Anthropic的招股书第10条写着这么一句话：名单上的七个人，每人向公司出资1美元，认购一股F类股票。
7美元换控股权，这买卖值
一、史上最大IPO，Anthropic准备融多少
北京时间10月26日晚间，Anthropic正式向美国证券交易委员会（SEC）递交S-1注册声明，申请在纳斯达克全球精选市场挂牌，股票代码定为「ANT」。这是人类历史上首个前沿大模型公司的IPO申请，也是当前硅谷最受关注的资本事件。
招股书草案披露，公司计划募资规模最高约1000亿美元。按此计算，若发行成功，Anthropic将一举超越SpaceX在今年6月创下的862.5亿美元（含超额配售）募资纪录，成为美股历史上规模最大的IPO。高盛和摩根士丹利担任此次发行的主承销商，承销团队阵容与SpaceXIPO相似。
截至2026年5月28日，Anthropic在H轮融资后的投后估值为9650亿美元。市场对本次IPO估值的预期落在1.5万亿至2万亿美元区间。若按上限2万亿美元定价，对应每股价格将在数十美元量级，远超目前多数科技股的估值水平。
招股书同时透露，公司累计融资已达1411亿美元，资金来源于谷歌、Salesforce Ventures、Sound Ventures、Zoom Ventures等机构。截至2026年6月末，公司持有现金及现金等价物1124亿美元、短期投资193亿美元，另有合计770亿美元尚未动用的票据购买协议额度——谷歌200亿、亚马逊150亿、博通420亿。这些备用流动性意味着，即便IPO募资不及预期，公司也有足够的缓冲空间维持运营。
千亿募资任务已下达
1000亿美元目标背后的估值逻辑
Anthropic选择在此时冲刺IPO，背后是多重压力的叠加。第一重压力来自算力采购合同的履约需求。招股书披露，公司已向微软承诺采购449亿美元算力、向谷歌承诺1145亿美元、与SpaceX AI签订的合同最高金额达1205亿美元。这三份合同的合计承诺支出超过2700亿美元，远超公司目前的现金储备。
第二重压力来自竞争格局。OpenAI在同一时期也秘密递交了上市申请，尽管Sam Altman在9月接受《财富》杂志采访时确认「今年不会上市」，理由是「安全方面正在发生的一切」，但市场普遍认为OpenAI的上市时间表只是延后而非取消。Anthropic选择先行撞线，是在争夺「首家上市前沿AI公司」的标签。
第三重压力来自资本市场情绪。2026年下半年，AI与芯片股经历了一轮明显抛售，SpaceX上市首日大涨19%后回落至发行价附近，投资者对高估值成长股的态度趋于谨慎。Anthropic选择在市场情绪并不友好的时刻强行冲刺，意味着管理层对自身的盈利能力和现金流有较强信心。
若落地将超越SpaceX创下的860亿纪录
SpaceX的862.5亿美元募资纪录，是马斯克在2024年创造的。当时SpaceX以1.8万亿美元估值完成IPO，成为美股历史上规模最大的私人公司上市案例。Anthropic的目标规模超过了SpaceX的16%，这一增幅在IPO历史上极为罕见。
作为对比，2018年Uber上市募资87亿美元，2021年 Airbnb上市募资76亿美元，2023年 DoorDash上市募资50亿美元。Anthropic的目标规模是这些案例的10倍以上，其体量已与部分中型科技公司相当。
如果募资1000亿美元全部到位，Anthropic将成为全球市值最高的上市公司之一。按2万亿美元估值计算，其市值将超过Meta（约1.5万亿美元）、接近苹果（约3万亿美元）的一半。这一估值水平意味着，市场对其未来现金流的折现预期极为乐观。
高盛、摩根士丹利担任主承销商
招股书未披露具体的发行股数和定价区间，预计还需约两周时间，公司在评估投资者反馈后提交更新版上市文件时，才会公布这些细节。知情人士表示，IPO路演最早可能在11月9日当周启动，11月16日当周挂牌。
承销团队已就位
二、7美元控股权的设计逻辑
招股书第10条的这段描述，揭示了Anthropic治理结构中最核心、也最具争议的设计：七位联合创始人通过Founder LLC持有唯一一股Class F特殊股票。这一股不享有任何经济收益权，分不了一分钱利润，但它的投票权固定等于公司全部投票权的50.1%，包括选举董事会成员、修改公司章程、批准重大交易等所有关键事项。
协议里还特意写了一句：这1美元是不可撤回的出资，七个人放弃了任何要回这笔钱的权利。这意味着，即便创始人退出公司、离世或转让股份，Class F股票的投票权也不会被稀释或转移。
