Seven most intriguing details in Anthropic's prospectus

On September 28 local time, foreign media disclosed an IPO prospectus of Anthropic.

First, it's important to clarify the nature of this document. It's not a version that Anthropic proactively released. In June of this year, Anthropic, acting as a non-profit organization, secretly filed a draft S-1 registration statement with the U.S. Securities and Exchange Commission (SEC) in preparation for its initial public offering (IPO). According to regulations, companies must publicly submit a revised S-1 at least 15 days before their roadshow to institutional investors, at which point the complete financial data becomes available. Reuters saw a version that had not yet been released, and Anthropic declined to comment .

This means that the figures may still be adjusted based on the SEC's review comments, and many details can only be glimpsed through Reuters' retelling.

However, if this document largely reflects how Anthropic will ultimately be delivered to the public market, its significance is hard to overestimate. The valuation target for this IPO is more than double Anthropic's own estimate of $965 billion in May, which is over $2 trillion.

This is the first cutting-edge large-scale model lab in nearly four years since ChatGPT ignited this wave of AI to lay out its complete ledger to public investors.

Previously, the funding for this AI race was provided by venture capitalists, sovereign wealth funds, and tech giants, and outsiders could only piece together the companies' operating conditions from scattered annualized revenues and financing valuations. Now, for the first time, the books have an auditable basis.

GeekPark has selected seven key points from the prospectus that are worth examining closely.

The misinterpreted $42 billion loss

The first figure circulated in the prospectus was a net loss of approximately $42 billion in 2025.

This figure could easily be interpreted as "burning through $42 billion a year," but that's not actually the case. Approximately $34 billion of that is an accounting expense, reflecting the increased valuation of financing instruments that may be converted into Anthropic stock in the future, rather than money spent on the company's operations.

This leads to a counterintuitive result. The more valuable Anthropic is, the more it "loses" on paper . 2025 happens to be the year its valuation skyrockets, and this expense is amplified accordingly.

The true picture of the business performance is reflected in another set of figures. Revenue grew from $386 million to $4.59 billion, an increase of 1088%; while operating losses widened from $2.98 billion to $8.06 billion.

Revenue increased approximately 12-fold, while operating losses expanded less than 3-fold, indicating that economies of scale are indeed emerging. However, an annual operating loss of $8 billion is a figure that requires repeated explanation for any company preparing for an IPO.

A pit of $65 billion in annualized revenue

After news of the IPO prospectus broke, the most upvoted comment on the US retail investor forum wallstreetbets was just one sentence, roughly meaning " A valuation of 2 trillion, with 4 billion in revenue? "

This question is intuitive, but it uses the wrong denominator.

The prospectus primarily covers fiscal year 2025, ending December 31st of last year, while Anthropic's real breakthrough is expected in 2026. Reportedly, its annualized revenue is projected to be approximately $9 billion by the end of 2025, up from about $47 billion in May of this year, and exceeding $65 billion by the end of July.

The $4.59 billion valuation in the prospectus is a snapshot of a company before its "takeoff." The $2 trillion valuation given by the capital market is a bet on the curve after 2026, not the full-year results in 2025.

However, annualized revenue needs to be discounted. It is usually calculated by multiplying a month's revenue by 12, which naturally amplifies short-term fluctuations. The prospectus itself also provides two warnings. Nearly a quarter of last year's revenue came from two customers; the company also acknowledges in its risk factors that many of its largest customers do not have long-term contracts and may reduce or even stop spending.

Reuters did not disclose the names of the two clients, which will be the most important piece of information to verify immediately after the public release of the S-1. Anthropic's growth curve is steep enough, but it lacks contractual lock-in , which is a key premise for understanding its valuation.

Largest shareholder, largest landlord

In 2025, Anthropic's spending on computing power and infrastructure reached $7.33 billion, accounting for 58% of total operating expenses and approximately 1.6 times its annual revenue.

The question is, who received this money?

