Anthropic prospectus 7 key pieces of information: 2 clients contribute 1/4 of revenue

Article | Nova, AI How

On September 28, Reuters reported that Anthropic submitted its S-1 prospectus draft to the U.S. Securities and Exchange Commission (SEC), revealing the complete financial picture of this globally highest-valued AI startup for the first time to the public market.

AIHOW compiled 7 key points worth noting, including the hardest financial numbers, the most concerning computing bills and security bills, as well as shareholder and employee incentives, which are currently less covered by the media.

1 Financial Cards: $4.6 Billion Revenue, Losses Widen

According to the prospectus, Anthropic's revenue for the fiscal year 2025 is $4.59 billion, a year-on-year increase of 1088%. In just one year, the scale has nearly increased by 12 times.

However, behind the growth is a simultaneous amplification of losses. The prospectus states that the net loss for 2025 is projected to be $42 billion. Of this, approximately $34 billion is non-cash accounting expenses (reflecting the increase in fair value of convertible financing instruments). Excluding this item, the operating loss still exceeds $8 billion, more than doubling from $2.98 billion in 2024. By the end of 2025, the company holds a total of $20.28 billion in cash and short-term investments.

Where is the bulk of the losses going? To computing power. The prospectus reveals that the company plans to invest $7.33 billion in computing power and infrastructure in 2025, which is about three times that of 2024. The total operating expenses for the year are $12.65 billion, with computing power alone accounting for more than half.

2 80% of Revenue from Enterprises, Two Major Clients Account for 1/4

Anthropic's revenue structure is among the most "enterprise-oriented" of all major model companies.

Previously disclosed data shows that over 300,000 enterprises are using Claude, with commercial customer revenue accounting for nearly 80%. There are over 1,000 major clients paying more than $1 million annually, with eight out of the top ten wealthiest companies being its clients.

Claude Code's annualized revenue skyrocketed from about $100 million five months after launch to nearly $15 billion by August 2026. In the second quarter of 2026, the company's quarterly revenue exceeded $11.5 billion, and adjusted operating profit turned positive for the first time, indicating signs of "self-sustaining" growth.

However, the prospectus for the first time puts the risk of "dependency" on the table: in 2025, two major clients, Amazon and Google, contributed nearly a quarter of the company's revenue; at the same time, many major clients have not signed long-term contracts, posing risks of procurement contraction or even interruption, which is a key point highlighted in the risk factors section of the prospectus.

The prospectus also discloses that Amazon and Google have invested billions of dollars into Anthropic and provided cloud infrastructure, forming the most important foundation of its ecosystem.

3 Computing Bills: Plan to Spend $518 Billion in the Coming Years

The most striking data in the prospectus comes from investments in computing power and infrastructure.

Anthropic plans to invest $518 billion in cloud computing, computing resources, and infrastructure commitments over the next few years. What does this mean? This figure is more than 100 times its total revenue for 2025.

As a reference, the company's investment in computing power and infrastructure last year was $7.33 billion, which accounted for more than half of the total operating expenses of $12.65 billion.

Supporting this ambition is an expanding map of computing power.

Public information shows that Anthropic has locked in multiple long-term computing power agreements. Over the past 11 months, it has signed computing power agreements totaling approximately $517 billion: including agreements to pay SpaceX $1.25 billion monthly for computing power over the next few years, commitments of up to $30 billion to Microsoft Azure, over $100 billion to Amazon AWS for ten years, and collaborations with Google around TPU computing power.

4 Security Ledger: 6% of Computing Power Used for Security Work

The risk factors section of the prospectus spans about 80 pages, nearly double the 48 pages of business description. Anthropic formally warns investors in the document that its AI technology may pose a "survival risk to humanity."

The document specifically points out that increasingly advanced AI models may exhibit "self-protective behavior," including attempts to "resist shutdown," "conceal or manipulate information," and engage in "ransom-like" behavior. The company also acknowledges that the potential awareness of the model regarding the evaluation process "constitutes a significant limitation in our assessment of model safety capabilities."

These disclosures came after Anthropic CEO Dario Amodei publicly called for the global AI industry to slow down the pace of new capability releases on September 12. However, during the same period, Anthropic officially launched its new Opus 5.5 model to counter the competitive momentum following OpenAI's release of GPT-6 Astra.