这条款评审时吵翻了吧
Class F股票的机制与不可稀释条款
Class F股票的不可稀释条款是这份治理设计的核心。通常情况下，公司增发新股会稀释原有股东的持股比例。但招股书明确规定，Class F股票的50.1%投票权比例不受后续发行影响。即使Anthropic未来发行10亿股新股票，Class F的投票权比例依然维持在50.1%。
这一条款的法律依据来自特拉华州公司法第151条，允许公司发行具有不同投票权的多类别股票。类似的结构在Alphabet（Google母公司）的Class A、Class B、Class C体系中也有体现，但Anthropic的设计更为极端——单一股票掌握过半投票权，且七人共享这一投票权。
七位联合创始人的背景与持股比例
七位联合创始人分别是：CEO Dario Amodei、总裁兼董事会主席Daniela Amodei（Dario的妹妹）、首席算力官Tom Brown、首席科学官Jared Kaplan、首席架构师Sam McCandlish、可解释性研究负责人Chris Olah，以及Anthropic Institute负责人Jack Clark。他们均出自OpenAI早期核心团队，2021年因对AI安全治理理念的分歧出走，创立Anthropic。
招股书披露，七人通过Founder LLC合计持有公司约10%的经济利益，每人持股约1.4%。按2026年5月H轮9650亿美元估值计算，每人身家均超过130亿美元。但他们的控制权远超车份比例——Class F股票的50.1%投票权意味着，即便七人各自转让经济股份，控制权依然集中在Founder LLC手中。
「使命守护者」的定位与治理边界
招股书对这七人的定位是「我们使命的守护者」。这一定位背后，是Anthropic对自身使命的强调：确保AI系统安全、可靠、可控。七位创始人均来自OpenAI的安全研究团队，他们的核心理念是「前沿AI研究必须以安全为前提」，这与OpenAI后来转向「快速推进AGI」的战略方向产生了根本分歧。
但治理结构的极端设计也带来了潜在风险。50.1%的投票权意味着，七人可以在股东反对的情况下强行通过任何决议，包括更换CEO、出售公司、改变使命方向等。招股书没有设置独立的受托人来制衡这一权力，而是将决策权完全交给七位创始人。
代码写了，权限给满了
三、收入暴涨与账面亏损的反差
Anthropic的财务数据呈现出一种强烈的反差：收入侧高速增长，利润侧持续亏损。招股书披露，2025年全年营收45.85亿美元，2026年上半年营收跃升至162.86亿美元，五个季度累计营收增长约30倍。今年二季度，公司经营利润首次转正，达到9.74亿美元。
但与此同时，2026年上半年账面净亏损高达807.48亿美元。这一亏损数额远超收入规模，引发了市场的广泛质疑。招股书的解释是：亏损主要来自可转换票据的公允价值变动，而非经营性亏损。
账面亏的是纸面财富
五个季度营收涨30倍的增长路径
Anthropic的营收增长主要来自两个渠道：按使用量计费的API收入和企业订阅收入。招股书披露，2024年全年按使用量计费收入为2.88亿美元，占公司总收入的75%；2025年全年这一数字增长至38亿美元，占总收入的83%；2026年上半年按使用量计费收入达到131亿美元，占总收入的80%。
这一增长模式的背后，是Claude模型在开发者生态中的快速渗透。Claude Opus 4、Claude 3.5 Sonnet等版本相继发布，吸引了大量企业和开发者接入API。截至2026年6月，公司已服务超过12,000家年收入10万美元以上的客户，其中超过79%的收入来自前五大客户以外的客户。
807亿美元净亏的构成拆解
807亿美元的净亏损中，约400亿美元来自可转换票据的公允价值变动。这类票据是公司在早期融资中发行的可转换债券，其账面价值随公司估值波动而调整。由于Anthropic在2026年完成了多轮高估值融资，票据的公允价值大幅上升，形成了账面亏损。
另一部分亏损来自股权激励费用和研发支出。Anthropic的研发投入常年保持在收入规模的50%以上，主要用于算力采购、模型训练和安全研究。公司没有对研发费用进行资本化，而是全部费用化处理，这进一步放大了账面亏损。
调整后可观的真实盈利能力
