内容来源：https://www.wired.com/story/anthropic-prospectus-leaves-out-some-numbers/

内容总结：

人工智能公司Anthropic于周一提交了备受期待的IPO招股说明书，首次公开披露了公司的业务、战略及风险状况，为其上市做准备。这份名为S-1的文件长达数百页，但存在若干明显遗漏，尤其是部分财务数据。

招股书中写道：“Claude经历了历史上企业采用速度最快的技术应用，商业化于2023年第一季度启动时收入为零，而截至2026年9月30日的三个月收入已达_____美元。”此类留白在文件中出现了数十处。

企业在S-1文件中隐去部分信息并不罕见，美国证券交易委员会的规定对此有一定宽容度，尤其涉及被视为“机密”的信息。但了解企业究竟选择隐去哪些内容仍具参考价值。就Anthropic而言，缺失的细节主要集中在截至9月底的最近一个季度财务数据，以及与其算力投资相关的租赁协议。专家预计，许多细节将在Anthropic正式上市前披露，但目前的缺失或许暗示了投资人工智能的不确定性——各公司正努力确保充足的算力资源。

Anthropic未回应置评请求。

纽约州立大学布法罗分校管理学院会计与法律教授迈克尔·丹布拉在接受《连线》杂志采访时表示，第三季度数据缺失有“实际原因”：9月30日才过去几周，“对于这种规模的公司来说，完成账目结算需要时间”。丹布拉还指出，公司可能因注册声明中的错误陈述而承担法律责任，“因此企业有强烈动机在数据经过充分审核之前不对外公布”。他表示，根据SEC规则，如果Anthropic在11月中旬之后上市，“将需要更新文件并补充第三季度财务数据”。

至于数据中心租赁协议，扬斯敦州立大学会计学教授埃里卡·纽曼表示，企业在“合同附件”中隐去机密细节“并不罕见”，其中可能包括租赁条款。但纽曼也指出，值得审视的是，Anthropic的隐去处理是否会妨碍或增加投资者分析该公司财务义务的难度，以及这些义务是否可能影响公司现金流。

根据S-1文件，Anthropic的许多财务义务与算力相关，算力是“我们业务中最重要的运营成本”，同时也是公司的商业风险，文件从供给不足和过度投资两个角度进行了阐述。（部分AI行业人士一直警告称，芯片短缺可能放缓AI发展进程。）Anthropic的招股书更强调过度投资的风险，称公司需要在前瞻性算力资源上进行投资，这些投资乍看之下可能显得“不负责任地庞大”。

Anthropic解释称，芯片被视为来自多个战略合作伙伴的“单一可互换资源池”，合作伙伴包括亚马逊、谷歌、英伟达和AMD。公司指出，已签订的合同“总承诺支出超过_____美元，其中绝大部分与算力和基础设施相关”。部分合同预计在2030年12月前完成，其他合同可能延续至2045年。

尽管招股书提及了因土地和电力资源有限等因素带来的算力不确定性，但Moor Insights & Strategy首席执行官兼首席分析师帕特里克·穆尔黑德认为，投资者理应获得更多细节。他告诉《连线》：“关键在于逐一描述每个领域的概率和影响。Anthropic欠投资者更多信息，比如这些个别问题的发生概率，量化百分比几率和影响程度。而不是笼统地说‘可能产生重大影响’。”

Anthropic在脚注中确实提供了一些引人注目的估算数据。自2025年12月31日以来，该公司已启动或修订了“算力和基础设施”租赁协议，“未来最低支付义务”合计约4170亿美元。自6月30日以来，其未来支付义务又增加了2450亿美元，均与基础设施协议相关。此外，公司还扩展了“某些现有商业协议”——并注明“尚未最终确定”——新增金额在1250亿至1500亿美元之间。Anthropic表示，若计入具有“最低不可取消付款”的“短期”协议，还有210亿美元。如果全部履行，公司最低将欠约8080亿美元。

