大多数标题不会告诉你的关键澄清（The Verge 等多家媒体核对后一致）：这 807 亿的大头是公司发行的融资票据带来的会计计提，不是真金白银烧掉的现金。也就是说，"账面亏损"和"经营烧钱"是两回事——当然，计提也不是没有代价，它对应着未来真实的偿付义务或股份摊薄，只是不在今天发生。
真要看家底：截至披露，Anthropic 累计融资 1411 亿美元，账上现金及等价物 1317 亿美元。这大概是商业史上上市前最厚的钱包之一。
算力账单：真正的悬念所在
比亏损更值得盯的是算力承诺。HN 上的高赞分析把 S-1 里的多年期承诺逐项加了一遍：Google 1140 亿 + AWS 1100 亿 + Broadcom 1610 亿 + SpaceX 1200 亿，合计约 5050 亿美元量级。S-1 还提示，未来数年算力获取的累计花费可能高达 7500 亿美元。
这是什么概念？Anthropic 一家公司对算力的多年期承诺，已经超过很多国家的年度财政预算。这也是为什么它要上市——一级市场的钱快喂不动这种量级的开支了，IPO 锁定的是公开市场的长期融资通道。
顺带一提，这份文件里连"怪细节"都很有信息量（The Verge 专门整理了五个）：CEO Dario Amodei 去年总薪酬近 1800 万美元（大头是股票期权，在大厂里算低的），妹妹兼总裁 Daniela 1640 万；公司为 Dario 花了 275 万美元个人安保费、为 Daniela 花了 129 万住宅安保费；Dario 个人使用公司飞机的开销是 97,055 美元。以及最"Anthropic"的一条——风险因素章节正经写入了"有意识的 AI"和"灾难性风险"。一家公司在向公众卖股票的文件里提醒人类注意生存风险，这确实前所未见。
估值：从 2 万亿到 1.5 万亿的下调
市场传的目标估值一度高达 2 万亿美元——超过 SpaceX 的 1.77 万亿，直逼英伟达（5 万亿+）。但 Reuters 报道了一个耐人寻味的细节：在 Anthropic 于投资者活动披露部分财务后，部分投资者已把估值下调到约 1.5 万亿。
下调的原因不难猜：Reuters 8 月的独家报道说，这个估值锚定的是"2028 年营收 1900-2000 亿美元"的预测——也就是说，今天的价格买的是两年后再造三个"现在的 Anthropic"。信这个故事，1.5 万亿不贵；不信，那就是泡沫顶点。1.5 万亿对应 2026 上半年年化营收约 46 倍 PS，这笔账怎么算，取决于你对 AI 需求持续性的判断。
有个插曲能说明市场情绪的分裂：早在 10 月 12 日，就有研究机构公开把这次 IPO 称为"对投资者轻信程度的空前考验"（an unprecedented test of investor gullibility）。两周后招股书公开，HN 主帖拿到 1444 分、1411 条评论，"惊叹增长"和"质疑泡沫"吵成一团——这个分裂大概会一直延续到定价日。
治理结构：想买它股票的人得先想清楚一件事
这次 IPO 还有个绕不开的争议点：Anthropic 不是普通公司，它是一家公益公司（PBC），外加一套创始人治理安排——七位联合创始人组成的新 LLC 承担"使命守护者"角色，另有 Long-Term Benefit Trust（长期受益信托）约束董事会构成。
翻译成人话：买它股票的公众股东，权利会比一般上市公司小——公司可以合法地以"人类长期利益"为由做出不符合短期股东回报的决策。支持者认为这是 AI 公司应有的安全护栏；质疑者认为这是同股不同权的极端形态。HN 上已有讨论指出投资者签署前即已知情、事后诉讼理由不成立——但"知情"和"接受"之间，每个打新人还是得自己掂量。
把整件事说清楚：站得住的是营收增长（数字多源咬合、客户分散度改善）、现金储备、以及"净亏大头是计提"的澄清；有争议的是估值锚（2 万亿预测 vs 1.5 万亿现实）和治理结构；没证实的包括最终定价、7500 亿算力开支的具体口径。11 月 16 日当周挂牌（WSJ 口径）之后，市场会给出它的投票——而这次投票的结果，大概率会重新定义整个 AI 行业的融资水温。
参考来源
- SEC EDGAR：Anthropic, PBC S-1 原文（一手文件）
- Anthropic 官方公告（一手确认）
- The Verge：Anthropic files for its IPO（净亏损会计计提澄清）
- The Verge：Five weird details（高管薪酬等细节）
- Reuters（via Livemint）：估值下调报道
- WSJ：S-1 Shows Red-Hot Growth, Risks（7500 亿算力开支）
- 华尔街见闻：Anthropic 正式提交 IPO 申请（11/16 挂牌计划、客户结构）
- Hacker News：Anthropic S-1 主帖（算力承诺拆解等社区分析）
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。
One sentence: Claude Developer Anthropic publicly presented S-1 on October 26, with the code ANT landing at NASDAQ — a semi-annual gain of $16.3 billion, an increase of 130 per cent, a net deficit of $80.7 billion (largely accounted for rather than cash), anecdotal valuation of $1.5-2 trillion, which is expected to be one of the largest IPOs in history; the plan is to be put up by mid-November.
There's a hunch in the past six months that Anthropic will be on the market sooner or later. In June, it confirmed confidential filings, and since then the valuation rumours have gone from "billions of dollars club" to "2 trillions above SpaceX". October 26th, boots landed - S-1 officially open, code one word: ANT.