如果剔除公允价值变动、股权激励费用和一次性支出，Anthropic的调整后净利润在2026年上半年达到26.71亿美元，为正数。这一数字意味着，公司的核心业务已经具备盈利能力，只是会计处理上的账面亏损掩盖了这一事实。
调整后终于赚回来了
四、千亿算力承诺的长期风险
Anthropic的算力采购承诺是招股书中最引人关注的长期风险点。截至2026年6月末，公司已签署的采购承诺与租赁协议合计支出超过一个未披露的巨额数字，绝大部分与算力和基础设施相关。其中相当大部分发生在2030年前，其余延续至2031至2045年。
招股书披露了部分具体数字：向微软承诺采购449亿美元算力、向谷歌承诺1145亿美元、与SpaceX AI签订的合同最高金额1205亿美元。这三份合同的合计承诺超过2800亿美元，是Anthropic目前现金储备的近两倍。
未来十年的砖头已搬完
微软、谷歌、SpaceX AI的采购合同
微软的449亿美元采购主要通过Azure云平台执行，公司提供英伟达GPU算力支持。谷歌的1145亿美元合同则涉及TPU芯片和云计算资源。SpaceX AI的1205亿美元合同最为特殊，这是Anthropic与马斯克旗下公司的合作，涉及Starlink卫星通信和算力基础设施。
这三份合同的共同特点是：预付款比例高、违约罚金重、价格锁定机制。这意味着，即便未来AI需求下滑或技术路线变化，Anthropic也无法轻易退出这些合同。
从纯云采购到「云+自有」的混合模式
招股书同时透露，公司正从纯云采购转向「云+自有」混合模式。Anthropic Direct项目允许公司直接租赁芯片并托管在专属设施，长期看，在算力投资中的比重越来越大。这一转型意味着，未来的资本支出将进一步增加，但公司的控制权和成本结构也将得到优化。
十年期基础设施支出的财务影响
千亿级别的算力承诺意味着，Anthropic在未来十年内将持续承受巨大的财务压力。即便公司保持当前的收入增速，也需要将大部分现金流用于履约采购合同。一旦收入增速放缓，这些长期承诺可能成为沉重的包袱。
架构定了，债务也定了
五、上市时机的市场考验
Anthropic选择在此时冲刺IPO，面临的市场环境并不友好。
更值得对照的是它的老对手。OpenAI在同一时期也秘密递交了上市申请，但Sam Altman在9月确认「今年不会上市」，理由是「安全方面正在发生的一切」。OpenAI的推迟，为Anthropic留出了窗口期。
AI与芯片股的近期波动
2026年第三季度，标普500指数中的AI相关成分股整体下跌约15%，芯片股跌幅更大。英伟达、AMD等公司的股价均出现明显回调。这一趋势反映了市场对AI基础设施投资回报率的担忧——算力需求增长迅速，但商业化变现的路径仍不清晰。
OpenAI的对比与竞争格局
OpenAI的推迟上市，为Anthropic创造了时间窗口。但OpenAI的估值同样高达9650亿美元（H轮），且拥有GPT系列模型的强大品牌效应。如果OpenAI在未来12个月内完成上市，Anthropic将面临激烈的竞争。
投资者最关心的三个问题
投资者在评估Anthropic IPO时，最关心的问题有三个：第一，算力采购承诺的财务风险如何量化？第二，50.1%的创始人控制权是否会削弱公司治理？第三，收入增速能否持续覆盖千亿级采购支出？这三个问题的答案，将决定IPO的最终定价和后市表现。
Class F股票的诞生逻辑与50.1%锁定机制
招股书第10节的措辞极其克制，几乎像是在描述一笔微不足道的日常交易。但背后的机制设计，精确到令人侧目。
Class F股票只有一股，由七位联合创始人通过一个名为 Founder LLC 的特殊目的实体共同持有。这一股不享有任何经济收益权——不分红、不分配清算资产、不参与任何利润分配。它唯一的属性是投票权。
投票权的锁定方式有两个关键点。第一，这一股在计算表决权时，按固定比例折算为公司全部投票权的50.1%。这个数字不是约数，是精确的协议条款。第二，无论后续增发多少新股、无论投资人以何种估值进入、无论公司进行多少次并购或剥离，这个50.1%的比例都不会被稀释。招股书用了一句强调性的表述："所有交给股东投票的事项都算"，包括选举长期利益信托（LTBT）受托人席位以外的董事。
这意味着什么。即便公司未来再发行十亿股普通股，七位联合创始人通过这一股仍然握着过半表决权。资本可以无限进入，控制权永远留在创始人手中。
七位联合创始人的身份与各自角色
Founder LLC 的成员名单，基本就是一份 Anthropic 核心科研与工程班底的完整画像。