追踪芯片行业的Creative Strategies首席执行官兼首席分析师本·巴贾林表示，这很可能只是Anthropic招股书的初稿，他预计更新版本将填补大量空白。

尽管S-1文件中的隐去处理十分常见，丹布拉仍将其描述为“有争议的”。虽然企业可能辩称这是“保护专有信息”，但丹布拉表示，怀疑者指责企业“投机性地”利用隐去处理来“回避披露在政治或社会层面敏感的信息”。不过，丹布拉认为隐去信息可能带来代价。他指出，S-1文件中存在隐去处理的公司，在IPO时更有可能“低估”其股价。丹布拉称，当企业在S-1中隐去信息时，它们通常明白自己与外部投资者之间制造了一道“信息鸿沟”。由于预期投资者会以“怀疑”态度对待这一鸿沟，企业可能倾向于通过设定较低的发行价来“吸引”投资者参与IPO。

目前尚不清楚Anthropic的发行价最终会是多少，也不清楚投资者如何解读该公司S-1文件中的隐去处理。但据巴贾林所言，就目前基于芯片和基础设施获取情况来推算数据而言，整个人工智能行业在某种程度上都是一片“空白”。所以，这确实都充满风险。他说：“所有这些，恰恰说明我们在企业采用AI方面还处于极其早期的阶段，而这个市场每年将超过3万亿美元，且很可能还在持续增长。”

中文翻译：

Anthropic于周一提交了备受期待的IPO招股说明书，在上市前披露了公司的部分业务、战略和风险状况。尽管这份被称为S-1的文件长达数百页，但其中有一些明显的遗漏—— namely，某些财务数据。

“Claude经历了历史上任何技术中最快的企业采用速度，体现在营收从2023年第一季度商业化时的0美元增长至截至2026年9月30日止三个月的_____美元，”数十个此类示例中的第一个这样写道。

公司在S-1文件中隐去某些信息相对常见，美国证券交易委员会的规则对此类做法允许一定程度的宽容，尤其是涉及被视为“机密”的信息时。

尽管如此，了解这些公司究竟决定省略哪些内容仍然是有帮助的。在Anthropic的案例中，缺失的细节主要围绕其最近一个季度（截至9月底）的财务数据，以及与其算力投资相关的租赁协议。尽管专家预计在Anthropic上市前许多细节将会披露，但目前这些信息的缺失可能暗示了投资人工智能的不确定性，因为各公司都在努力确保充足的算力。

Anthropic未回应置评请求。

布法罗大学管理学院会计与法学教授迈克尔·丹布拉告诉《连线》杂志，第三季度数据缺失有“实际原因”：9月30日才过去几周，“为这样规模的公司结账需要时间”。

丹布拉表示，公司还可能因注册声明中的错误陈述而承担责任，“因此存在强烈的动机在数据经过充分审查之前不公布”。他说，SEC规则要求“如果Anthropic在11月中旬之后上市，则需要更新文件中的第三季度财务数据”。

至于数据中心租赁协议，扬斯敦州立大学会计学教授埃里卡·诺伊曼表示，公司从“合同附件”（可能包括租赁条款）中隐去机密细节“并不罕见”。

然而，诺伊曼补充说，值得仔细审视Anthropic的隐去处理是否会阻止或增加投资者分析Anthropic财务义务的难度，以及这些义务是否可能影响公司的现金流。

根据S-1文件，Anthropic的许多财务义务与算力相关，算力是其“业务中最重要的运营成本”。这也是公司的一项业务风险，文件从供给不足和过度投资两个方面进行了阐述。（人工智能行业的部分人士一直警告称，芯片限制可能减缓AI的进展。）