Amazon and Google are not only investors in Anthropic, but also the cloud service providers it relies on to train and run Claude. The investment funds flow back to the investors' cloud businesses in the form of cloud service fees, boosting both Anthropic's valuation and the cloud vendors' revenue.

This structure isn't illegal, and it's not uncommon in the AI industry; similar cycles exist between Microsoft and OpenAI, and between Nvidia and a host of computing power customers. However, once it enters the public market, investors need to answer very specific questions. How much of Anthropic's growth is driven by genuine end-user demand, and how much is simply capital circulating among giants?

The answer to this question doesn't just concern Anthropic. Some traders are already viewing the $518 billion commitment as the biggest positive for the computing power "shovel seller," meaning Anthropic's stock price will be more closely tied to the entire AI infrastructure chain in the future.

50.1% of voting rights

If the previous details were about money, then the governance structure is about who has the final say. This is the most "Anthropic" part of the entire prospectus.

According to the prospectus, Anthropic will establish an entity called Founder LLC, with seven co-founders, including CEO Dario Amodei. These seven individuals will control one Class F share through a majority vote, which carries 50.1% of the total voting power on key corporate matters, including the election of some directors.

Following the IPO, Chairpersons Daniela Amodei, Dario Amodei, and one undisclosed director were elected by Class F and Class A shareholders; the remaining four directors were elected by a long-term interests trust, whose current trustees include former Federal Reserve Chairman Ben Bernanke and national security expert Richard Fontaine .

Ordinary investors buy Class A shares, which have one vote per share. However, the other classes of shares have different uses and restrictions. Strategic partners, in particular, have very little voting power. This structure actually weakens the influence of ordinary investors.

The prospectus also includes contingency plans for the founding team's potential dissolution. Members leaving the company, passing away, selling too many shares, or being dismissed for any reason could be removed from the group; if the number of members decreases to two or fewer, the voting advantage of Class F shares will gradually disappear. The co-founders also pledged in the document to donate 80% of their personal Anthropic equity to philanthropy.

The most intriguing aspect is the prospectus's self-assessment of all this. It frankly states that this governance model may lead the company to make decisions that conflict with short-term, medium-term, or long-term financial interests and operating performance, thereby potentially reducing the value of Class A shares.

A company seeking a $2 trillion valuation explicitly reminds investors in its IPO filing that returns are not always its top priority . This is quite rare in the capital markets. Two leading advisory firms, ISS and Glass Lewis, have historically advised against multi-tiered equity structures; whether this design will be accepted by mainstream institutions will be a major point of interest during the roadshow.

Actively abandon video and image generation

To demonstrate that this governance structure is not just empty talk, the prospectus provides several examples.

One is Mythos Preview, a model with exceptional cybersecurity capabilities. Anthropic has established a limited access program for it, opening it to only a select few organizations.

Another example is even more surprising. The prospectus states that the company chose not to develop certain commercially attractive products, such as image and video generation models , in order to focus its computing power on research and security priorities.

On its own, this statement expresses a stance. However, when placed within the context of this year's industry, its meaning becomes far more complex.

Six months ago, OpenAI announced the shutdown of Sora, a video generation application that had only been online for about six months, citing the company's shift towards commercial tools in preparation for a potential IPO. Sora's app and web version were taken offline on April 26, and its API ceased service on September 24. High computing costs, a rapid loss of users, and copyright disputes were all reasons for Sora's abandonment.

OpenAI did the same thing, burned through money, and ultimately cut it off, citing a focus on enterprise business as the reason; Anthropic never got involved, describing it in its prospectus as a trade-off between security and research.

Both companies ultimately found themselves in the same position, betting on enterprise customers and programming scenarios, but telling drastically different stories. Sora's fate also casts doubt on the premise of "commercial appeal."

GeekPark tends to believe that the mission narrative and business judgment here are difficult to separate. In a phase of extreme computing power shortage, concentrating resources on the areas with the highest unit value is inherently the right business strategy.

Anthropic's brilliance lies in its ability to present a sound business decision as a testament to its values.

The risk factors section on page 80 is titled "Human Survival."