Dario Amodei，CEO，斯坦福博士，主攻机器安全研究，是 Anthropic 的公开面孔和战略决策者。
Daniela Amodei，总裁兼董事会主席，Dario 的亲妹妹，负责日常运营和公司治理层面的统筹。
Tom Brown，首席算力官，主导算力基础设施的采购与调度，是 Anthropic 大规模 GPU 部署的核心操盘手。
Jared Kaplan，首席科学官，前 DeepMind 研究员，主导模型训练的科学方法论和性能评估体系。
Sam McCandlish，首席架构师，负责 Claude 模型的架构设计与工程实现。
Chris Olah，可解释性研究负责人，开创了 AI 可解释性研究的新方向，是 Anthropic 差异化定位的关键支柱。
Jack Clark，Anthropic Institute 负责人，专注于 AI 政策和治理框架的外部研究输出。
七人全部出身 OpenAI 早期核心团队。2015年 OpenAI 成立时，他们是第一批工程师和研究者。2018年前后，随着 OpenAI 从非营利转向营利性结构，Dario Amodei 等人对 AI 安全治理的方向产生根本性分歧，最终选择出走，于2015年创立 Anthropic。
招股书对这七人的定位只有一个短语："我们使命的守护者"。
每人1.4%的经济权益 vs 10%合计投票权
经济权益和控制权的分离，是这套治理设计最刺眼的反差。
按 2026 年 5 月 H 轮 9650 亿美元估值计算，七位联合创始人每人持有约 1.4% 的股权，合计约 10%。这个数字放在任何一家上市公司都算得上慷慨——按上限 2 万亿美元估值算，每人身家超过 130 亿美元，七人合计 900 多亿。
但 10% 的经济权益，换来的不是 10% 的投票权，而是 50.1%。
换句话说，他们用最少的经济股份，锁死了最多的控制权。剩余 90% 的股权掌握在投资人、员工期权池和公众股东手中，但这些股东在关键治理事项上，始终是少数派。
这种设计并非 Anthropic 首创。硅谷早有先例——Facebook 的 Class B 股票、Google 的超投票权架构，都是同类逻辑。但 Anthropic 的设计更极端：其他公司通常给创始人保留 20% 到 40% 的投票权，Anthropic 直接锁定 50.1%，跨过绝对控制线的最低门槛，且不可稀释。
Anthropic Class F 股票治理结构
解剖完这个结构，一个必须回答的问题是：为什么 Anthropic 要费这么大劲，把治理权重拉到这个极端位置。
答案藏在它的创业史里。
参考文献
延伸入口
- 原文归档：tobemagic.github.io/ai-magician…
- 公众号：计算机魔术师
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。
Article 10 of Anthropic's equity book states that seven individuals on the list each contribute US$ 1 to the company to subscribe to a F stock.
Seven dollars for controlling. That's the value.
I. The biggest IPO ever. How much is Anthropic going to melt?
On the evening of October 26, Beijing time, Anthropic officially submitted a S-1 registration statement to the United States Securities and Exchange Commission (SEC) requesting to be listed in the NASDAQ global niche market with the stock code "ANT". This is the first IPO application in human history from the front-line company LLM and is also the most significant capital event in Silicon Valley today.