Anthropic的招股说明书更加强调过度投资的风险，指出公司需要对算力资源进行前瞻性投资，而这些投资乍看之下会显得“不负责任地巨大”。

Anthropic解释说，芯片被视为来自“多个战略合作伙伴的单一可替换资源池”，包括亚马逊、谷歌、英伟达和AMD。公司指出，它已签订的合同“总承诺支出超过_____美元，其中绝大部分与算力和基础设施相关”。其中一些合同预计在2030年12月之前完成，而另一些可能延续至2045年。

尽管招股说明书提到了Anthropic因土地和电力资源有限等因素而面临的算力不确定性，但Moor Insights & Strategy首席执行官兼首席分析师帕特里克·穆尔黑德认为，投资者理应获得更多细节。

“关键在于逐一描述每个领域中风险和影响的概率。Anthropic欠投资者更多信息，比如这些个别问题的发生概率，描述百分比概率和影响程度，”他告诉《连线》杂志。“而不是‘可能产生重大影响’。”

Anthropic确实在脚注中提供了一些引人注目的估算。自2025年12月31日以来，它已启动或修订了“算力和基础设施”租赁协议，“未来最低付款义务”总计约4170亿美元。自6月30日以来，其未来付款义务又增加了2450亿美元，全部与基础设施协议相关。它还扩展了“某些现有商业协议”——并附注说明这些协议“尚未最终确定”——追加1250亿至1500亿美元。

Anthropic表示，再加上带有“最低不可取消付款”的“短期”协议，又是210亿美元。总计而言，如果全部履行，它至少将欠约8080亿美元。

追踪芯片行业的Creative Strategies首席执行官兼首席分析师本·巴贾林表示，这可能只是Anthropic招股说明书的初稿，他预计更新版本中将填补大量空白。

尽管S-1文件中的隐去处理很常见，丹布拉将其描述为“有争议的”。虽然公司可能辩称这是“保护专有信息”，但丹布拉说，怀疑者指责他们“投机性地”利用隐去处理来“避免披露政治上或社会上敏感的信息”。

然而，丹布拉认为隐去信息可能是有代价的。他指出，S-1文件中存在隐去处理的公司也更可能在IPO时“低估”其股价。

丹布拉说，当公司在S-1文件中隐去信息时，它们通常明白自己与外部投资者之间制造了“信息差”。由于知道投资者会以“怀疑”态度对待这一差距，公司可能倾向于通过设定较低的初始价格来“吸引”他们参与IPO。

目前尚不清楚Anthropic的开盘价最终会是多少，也不清楚投资者可能如何解读该公司S-1文件中的隐去处理。但据巴贾林称，目前整个AI行业在基于芯片和基础设施获取来核算数据方面，都处于某种形式的“空白”状态。所以，是的，这都是风险。

“所有这一切，”他说，“恰恰表明我们在企业采用AI市场方面还处于极其早期的阶段，而这个市场每年将超过3万亿美元，并且可能还在增长。”