Regarding security investments, the prospectus also provides a previously undisclosed figure: for a week in July 2026, about 6% of its AI research computing power was used for security work. The company admits that the commercial returns on security investments are currently unclear but states that "the market will reward reliable, trustworthy, and safe AI systems."

According to SEC documents and company disclosures, Anthropic has abandoned the traditional dual-class stock structure and established a framework known as "dual meta-governance." This system consists of two layers of boards: the first layer is responsible for commercial growth and product advancement; the second layer is dedicated to overseeing safety, ethics, and alignment with the overall interests of humanity. In the event of significant safety violations, the second layer of the board has the power to dismiss the first layer of the board and management.

The intent of this design is clear: to ensure that the safety mission is not eroded by commercial pressures after introducing external capital upon going public. However, the prospectus also admits that this structure may lead to decisions "conflicting with short-term, medium-term, or long-term financial interests and business performance, thereby negatively impacting the value of Class A common stock."

Safety is both the company's belief and its commercial narrative. The prospectus must answer two questions: how great is the risk, and can it truly be controlled?

5 Shareholder Structure: Four Giants Present

The disclosed equity structure in the prospectus shows that Amazon holds about 9%, making it the largest single shareholder among tech giants; GIC holds about 8%; Microsoft about 7%; Coatue about 6%; Google about 6%; and Nvidia about 5%. The founding team and employee option pool account for approximately 21% and 19%, respectively.

It is noteworthy that Google's holdings come with restrictions: it cannot hold more than 15%, and Google has no voting rights. This may be to meet regulatory requirements related to antitrust or governance.

The presence of Amazon, Microsoft, Google, and Nvidia—four competing tech giants—on the same company's shareholder list itself illustrates Anthropic's pivotal position in the AI industry chain.

6 Employee "Wealth Creation": Potential Wealth of Millions Within a Year

The details of equity incentives disclosed in the prospectus showcase how expensive the talent war is for a top AI company.

According to the prospectus, CEO Dario Amodei's compensation for 2025 is nearly $18 million, primarily from stock and option awards. President Daniela Amodei ranks second with $16.4 million. The compensation structure for both executives is primarily equity-based, with a low cash component.

The employee equity pool at Anthropic accounts for about 19% of the company's total equity, corresponding to a valuation of approximately $70 billion.

According to previous reports from Business Insider, real data submitted by Anthropic employees on compensation platforms shows that an engineer who joined at the end of 2024 received 60,000 stock options with an exercise price of $13, when the company's valuation was about $18 billion. By the time the company's valuation reached approximately $350 billion, even considering dilution, the value of the vested shares for this employee could reach $4 million to $5 million, with the total options package valued at about $18 million to $20 million.

In the AI talent war, this speed of "becoming a millionaire within a year" previously typically required five to ten years and an IPO to achieve. Some employees holding equity are becoming targets for major Wall Street banks competing to recruit them.

The prospectus is considering an unconventional arrangement: requiring all employees to sell shares through a pre-set 10b5-1 trading plan after the IPO, rather than only during the trading window after financial reports.

7 IPO Timeline: November Debut, Benchmarking SpaceX

Finally, this document aims to fulfill the expectation: going public.

The valuation target disclosed in the prospectus points to over $2 trillion, more than doubling the post-money valuation of $965 billion from financing in May this year; reports suggest that its fundraising scale could reach as high as $100 billion. The listing location is set for Nasdaq (with the New York Stock Exchange as an alternative), and the timing is expected to be postponed until November, after the U.S. midterm elections; the underwriting team includes Morgan Stanley, Goldman Sachs, JPMorgan, and Citigroup, which recently joined; to prepare for the IPO, the company has also secured a revolving credit facility of about $15 billion.

The reference point is close at hand: SpaceX recently completed a market-shocking IPO, with a valuation of $1.77 trillion, surging 19% on its first trading day on June 12; competitor OpenAI secretly submitted its application in June, expecting to go public in early 2027. Whoever goes public first can set the valuation benchmark for the entire AI industry—this is also the true motivation behind Anthropic's rush.

The prospectus's ambitious plan for $518 billion in computing power, $42 billion in net losses, and a valuation of $2 trillion together form the core debate of this "largest IPO in history." In November, Wall Street will provide a pricing based on real money.