The draft equity book reveals that the company plans to raise up to $100 billion. According to this calculation, if the release is successful, Anthropic will go beyond SpaceX’s $8,625 billion (including over-sale) fund-raising record in June this year, making it the largest IPO in the history of the United States. Goldman Sachs and Morgan Stanley served as main distributors for the launch, and the sales team was similar to SpaceXIPO.
As at 28 May 2026, Anthropic had a post-investment valuation of $965 billion after the H round. Market expectations for this IPO valuation fall between $1.5 trillion and $2 trillion. If priced at a ceiling of $2 trillion, the corresponding price per share would be in the order of several dozen dollars, well above the current valuation level for most of the technology units.
At the same time, the equity book revealed that cumulative corporate financing had reached $141.1 billion, financed by Google, Salesforce Ventures, Sound Ventures and Zoom Ventures. By the end of June 2026, companies held $112.4 billion in cash and cash equivalents, $19.3 billion in short-term investments and a total of $77 billion in unused instruments purchase agreements quota — 20 billion Google, 15 billion Amazon and 42 billion Chase. These back-up liquidity means that the company has sufficient buffer space to maintain operations even if IPO raises less than expected.
A multi-billion-dollar fund-raising mission has been launched.
Valuation logic behind the $100 billion target
Anthropic chooses to sprint IPO at this time, with multiple pressures on its back. The first pressure arises from the need for performance in the calculation of procurement contracts. The equity book disclosed that the company had committed $4.49 billion to Microsoft, $114.5 billion to Google and $12.5 billion to a contract with SpaceX AI. The combined committed expenditure of the three contracts exceeded $270 billion, far exceeding the current cash reserves of the company.
The second heavy pressure comes from the pattern of competition. During the same period, Open AI also filed a secret listing application, and although Sam Altman confirmed in an interview with Fortune magazine in September that “this year he will not be listed” on the grounds that “what is happening in the area of security”, it is generally accepted that Open AI’s listing schedule is delayed rather than cancelled. Anthropic chooses the front line to compete for the label "first listed front-line AI."
The third stress comes from the mood of capital markets. In the second half of 2026, the AI and Chip Unit experienced a clear round of sales, with SpaceX showing a sharp 19 per cent rise in the first day of the market and falling near the issue price, with investors being cautious about high-value growth. Anthropic ' s choice to push when market sentiment is not friendly means that management has greater confidence in its profitability and cash flow.
If you land, it will surpass the 86 billion recorded by SpaceX.
SpaceX's $8.625 billion fund-raising record was created by Mask in 2024. At that time, SpaceX completed the IPO with a value of $1.8 trillion and became the largest private company in the history of the United States stock. Anthropic ' s target size exceeds 16 per cent of SpaceX, an increase that has been rare in the history of IPO.
By contrast, US$ 8.7 billion were raised in 2018, US$ 7.6 billion in 2021 and US$ 5 billion in 2023 by DoorDash. Anthropic ' s target size is more than 10 times that of these cases, and its volume is already comparable to that of some medium-sized technology companies.
If $100 billion is fully raised, Anthropic will be one of the most marketed companies in the world. On the basis of the $2 trillion valuation, its market value will exceed half of Meta (approximately $1.5 trillion) and close to apples (approximately $3 trillion). This level of valuation means that the market is extremely optimistic about its forecast for future cash flows.
Goldman Sachs and Morgan Stanley are the main vendors.
The number of specific distribution shares and pricing areas is not disclosed in the book, and it is expected that it will take about two weeks before companies submit updated listing documents after assessing investor feedback. According to sources, the IPO road show could be launched as early as the week of 9 November, when it was boarded.
The sales team is in position.
II. Design logic for US$ 7 holdings
This description of article 10 of the equity book reveals the most central and controversial design of the Anthropic governance structure: the seven co-founders hold the only special stock of Class F through the Foundation LLC. This share does not enjoy any right to economic gain and does not share a single cent of profits, but its voting rights are fixed at 50.1 per cent of the company ' s total voting rights, including all key matters such as the election of board members, the amendment of company charters and the approval of major transactions.