评论

返回顶部

英文来源：

Anthropic filed its hotly anticipated IPO prospectus on Monday, spelling out parts of the company’s business, strategy, and risk profile ahead of its public debut. Though the filing, known as an S-1, is several hundred pages long, it has some notable omissions—namely, certain financial figures.
“Claude has experienced the fastest enterprise adoption of any technology in history, as demonstrated by revenue growing from $0 at commercial launch in the first quarter of 2023 to $_ for the three months ended September 30, 2026,” reads the first of dozens of such examples.
It’s relatively common for companies to redact certain information in their S-1 filings, and Securities and Exchange Commission rules allow for a certain degree of leniency around the practice, particularly when it comes to information considered “confidential.”
Still, it’s helpful to know what, exactly, these companies decide to leave out. In Anthropic’s case, the missing details largely revolve around financials for its most recent quarter, which concluded at the end of September, and lease agreements tied to its compute investments. While experts expect many details to be disclosed before Anthropic goes public, their current absence may hint at the uncertainties of investing in artificial intelligence, as companies work to secure sufficient compute.
Anthropic did not respond to a request for comment.
Michael Dambra, accounting and law professor at University at Buffalo School of Management, tells WIRED that there are “practical reasons” for its third quarter numbers to be absent: September 30 was only a few weeks ago, and “closing the books for a company of this size takes time.”
Companies can also be liable for misstatements in their registration statements, “so there is a strong incentive not to publish numbers before they have been fully vetted,” Dambra says. SEC rules require that “Anthropic would need to update the filing with third-quarter financials” if the company goes public any later than about mid-November, he says.
As for data center lease agreements, Erica Neuman, an accounting professor at Youngstown State University, says it’s “not unusual” for companies to redact confidential details from “contract exhibits,” which could include lease terms.
However, Neuman adds that it’s worth scrutinizing whether Anthropic’s redactions would prevent, or make it more difficult, for investors to analyze Anthropic’s financial obligations and whether they may impact the company’s cash flow.
Many of Anthropic’s financial obligations relate to compute, which is the “most significant operating cost in our business,” according to the S-1. It’s also a business risk for the company, and it’s laid out both in terms of undersupply and overinvestment. (Parts of the AI industry have been warning that chip constraints could slow the progress of AI.)
Anthropic’s prospectus puts more emphasis on the risks of overinvestment, stating that the company needs to make forward investments in compute resources that will seem “irresponsibly large” at first glance.
Chips, Anthropic explains, are approached as a “single, fungible pool from multiple strategic partners,” including Amazon, Google, Nvidia, and AMD. The company notes that it has entered into contracts “with an aggregate committed spend of more than $, substantially all of which is related to compute and infrastructure.” Some of these contracts are set to be completed before December of 2030, while others may extend through 2045.
Though the prospectus touches on Anthropic’s compute uncertainties due to factors like limited land and power resources, Patrick Moorhead, CEO and chief analyst at Moor Insights & Strategy believes investors are owed more detail.
“It’s about characterizing the odds and impact individually across each of the areas. Anthropic owes investors more information, such as a probability of these individual issues, characterizing the percentage chance and the degree of the impact,” he tells WIRED. “Not ‘could materially impact.’”
Anthropic does provide some eye-catching estimates in footnotes. Since December 31, 2025, it has started or amended “compute and infrastructure” leases with “future minimum payment obligations” worth approximately $417 billion in aggregate. Since June 30, it’s added another $245 billion to its future payment obligations, all related to infrastructure agreements. It also expanded “certain existing commercial agreements” that it caveats “have not yet been finalized,” for an additional $125 billion to $150 billion.
Include “short term” agreements with “minimum non-cancelable payments,” Anthropic says, and that’s another $21 billion. In total, it will owe, at minimum, an estimated $808 billion if it follows through on all of this.
Ben Bajarin, chief executive and a principal analyst at Creative Strategies who tracks the chip industry, says that this is likely just the first draft of Anthropic’s prospectus, and that he expects a lot of gaps to be filled in an updated version.
Despite how common S-1 redactions are, Dambra describes them as “controversial.” While companies may argue that they’re “protecting proprietary information,” Dambra says that skeptics accuse them of using redactions “opportunistically” in order to “avoid disclosing politically or socially sensitive information.”
However, Dambra argues that redacting information may come at a cost. He notes that companies with redactions in their S-1s were also more likely to “underprice” their shares at the initial public offering.
When companies redact information in their S-1, Dambra says they typically understand that they’ve created an “information gap” between them and outside investors. Understanding that investors will treat this gap with “suspicion,” companies may feel inclined to “entice” them to participate in the IPO by setting a lower starting price.
It’s unclear what Anthropic’s opening price may end up being, or how investors may be interpreting the redactions in the company’s S-1. But according to Bajarin, the whole AI industry is one form of “blank” or another when it comes to crunching the numbers based on access to chips and infrastructure right now. So, yes, it’s all a risk.
“All of this,” he says, “just shows how absurdly early we are in enterprise adoption of an AI market that will be more than $3 trillion yearly and likely growing.”
Comments
Back to top