Anthropic presented a book of stock-taking: income grew 12 times to $4.6 billion, and the risk component warned that it was "a threat to human survival."

Author: Baldron, see you on Wall Street

Anthropic disclosed to the outside world his first open-equity book, which outlined an artificial intelligence company that tried to access capital markets in astronomical terms and admitted that its core technology could pose "a catastrophic or even existential risk" to humans.

On 28 September, according to a book of admissions obtained by Reuters,Anthropic grew 12 times to nearly $4.6 billion in 2025, while the operating deficit increased from $2.98 billion in 2024 to $8.06 billion over the same period。

CompanyExpenditure on capacity and infrastructure last year amounted to $7.33 billion, a threefold increase from 2024, accounting for more than half of total operating expenditure of $12.65 billion.

The book of admissions shows thatAnthropic plans to invest $51.8 billion in cloud computing, computing and infrastructure over the next few yearsThe actual voice of ordinary shareholders will be significantly limited after listing.The seven co-founders will hold 50.1 per cent of voting rights on key matters through a new entity called the Foundation LLC.

The listing plan will further test the market ' s resilience to the investment in AI at a time when both the AI and chip units have recently been sold out.

As mentioned earlier in Wall Street, Anthropic's public fund-raising period could be postponed until after the mid-November elections in the United States.

The valuation sword is $2 trillion, twice as much as it was expected six months ago.

In May of this year alone, the company’s own estimate for the IPO target was $965 billion. The valuation is expected to more than double in just a few months.

Of the nearly $42 billion in net losses, approximately $34 billion represented non-cash accounting costs and reflected an increase in the fair value of instruments that could be converted to equity under previous financing arrangements rather than operating cash outflows.As at 31 December 2025, companies held cash, cash equivalents and short-term investments totalling $20.28 billion.

The book also revealed a number of commercial risks:Nearly one quarter of the operations were received from two clients, and most large clients did not have long-term contracts, which could be reduced or discontinued at any time.

If IPO were to go on this trip, it would be one of the best-performing listings in the United States market since 2021.

The reference system may be SpaceX’s June IPO, where the company’s first-day stock price rose by 19 to 160 dollars under the flag of Mask, but has now fallen to about $147, making investors cautious about high valuations of high-growth companies.

Risk warning: 80 pages of book calls describing the AI threat

In the main body of this 261-page book, the risk factor section is approximately 80 pages long, almost twice as large as 48 pages in the business description section.In contrast, the risk portion of SpaceX ' s 277-page equity book is only about 38 pages.

Anthropic clearly warns in his paper that he will not be able to do so.Its AI model may demonstrate "self-protection behaviour", including "resistance of shut-downs", "hidden or manipulated information" and "similar blackmail," noting that "the development and extension of highly advanced models may further increase the risk of harm caused by models".

The company also admitted that the model might develop unexpected capabilities in the course of training, which "may not be discovered until the model is deployed and a major security event has been triggered".

Ivan Hubinger, a company safety researcher, estimates that the government is not in a position to do so.AI is more than 10 per cent likely to cause human deaths in the next decade.

Nevertheless, the actual size of Anthropic ' s input in the area of security was not disclosed in the document. The company indicated that, for example, some 6 per cent of its AI Institute ' s computing power was spent on safety work in one week of this July.

While acknowledging that the commercial returns on safe investments are not yet clear, the company states:

The market will be rewarded by reliable, trustworthy and secure AI systems.

Governance structure: Team of founders led and ordinary investor voice limited

Anthropic designed a unique corporate governance architecture to strike a balance between business interests and the AI security mission.

The company will continue to operate as a Delaware Public Interest (PBC) while creating the "Funder LLC" entity, consisting of seven co-founders. This entity holds F-stocks and has 50.1 per cent of total voting rights in key corporate matters, including major issues such as selected board elections.

General A shares sold to ordinary investors have only one vote per share, but under the above-mentioned structure the real impact will be significantly limited.It was also admitted that this structure could lead to some decisions "contrary to short-, medium- or long-term financial interests and operational performance, with a negative impact on the value of generic A units".

Among the seven founders, CEO Dario Amodei was at the core with his sister, Daniela Amodei, president and chairman of the company.

In 2025, Dario’s remuneration was about $18 million and Daniela’s $16.4 million, both of which were based on equity and options incentives. Both individuals and other co-founders have pledged to spend 80 per cent of their personal Anthropic shares on philanthropy.

Anthropic's board of directors, with four additional members to be elected by the company's long-term interest trust, is currently held by Ben Bernanke, former Federal Reserve Chairman, and Richard Fontaine, national security expert.

trillions of dollars in AI arms race: OpenAI is the biggest rival.

Anthropic is facing intense competition in the AI field, with OpenAI as its chief counterpart.

Both companies have a full-scale presence on corporate clients, top talent, and the influence of Washington’s policies. According to media reports, OpenAI submitted an IPO application secretly in June this year and is expected to be listed by early 2027.

Anthropic was created about five years ago by a group of researchers who left OpenAI because of differences in corporate governance and AI security concepts, and published the first large-language model in March 2023, which directly competes with OpenAI. In addition, it competed with SpaceX for XAI, Alphabet and Meta in the area of AI infrastructure.

Amazon and Google were two of the early strategic partners in Anthropic, each with billions of dollars invested in them while providing cloud computing infrastructure support for the training and deployment of the Claude model.

The analysts believe that the first completed listing in the AI area will set a benchmark for the whole industry ' s valuation system and provide direct access to the AI track for investors who have been waiting for years before.