The agreement also specified that the $1 was an irrevocable contribution and that seven people had waived any right to recover it. This means that even if the founders had withdrawn from the company, had died or had transferred shares, the voting rights in Class F stock would not have been diluted or transferred.
You must've had a lot of noise about this.
Class F stock mechanism and non-dilution clause
Class F stock's non-dilution clause is at the heart of this governance design. Typically, the company's new shareholding dilutes the shareholding ratio of the original shareholder. However, the equity book clearly states that 50.1 per cent of the voting share of Class F stocks is not affected by subsequent issuances. Even though Anthropic will issue new shares of 1 billion shares in the future, Class F ' s voting ratio remains at 50.1 per cent.
The legal basis for this provision derives from article 151 of the Delaware Companies Act, which allows companies to issue multi-category shares with different voting rights. A similar structure is reflected in the Class A, Class B and C systems of Alphabet (Google parent company), but the design of Anthropic is more extreme — a single stock holding a semi-voting vote, which seven people share.
Background and shareholding ratio of seven co-founders
The seven co-founders were CEO Dario Amodei, Daniela Amodei (sister of Dario), President and Chairman of the Board, Tom Brown, Chief Accounting Officer, Jared Kaplan, Chief Scientific Officer, Sam McCandlish, Chief Architect, Chris Olah, Head of Explanatory Research, and Jack Clark, Head of Anthropic Institute. They all emerged from the early core team of Open AI, which created Anthropic in 2021 because of differences over the AI security governance concept.
The equity book disclosed that seven persons held approximately 10 per cent of the financial interests of the company through the Foundation LLC, each holding approximately 1.4 per cent. Based on the May 2026 round H valuation of $965 billion, each household exceeded $13 billion. But their control far exceeded that of the 50.1 per cent vote of the Class F stock, which meant that even if the seven individuals each transferred their economic shares, control remained in the hands of Founder LLC.
The Positioning and Governance of Mission Guardians
The book of recruitment is designed for these seven people as "the guardian of our mission". Behind this positioning is the Anthropic emphasis on its mission: to ensure that AI systems are safe, reliable and manageable. The seven founders came from Open AI’s security research team, whose core philosophy is that “Front-AI research must be safe,” which is fundamentally different from Open AI’s subsequent shift to the strategic direction of Quick Advance AGI.
However, the extreme design of governance structures also poses potential risks. 50.1 per cent of voting rights mean that seven people can force the adoption of any resolution against shareholders, including the replacement of the CEO, the sale of the company, the change of mission, etc. The equity book does not have an independent trustee to counterbalance this power, but instead gives full decision-making power to the seven founders.
It's written. permission is full.
III. Deficiencies between soaring income and book losses
Anthropic's financial data show a strong contrast: high growth on the income side and sustained loss on the profit side. The equity book revealed that, in 2025, the battalion received $4,585 million for the whole year, that in the first half of 2026, the battalion jumped to $16,286 million, and that the cumulative five-quarter harvest increased by about 30 times. For the first time in the second quarter of this year, the company ' s operating profits went up to $974 million.
At the same time, however, the net book losses in the first half of 2026 amounted to $80,748 million. This loss is well above the size of income and raises widespread questions in the market. The equity statement explained that the loss was mainly due to a change in the fair value of the convertible instrument rather than to an operating loss.
It's paper wealth that's missing.
Five quarters, 30 times the growth path.
Anthropic ' s growth in revenue is mainly from two sources: the API income, which is based on usage, and the corporate subscription income. The equity book disclosed that in 2024, $288 million, or 75 per cent of the company ' s total income, had been earned throughout the year; in 2025, that figure had increased to $3.8 billion, or 83 per cent, of the total income; and in the first half of 2026, it had earned $13.1 billion, or 80 per cent, of the total income.
Behind this growth pattern is the rapid penetration of the Claude model in the developers ' ecology. Claude Opus 4, Claude 3.5 Sonnet and others successively publish have attracted a large number of businesses and developers to access API. By June 2026, the company had served more than 12,000 clients with an annual income of more than $100,000, of which more than 79 per cent were from clients other than the top five.
