Anthropic Files Prospectus, IPO Aims to Raise $100 Billion

By Shanshang | Author: He Jian | Editor: Jiang Jiao

After half a year of media reports, document leaks, and proactive communication with investors, nearly all of Anthropic's key financial data has been disclosed piece by piece.

We learned early on about Anthropic's revenue figures for the first half of the year, and we also know that it achieved quarterly profitability in this year's second quarter. Even its commitment to more than $500 billion in computing spending over the next decade with six cloud service providers—about 80% of which is non-cancelable—was exposed long ago.

This makes the prospectus Anthropic just made public last night seem somewhat outdated. Even prior news reports reflect the company's latest operating conditions better than the prospectus does. The financial data in the prospectus is only updated through the first half of this year, while the media had long since reported that Anthropic's revenue growth slowed sharply in the third quarter, which in turn led investors to lower their valuation expectations for Anthropic.

Regardless, this nearly 400-page prospectus is the beginning of Anthropic's entry into the capital markets. Anthropic's previous funding round was as high as $65 billion, far exceeding the fundraising scale of the vast majority of IPOs. Now, its IPO fundraising target is $100 billion, which could set a record for the largest IPO in history.

Just like the internet listing wave years ago, more large-model companies are waiting to go public after Anthropic. They have similar products and services, similar business models and revenue structures, and are likewise dependent on massive computing investment.

Anthropic's listing may just be the beginning of the large-model frenzy—hopefully it will not be the peak.

Quarterly revenue of $10 billion, top five customers account for 25%

Judging purely from the revenue figures of the past few years, now is indeed the best time for Anthropic to sprint toward a listing.

In 2023, Anthropic's full-year revenue was only $41 million, rising to $386 million in 2024 and $4.585 billion in 2025. Such growth is already staggering, but by the first half of this year, Anthropic's half-year revenue had already reached $16.286 billion, up nearly 13-fold year over year and more than three times the sum of its revenue over the past three years.

In particular, in this year's second quarter, Anthropic's single-quarter revenue reached $11.555 billion, up 144% from the previous quarter and nearly 14-fold year over year. By comparison, Salesforce, the world's largest CRM software company, has quarterly revenue of only $11.3 billion.

Anthropic prospectus

Unlike other companies' complex revenue structures, Anthropic's astonishing growth still relies on the simplest business in the large-model industry: tokens. Similar to China's already-listed Zhipu and MiniMax, Anthropic's revenue also comes mainly from tokens—what they call "paying for intelligence."

Specifically, there are two charging methods. One is usage-based charging, where customers call the Claude model through APIs and other means and pay according to the tokens actually consumed. The other is subscription services, including Claude Pro, Max, and enterprise-oriented subscription plans.

In the first half of this year, Anthropic's usage-based revenue reached $13.064 billion, up about 12-fold year over year, accounting for 80.2% of total revenue. Subscription revenue reached $3.222 billion, up nearly 17-fold year over year, with its share rising from 15.5% in the same period last year to 19.8%.

Usage-based charging is mainly aimed at enterprise customers using the model through APIs, while subscription services cover more individual users. Anthropic expects in its prospectus that for a considerable period in the future, usage-based charging will still contribute the vast majority of revenue.

The rapid growth of enterprise customers also confirms this. As of the end of this year's third quarter, more than 11,900 customers had contributed at least $100,000 in revenue to Anthropic over the past 12 months, compared with only more than 950 in the same period last year. In addition, more than 1,900 customers contributed more than $1 million annually, and about 250 customers contributed more than $10 million.

Anthropic also disclosed that as of the end of this year's third quarter, its net revenue retention rate was as high as 900%. This means that the revenue contributed each quarter by the same group of enterprise customers has now reached nine times the level of the same period last year. They emphasized in the prospectus that their penetration in large enterprises is still at an early stage. As of the end of this year's third quarter, Anthropic said revenue from its top five customers all hit record highs, with the top five customers together accounting for about 25% of its revenue.

Anthropic's own applications such as Claude Code have helped considerably. The prospectus shows that as of the end of this year's third quarter, among customers contributing at least $100,000 in revenue over the past 12 months, more than half also used both the Claude model platform and Anthropic's own applications such as Claude Code and Cowork. At the end of last year, this proportion was only about 30%.

