Anthropic files for IPO: lost $4.2 billion last year, revenue grew 12x to $4.6 billion, risk section warns of "threats to human survival"

Anthropic has disclosed its initial public offering prospectus, outlining a company aiming to enter the capital markets with an astronomical valuation, while also admitting that its core technology may pose "catastrophic or even existential risks" to humanity.

According to the prospectus obtained by Reuters on September 28, Anthropic's revenue is set to increase 12-fold in 2025 to nearly $4.6 billion, while operating losses will grow from $2.98 billion in 2024 to $8.06 billion in the same period.

Last year, the company spent $7.33 billion on computing power and infrastructure—triple the expenditure in 2024—making up over half of its total operating expenses of $12.65 billion.

The prospectus shows that Anthropic plans to invest $518 billion in cloud computing, compute power, and infrastructure over the coming years, while the actual voting power of common shareholders post-listing will be significantly limited. The seven co-founders will, through a new entity named “Founder LLC,” hold 50.1% of the voting rights on key matters.

This IPO comes at a time when AI and chip stocks have recently faced sell-offs, further testing the market’s appetite for AI investments.

Wallstreet Vision previously mentioned that Anthropic’s public offering might be postponed until after the US midterm elections in November.

Anthropic is targeting a valuation of over $2 trillion in this IPO, while as recently as May, the company's own forecast was just $965 billion. In just a few months, expectations have more than doubled.

Of the nearly $42 billion in net losses, about $34 billion are non-cash accounting charges, reflecting an increase in the fair market value of instruments convertible to equity from prior financing, rather than operational cash outflows. As of December 31, 2025, the company’s cash, cash equivalents, and short-term investments totaled $20.28 billion.

The prospectus also reveals several business risks: Nearly a quarter of revenue comes from two clients, and most major customers have not signed long-term contracts, meaning they could reduce or stop purchases at any time.

Should this IPO be successful, it will become one of the best-performing listings in the US since 2021.

A relevant comparison might be SpaceX’s IPO in June this year. On its first trading day, shares soared 19% to $160 but have since dropped back to about $147, causing investors to remain cautious about the high valuations of high-growth companies.

In this 261-page prospectus, the risk factors section spans about 80 pages—nearly twice the 48 pages devoted to business descriptions. By comparison, SpaceX’s 277-page prospectus included only about 38 pages on risks.

Anthropic explicitly warns in the filing that its AI models may demonstrate “self-preservation behaviors,” including “resisting shutdown,” “concealing or manipulating information,” and even “ransomware-like” actions. The company notes that “the development and expanded use cases of highly advanced models may further increase the risk of harm caused by these models.”

The company further admits that models may develop unforeseen abilities during training, which “might not be discovered until after deployment and a significant security incident occurs.”

Company safety researcher Evan Hubinger estimates that there is over a 10% probability that AI could lead to human deaths within the next decade.

Despite this, the exact scale of Anthropic’s investment in safety was not disclosed in the document. The company stated that, for one week in July this year, about 6% of the compute power for its AI research was allocated to safety work.

The company recognizes that the commercial returns of investing in safety remain unclear but also states:

The market will reward reliable, trustworthy, and safe AI systems.

To balance commercial interests with its AI safety mission, Anthropic has designed a unique corporate governance structure.

The company will continue operating as a Delaware Public Benefit Corporation (PBC) and will establish a new “Founder LLC” entity composed of all seven co-founders. This entity will hold F-class stock, with 50.1% of total voting rights on key company matters, including major board appointments.

Class A common stock offered to ordinary investors will carry only one vote per share, but under this structure, their real influence will be significantly limited. The prospectus also admits that this structure could result in decisions “that conflict with the short-term, medium-term, or long-term financial interests and business performance, negatively impacting the value of the Class A common stock.”

The core founders are CEO Dario Amodei and his sister, President and Chairwoman Daniela Amodei.

In 2025, Dario’s compensation will amount to about $18 million and Daniela’s will be $16.4 million, mostly in shares and options. Both, along with other co-founders, have pledged to donate 80% of their personal Anthropic holdings to charitable causes.

The Anthropic board also includes four additional members, elected by a company-established “Long-term Benefit Trust.” Current trustees include former Federal Reserve Chairman Ben Bernanke and national security expert Richard Fontaine.

Anthropic faces intense competition in the AI sector, with OpenAI as its main adversary.

The two firms are engaged in fierce competition for enterprise clients, top talent, and influence in Washington, D.C. According to media reports, OpenAI covertly filed for an IPO in June this year and is expected to go public by early 2027 at the latest.

Anthropic was founded about five years ago by ex-OpenAI researchers who left over corporate governance and AI safety disagreements. In March 2023, Anthropic launched its first large language model, entering direct competition with OpenAI. The company also competes with xAI under SpaceX, Google under Alphabet, and Meta in AI infrastructure.

Amazon and Google are Anthropic’s two main early strategic partners, both investing billions and providing cloud computing resources to train and deploy the Claude model.

Analysts believe that the first pure-play AI company to go public will set a valuation benchmark for the entire sector and provide long-waiting investors direct access to the AI race.