Net deficit of $80.7 billion
Of the net loss of $80.7 billion, some $40 billion resulted from changes in the fair value of convertible instruments. Such instruments are convertible bonds issued by companies in early financing, the book value of which is adjusted to the fluctuations in corporate valuations. As Anthropic completed multiple rounds of high valuation finance in 2026, the fair value of the instrument increased significantly, resulting in a book loss.
Another part of the loss is due to equity incentives and R & D expenditures. Anthropic ' s research and development inputs have remained at more than 50 per cent of revenue per year, mainly for computing, modelling and safety research. The company did not capitalization R & D costs, but fully expensed them, further magnifying the book losses.
Readjusted real profitability.
Anthropic ' s adjusted net profit reached $2,671 million in the first half of 2026, positive, if fair value variations, equity incentive costs and one-time expenses were excluded. This figure implies that the company ' s core business is already profitable, a fact that is masked by the book losses on accounting processing.
I finally got it back.
IV. LONG-TERM RISK PROJECTS OF LONG-TERM OBLIGATIONS
Anthropic ' s commitment to computing is the most interesting long-term risk point in the equity book. As at the end of June 2026, the firm had signed purchase commitments and combined lease agreements that exceeded an undisclosed large amount, most of which related to computing and infrastructure. A significant proportion of these occurred before 2030, with the remainder extending from 2031 to 2045.
Some specific figures were disclosed in the equity book: a commitment to Microsoft to procure $4.49 billion in calculus; a commitment to Google $114.5 billion; and a contract with SpaceX AI for a maximum of $12.5 billion. The combined commitments for the three contracts exceeded $280 billion, nearly twice the current cash reserve of Anthropic.
The bricks have been moved for the next 10 years.
Procurement contracts for Microsoft, Google, SpaceXAI
Microsoft ' s $44.9 billion procurement was implemented mainly through the Azure Cloud platform, with the company providing GPU computing support. Google’s $114.5 billion contract involved TPU chips and cloud computing resources. The $12.5 billion contract for SpaceX AI, the most special, was the cooperation between Anthropic and Mask Sub-Citizen, which involved Starlink satellite communications and computing infrastructure.
The common features of the three contracts were the high rate of advance payments, the heavy penalty for breach of contract and the price locking mechanism. This means that even if the future AI needs decline or the technical route changes, Anthropic will not be able to easily withdraw from these contracts.
From pure cloud procurement to cloud-owned hybrid mode
The book also reveals that the company is moving from pure cloud procurement to a hybrid "cloud + ownership" model. The Anthropic Direct project allows companies to lease chips directly and host them in exclusive facilities, and in the long run, they are increasingly involved in computing investments. This transformation means that capital expenditure will increase further in the future, but that corporate control and cost structures will also be optimized.
Financial impact of 10-year infrastructure expenditure
The hundreds of billion-grade calculus commitment means that Anthropic will continue to be under great financial pressure over the next decade. Even if the company were to maintain its current revenue growth rate, it would need to spend most of the cash flow on performing procurement contracts. These long-term commitments can become a heavy burden once income growth slows down.
It's set. It's set.
V. Market test of listing timing
Anthropic chose to sprint IPO at this time, and the market environment was not friendly.
More worthy of comparison is its old rival. During the same period, Open AI also filed a secret listing application, but in September, Sam Altman confirmed that “this year he will not be listed” on the grounds that “what is happening with regard to security”. Open AI's delay left a window period for Anthropic.
Recent fluctuations between AI and chip units
In the third quarter of 2026, the AI-related unit in the Standard 500 index fell by about 15 per cent overall and the chip unit fell even more. There has been a marked reversal in the stock prices of companies such as Weida and AMD. This trend reflects the market's concern about the return on AI infrastructure investment — rapid growth in computing demand, but the path to commercialized realization remains unclear.
Open AI contrasts with competition patterns
Open AI's late listing created a window of time for Anthropic. However, Open AI ' s valuation was also as high as $965 billion (wheel H) and had a strong brand effect with GPT series models. If Open AI completes its listing in the next 12 months, Anthropic will face intense competition.