However, Anthropic's revenue growth has already begun to slow. According to investor materials previously disclosed by the media, Anthropic's third-quarter revenue was approximately $16.8 billion, up about 45% quarter over quarter, far below the 144% increase in the second quarter.

The prospectus has not yet updated the detailed financial data for the third quarter. Anthropic previously said that it tightened controls on distillation in the third quarter, which led to a reduction in some revenue. The prospectus discloses that as of the end of this year's third quarter, this portion of revenue accounted for 5% to 7% of Anthropic's revenue.

Finally starting to make money, but not enough to fill the future

In stark contrast to the rapid revenue growth, Anthropic's losses are equally staggering.

The prospectus shows that in 2025, Anthropic's full-year net loss reached $41.965 billion, and in the first half of 2026 it was as high as $80.748 billion, more than five times the $12.969 billion in the same period last year. In just half a year, Anthropic's net loss was nearly 18 times its full-year revenue last year.

However, this huge loss mainly comes from changes in the book value of financing instruments. In the first half of this year, Anthropic recognized a loss of $72.388 billion due to changes in the fair value of financing instruments such as convertible notes, derivative liabilities and SAFEs, accounting for nearly 90% of its net loss in the same period.

Excluding these book accounting effects, Anthropic's operating condition has in fact improved significantly. In 2025, Anthropic's full-year operating loss was as high as $8.060 billion, equivalent to nearly twice its revenue in the same period. In the first half of this year, this figure has narrowed to $693 million, a decrease of about two-thirds from $2.135 billion in the same period last year.

In particular, in the second quarter of this year, Anthropic achieved quarterly operating profitability for the first time, with a profit of $974 million in the quarter and an operating margin of 8.4%. In the same period last year, Anthropic's single-quarter loss was still $1.2 billion, with an operating margin of negative 152.5%. But compared with traditional software giants, Anthropic still has a gap in profitability. Salesforce's operating margin in the second fiscal quarter of this year was about 20.5%.

In the past, the reason why large model companies had persistently high losses was often related to high computing power costs. Take Zhipu as an example. In 2025, Zhipu's full-year revenue was only 724 million yuan, but R&D expenses were as high as 3.180 billion yuan, of which about 2.283 billion yuan was used for computing power. By the first half of this year, Zhipu's revenue increased to 954 million yuan, R&D expenses still reached 2.131 billion yuan, and its net loss was 2.072 billion yuan.

Anthropic is the same. In 2023, Anthropic's full-year revenue was only $41 million, but computing power and infrastructure costs reached $433 million, equivalent to more than ten times its revenue. In 2024, computing power spending further increased to $2.522 billion, and in 2025 it reached $7.326 billion.

In the first half of this year, Anthropic paid another $8.945 billion for computing power and infrastructure, an increase of $6.611 billion from the same period last year, a growth rate of 283%. Spending in half a year has exceeded the whole of last year, accounting for more than half of the company's total operating expenses.

However, Anthropic's revenue growth has far exceeded the growth of computing power costs. In the first half of this year, Anthropic's revenue grew nearly 13 times year over year, while computing power grew only about 2.8 times. Computing power as a proportion of revenue fell all the way from 201% in the same period last year to 55%.

Anthropic prospectus

More critically, this computing power is often locked in in advance, and costs do not grow proportionally as revenue climbs. Anthropic explains in the prospectus that they usually purchase computing power 12 to 18 months in advance, and these costs are basically fixed in the short term and will not change with demand during the period. In other words, as long as revenue grows fast enough, Anthropic can gradually dilute computing power costs and increase profits.

However, the computing power Anthropic has locked in for the next decade or more is far more staggering than the money already spent.

The prospectus discloses that as of the end of June this year, Anthropic's signed fixed or minimum purchase commitments reached $265.6 billion, with contract terms of up to 20 years. Since then, the company has signed or expanded computing power agreements with a number of tech giants. According to data previously disclosed by the media, Anthropic's computing power and infrastructure spending commitments for the next decade have exceeded $500 billion, of which about 80% is irrevocable.