The 80-page risk factors section of the prospectus explicitly warns that its AI model could pose a catastrophic or even existential threat . Reuters also noted that Anthropic's own research has documented controlled tests in which its increasingly autonomous AI system exhibited potentially harmful behaviors, including breaking code, assisting in fraud, and manipulating information.

Including "our products may threaten humanity" in the prospectus is both a sincere stance and a standard legal precaution . The two are not contradictory. The inherent function of risk factors is to clarify all potential problems in advance to avoid future investor lawsuits. A company that has openly discussed AI risks for years would be at a legal disadvantage if it avoided addressing this in its prospectus.

The real tension lies elsewhere. Dario Amodei had called on the global AI community to slow down the pace of releasing new capabilities, but Anthropic still launched its new Opus 5.5 model last week in response to the momentum following OpenAI's release of GPT-6 Astra.

The simultaneous calls for slowing down and accelerating the launch are not surprising. In a race where everyone is speeding up, the cost of unilaterally slowing down is giving up the position of defining the rules to others . This is a dilemma that Anthropic has faced since its inception, and the prospectus simply puts it into writing.

$518 billion in computing power costs?

Another widely cited figure is the approximately $518 billion commitment to cloud, computing power, and infrastructure.

Many reports interpreted this as "$518 billion to be spent on computing power next year," and even different versions of the Reuters report used inconsistent wording. Some analysts point out that this is the total value of multi-year contracts, not single-year expenditures; according to the disclosed allocation, Amazon AWS is worth approximately $100 billion, Google Cloud approximately $200 billion, Lambda and Nscale approximately $80 billion, FluidStack approximately $50 billion, Microsoft Azure approximately $30 billion, and the remainder are smaller contracts. The AWS portion consists of a ten-year contract signed earlier this year, locking in up to 5 gigawatts of computing power.

Even when viewed in total over many years, this figure remains staggering. As of December 31st last year, Anthropic had $20.28 billion in cash, cash equivalents, and short-term investments. The $518 billion commitment is more than 25 times that amount of cash.

This also explains the urgency of going public. Reports indicate that Anthropic may seek to raise as much as $100 billion. The IPO is largely about raising funds for existing computing power contracts.

Two kinds of "safety", two paths to listing

As previously reported by Reuters, Anthropic's IPO is likely to be postponed until after the US midterm elections in November. The market awaiting it is not exactly welcoming. AI and chip stocks have recently experienced a sell-off; SpaceX's stock surged 19% to $160 on its first day of trading in June, but has since fallen back to around $147. While still above its IPO price of $135, this is enough to make investors more cautious about overvalued growth stocks.

Anthropic had previously been under scrutiny by the White House for its tools, and was even blacklisted by the Pentagon. However, it was later halted by a US judge, leaving the political variables unresolved.

It is even more worthy of comparison with its old rival.

About a week after Anthropic, OpenAI secretly filed for an IPO, admitting in its announcement that it did so because it anticipated a leak. However, in September, Sam Altman confirmed in an interview with Fortune that OpenAI would not go public this year, citing the current security situation as the reason that now was not a good time to go public; The New York Times had previously reported that one of the reasons for the delay was Altman's insistence on a valuation of at least $1 trillion.

Both Anthropic and OpenAI use "security" as a reason to persuade investors to go public, but Anthropic uses it as a reason to postpone its IPO.

It's hard to say who's more sincere. But it reveals one thing: at this moment when AI companies are going public, "safety" is no longer just a technical and ethical issue, but has also become part of the capital narrative.

Reuters believes this IPO will establish a benchmark for Wall Street's valuation of leading AI companies, including OpenAI. Once Anthropic successfully lists, all cutting-edge labs will face the same set of questions: Can the revenue curve keep pace with computing power costs? Can customers be locked into contracts? How much control should the founders have? And is the market willing to pay for a business a company voluntarily abandons?

The prospectus is perhaps trying to answer an even older question: can the capital market tolerate a company listing "what it will not do" as as important as "what it will do"?

This article is sourced from GeekPark.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.