This 300-page document is the first time that a page-by-page view of the company's books has been made. Read the feeling, it's quite appropriate in Fortune: Surreal — half the declaration of the Miracle Age, half the warning of the risk of survival, with a hugely growing and alarming financial statement in the middle.
You want to check the original numbers?
The two most beautiful sets of numbers in the books
The first group consists of receipts: 41 million in 2023 and 380 million in 2024 and 4585 million in 2025 and 162.6 million in the first half of 2026. Half a year of three-and-a-half times the last year's volume, 1.3 per cent. And this is going to happen at the speed of the technology, IPO, and Scylla couldn't find a few enemies -- in contrast, Anthropic was a company that was set up in 2021.
The client structure is also moving in the right direction: Q3 about 79 per cent of the accounts are received from clients other than the top five clients. For an AI company that's often questioned about a few large clients, this is a rare addition to the book.
The second group consists of losses: a net loss of $80.7 billion in the first half of the year. There's a key clarification that most of the titles won't tell you. This 80.7 billion is the bulk of the accounting generated by the financing instruments issued by companies, not by the burning of cash. That means "book losses" and "business burn money" are different things. Of course, it is not without cost, but rather without a real obligation or share in the future.
The bottom line: As of the disclosure, Anthropic cumulative financing $141.1 billion and cash and equivalents $131.7 billion. This is probably one of the thickest wallets in business history.
Calculator bills: real suspense.
What is more worthwhile than losses is a commitment to calculation. The high-profile analysis on HN adds to the multi-year commitments in S-1 one by one: Google 1140 billion + AWS 110 billion + Broadcom 161 billion + SpaceX 120 billion, totalling approximately $50.5 billion. S-1 It is also suggested that the cumulative cost of computing power in the coming years could be as high as $75 billion.
What is this concept? Anthropic's multi-year commitment to computing has exceeded the annual fiscal budgets of many countries. And that's why it's going to go on the market -- the first-level market money is fast feeding this level of spending, and IPO is locked in the open market's long-term financing path.
By the way, even the "unusual details" are available in this document (The Verge dedicated five): CEO Dario Amodei received nearly $18 million last year (largely stock options, low in large factories), sister and president Daniela 16.4 million; company spent $2.75 million on personal security for Dario and $1.29 million on residential security for Daniela; Dario spent $97,055 on personal use of company aircraft. And one of the most "Anthropic" -- the risk factor sections are actually written in "the conscious AI" and "the catastrophic risk." It is indeed unprecedented that a company is alerting people to the risks of survival in documents that sell stock to the public.
Valuation: downwards from 2 trillion to 1.5 trillion
Market-based target valuations were once as high as $2 trillion — 1.77 trillion more than SpaceX, pushing Britain (5 trillion+). However, Reuters reported an interesting detail: after Anthropic disclosed some of the financial aspects of investor activity, some investors have reduced the valuation to about 1.5 trillion.
It's not hard to guess: Reuters' exclusive in August reported that this valuation anchored a projection of "1900-2000 billion dollars in 2028." So today's price is two years to build three more "now Anthropic." Believe this story, 1.5 trillion is not expensive; believe it, that's the top of the bubble. 1.5 trillion corresponds to about 46 times the annualization of 2026 pS in the first half of the year.
There is an episode that points to the division of market sentiment: as early as October 12, an IPO was publicly called an unprescribed test of investor confidence. Two weeks later, the book was made public, HN received 1444 points, 1411 comments, "Expressed growth" and "question bubble" – a split that would probably last until pricing day.
Governance: Whoever wants to buy it needs to think first. Yes.
This time, IPO also has a point of contention: Anthropic is not an ordinary company, it is a public good company (PBC), plus a set of founders' governance arrangements - a new LLC consisting of seven co-founders who assume the role of "manager" and a Board of Directors bound by Long-Term Benefit Trust.
Translation: Public shareholders who buy their shares will have smaller rights than general listed companies — companies can legally use the term “long-term human interest” to make decisions that do not meet short-term shareholder returns. Proponents consider this to be the security fence that AI deserves; the challengers consider this to be an extreme form of different equity. It has been discussed that the investor was informed before signing, that the cause of action was not ex post facto - but that between " know" and "acceptance ", it is up to each new recruit to take his or her own account.
The whole thing is clear: what holds up is the clarification of revenue growth (digital multiple bites, improved customer dispersion), cash reserves, and the "net deficit is a figure"; what is at issue is the valuation anchor (2 trillion projections vs. 1.5 trillion reality) and the governance structure; and what is not proven is the specific calibration of final pricing, 750 billion computing expenses. On November 16th, when the cards were put up (WSJ caliber), the market would give its vote - and the result of this vote would probably redefine the funding temperature of the entire AI industry.
References
- SEC EDGAR: Anthropic, PBC S-1 original (first-hand document)
- Anthropic, official announcement.
- The Verge: Anthropic counts for its IPO (clarification of net loss accounting)
- The Verge: Five Weird details.
- Reuters (via Livemint): The downward reporting of the valuation
- WSJ: S-1 Shows Red-Hot Growth, Risks (750 billion in hard money)
- Wall Street news: Anthropic officially submitted an IPO application (11/16 listing plan, client structure)
- Hacker News: Anthropic S-1 main post (compute community analysis such as commitment to dismantling)
Auto-fetched and translated by AI Atlas. All rights belong to the original author.