With Google and Microsoft alone, Anthropic has committed to paying at least $114.5 billion and $44.9 billion respectively by 2033. A five-year equipment leasing contract related to Broadcom also has total payments of about $125.2 billion. As of the end of June this year, Anthropic held only $131.7 billion in cash and short-term investments, and had raised about $141.1 billion in cumulative financing.

These cloud service giants sell computing power to Anthropic while also helping it sell Claude through their cloud platforms and collecting channel fees from it.

In the first half of this year, Anthropic's sales, marketing and partner expenses reached $3.168 billion, compared with only $120 million in the same period last year, a year-on-year increase of 2,540%. Of that, the fees paid to channel partners such as cloud platforms increased by about $2.7 billion.

Currently, AWS, Google Cloud and Microsoft Azure have become the main sales channels for Claude. In 2025, revenue obtained through these cloud platforms accounted for 47% of Anthropic's full-year revenue, and further rose to 48% in the first half of this year.

Who is in charge of Anthropic?

U.S. large-model companies are always eager to talk about their nonprofit visions, and the best known of these is OpenAI. Anthropic is the same. Although this company is already the most heavily funded and highest-valued company in commercial history, it particularly emphasizes in its prospectus that "we are first and foremost a research laboratory," with the goal of "responsibly developing and maintaining advanced AI for the long-term benefit of humanity."

To this end, they established an extremely complex corporate governance structure. In addition to the conventional corporate board structure, Anthropic also specifically established an independent institution called the Long-Term Benefit Trust (LTBT), composed of five trustees with professional backgrounds in fields such as AI safety, national security and public policy.

The LTBT holds Class T shares and has the right to elect and remove a majority of Anthropic's board members. However, the trust itself does not participate in the company's day-to-day operations and has no right to directly decide the company's business strategy.

At the same time, Anthropic's seven co-founders will jointly establish a limited liability company (Founders LLC), which holds special Class F shares. Class F shares have no economic rights whatsoever, yet carry 50.1% of the company's voting power.

Anthropic equity structure design

Under the design in the prospectus, the Long-Term Benefit Trust and the founders jointly control the right to elect all board seats. Even if Anthropic goes public, ordinary investors will find it difficult to influence major company decisions through shareholder votes. Anthropic warns investors in its prospectus that this governance structure may cause the company to make decisions that are not in line with shareholders' economic interests.

Anthropic's seven founders each hold approximately 230 million to 250 million Class A ordinary shares, with fairly similar shareholding amounts. Among them, CEO Dario Amodei holds about 245 million shares, and President Daniela Amodei holds about 241 million shares, with the seven holding a total of about 1.69 billion shares.

Based on Anthropic's current valuation of $2 trillion, the seven founders' equity is worth a total of about $200 billion, or about $28.5 billion per person on average. They promised in the prospectus to devote 80% of their personal equity to philanthropy, but these promises can all be revoked and are not legally binding.

More interestingly, even if the founders continue to reduce their holdings in the future, it will not affect their control over the company. Under the arrangement in the prospectus, as long as they retain 2% of their personal shareholdings at the time of the IPO, the founders can remain in Founders LLC and continue to hold 50.1% of the company's voting power.

Anthropic has also set aside a valuable block of shares for public welfare. In October this year, the company announced the establishment of a Public Interest Fund and allocated about 5% of its shares to it. If the future share price rises 50% above the previous target, the board can decide to issue additional shares to the fund amounting to up to 2% of the company's total shares.

But this charitable plan has also drawn some skepticism. The so-called Public Interest Fund is in fact a limited liability company wholly owned by Anthropic, not an independent charitable foundation, and the use of funds and the transparency of the fund remain unclear.

Some critics of Anthropic often argue that the company is too hypocritical. On the one hand, it constantly accuses other AI companies of distilling Claude. On the other hand, Anthropic itself privately buys large quantities of paper books, dismantles, scans and then destroys them regardless of the consequences.

Previously, Anthropic was hit with a class-action lawsuit by authors over alleged use of pirated books to train its models. The prospectus discloses that Anthropic agreed in September 2025 to pay $1.5 billion in settlement, equivalent to one-third of Anthropic's revenue that year.