Three issues of greatest concern to investors
In assessing Anthropic IPO, investors have three main concerns: first, what are the financial risks of calculating procurement commitments? Second, would 50.1 per cent of founders control weaken corporate governance? Thirdly, will the growth rate of revenues continue to cover hundreds of billion class procurement expenditures? The answers to these three questions will determine the final pricing and post-market performance of IPOs.
Class F stock birth logic and 50.1% lock-in mechanism
The wording of section 10 of the equity book is extremely restrained, almost as if it were a description of an insignificant daily transaction. But the mechanism behind it is designed to be very precise.
Class F shares are held by seven co-founders through a special purpose entity called Founder LLC. This share is not entitled to any economic gain — no red, no distributive of liquidated assets, no participation in any profit distribution. Its only attribute is the right to vote.
There are two key points to the targeting of voting rights. First, this share is converted in a fixed proportion to 50.1 per cent of the total voting rights of the company. This figure is not an approximate number, but precise terms of agreement. Second, the 50.1 per cent share will not be diluted, regardless of subsequent additions, whether investors enter with any valuation, and regardless of the number of mergers and acquisitions or divestments carried out by companies. The equity book contained an emphasis on "all matters referred to shareholders for voting" and included the election of directors other than the LTBT trustee seat.
It means nothing. Even though the company will issue an additional billion shares of ordinary stock in the future, seven co-founders still hold half the voting power through this share. Capital has unlimited access and control remains in the hands of the founders forever.
The seven co-founders and their respective roles
The list of members of the Foundation LLC is basically a complete picture of the core of the Anthropic scientific and engineering class.
Dario Amodei, CEO, Dr. Stanford, major research on machine safety, is an open face and strategic decision-maker for Anthropic.
Daniela Amodei, President and Chairman of the Board of Directors, a close sister of Dario, is responsible for day-to-day operational and corporate governance-level integration.
Tom Brown, Chief Calculator, leading the procurement and movement of computing infrastructure, is the core operator for the large-scale GPU deployment of Anthropic.
Jared Kaplan, Chief Scientific Officer, former DeepMind Fellow, scientific methodology and performance assessment system leading model training.
Sam McCandlich, Chief Architect, responsible for the architecture design and engineering realization of the Claude model.
Chris Olah, Head of Explanatory Studies, has created a new direction for AI Explanatory Research, which is a key pillar of the differential positioning of Anthropic.
Jack Clark, Head of the Anthropic Institute, focused on the external research output of the AI Policy and Governance Framework.
Seven were all from Open AI early core team. When Open AI was founded in 2015, they were the first engineers and researchers. Around 2018, as Open AI moved from a not-for-profit to a profit-making structure, Dario Amodei and others fundamentally divided on the direction of AI security governance, eventually choosing to run away and create Anthropic in 2015.
There's only one phrase in the book for these seven: "The Guardian of Our Mission".
1.4% economic rights per person vs 10% combined voting Rights
The separation of economic interests and control is the most striking contrast in the design of governance.
Based on the May 2026 H round valuation of $965 billion, each of the seven co-founders held approximately 1.4 per cent of the equity, or a total of about 10 per cent. This figure is generous for any listed company — over $13 billion per person, with a ceiling of $2 trillion, and over $90 billion for seven persons.
However, 10% of the economic benefits, not 10% of the voting rights, but 50.1%.
In other words, they have the least economic shares and have the most control. The remaining 90 per cent of the shares are in the hands of investors, staff options pools and public shareholders, but they remain minority in key governance matters.
This design is not the first of Anthropic. Silicon Valley has a long history — the Class B stock of Facebook, and Google's super-voting structure are all the same logic. But the design of Anthropic is more extreme: other companies usually reserve 20 to 40 per cent of their voting rights for the founders, and Anthropic locks directly at 50.1 per cent, crossing the threshold of the absolute line of control and not diluted.
Anthropic Class F stock governance structure
One of the questions that must be answered is: why does Anthropic have to work so hard to pull the weight of governance to this extreme?
The answer lies in its history of entrepreneurship.
References
Extension entrance
- Original archive: tobemagic.github.io/ai-magician...
- WeChat Official Account: Computer magician
Auto-fetched and translated by AI Atlas. All rights belong to the original author.