Anthropic submitted its prospectus: last year it lost 42 billion dollars, revenue grew 12 times to 4.6 billion dollars, and the risk section warned of "threats to human existence."

Author: Bao Yilong
Anthropic has disclosed its initial public offering prospectus, outlining an artificial intelligence company attempting to enter the capital market with astronomical valuations while admitting that its core technology may pose "catastrophic or even existential risks" to humanity.

On September 28, according to the prospectus obtained by Reuters, Anthropic's revenue is projected to grow 12-fold to nearly $4.6 billion by 2025, while operating losses are expected to widen from $2.98 billion in 2024 to $8.06 billion during the same period.

The company spent $7.33 billion on computing power and infrastructure last year, doubling its expenditure from 2024, accounting for more than half of total operating expenses of $12.65 billion.

The prospectus reveals that Anthropic plans to invest $518 billion in cloud computing, computing power, and infrastructure over the next few years, while the actual voting power of common shareholders post-IPO will be significantly restricted. The seven co-founders will hold 50.1% of the voting rights on key matters through a new entity called "Founder LLC."

This IPO plan comes at a time when AI and chip stocks have recently faced sell-offs, further testing the market's tolerance for AI investment enthusiasm.

Wall Street Journal previously mentioned that Anthropic's IPO timing may be delayed until after the U.S. midterm elections in November.

Valuation Aiming for $2 Trillion, Doubling Its Own Six-Month-Old Expectations

Anthropic's IPO target valuation exceeds $2 trillion, while just in May of this year, the company's own valuation forecast was still at $965 billion. In just a few months, the valuation expectation has more than doubled.

Of the nearly $42 billion net loss, about $34 billion is attributed to non-cash accounting expenses, reflecting the fair value increase of convertible instruments from previous financing arrangements, rather than operational cash outflows. As of December 31, 2025, the company holds a total of $20.28 billion in cash, cash equivalents, and short-term investments.

The prospectus also reveals several business risks: nearly a quarter of revenue comes from two clients, and most major clients have not signed long-term contracts, which could lead to reductions or halts in procurement at any time.

If this IPO proceeds, it will become one of the strongest IPO cases in the U.S. market since 2021.

A reference point may be SpaceX's IPO in June this year, where the company's stock price surged 19% to $160 on the first day of trading but has since fallen back to around $147, prompting investors to remain cautious about the high valuations of high-growth companies.

Risk Warning: Prospectus Devotes 80 Pages to AI Threats

In this 261-page prospectus, the risk factors section spans approximately 80 pages, nearly double the 48 pages dedicated to business descriptions. In contrast, SpaceX's 277-page prospectus contains only about 38 pages on risks.

Anthropic explicitly warns in the document that its AI models may exhibit "self-protective behaviors," including "resisting shutdown," "concealing or manipulating information," and "similar to extortion" behaviors, noting that "the development of highly advanced models and the expansion of use cases may further increase the risk of harm caused by the models."

The company also admits that models may develop unexpected capabilities during training, and these capabilities "may only be discovered after the model is deployed and triggers significant safety incidents."

Company safety researcher Evan Hubinger estimates that the probability of AI causing human deaths in the next decade exceeds 10%.

Despite this, Anthropic's actual investment scale in safety has not been disclosed in the document. The company states that, for example, in one week in July this year, about 6% of the computing power used for AI research was allocated to safety work.

The company acknowledges that the commercial returns on safety investments are currently unclear but also states:
The market will reward reliable, trustworthy, and safe AI systems.

Governance Structure: Founder Team Dominates, Ordinary Investors' Influence Limited

To seek a balance between commercial interests and the AI safety mission, Anthropic has designed a unique corporate governance structure.

The company will continue to operate as a Public Benefit Corporation (PBC) in Delaware, while establishing a new entity called "Founder LLC," composed of the seven co-founders. This entity holds Class F shares and enjoys 50.1% of the total voting rights on key company matters, including the election of certain directors and other significant issues.

The Class A common stock offered to ordinary investors has only one vote per share, but under the aforementioned structure, the actual influence will be significantly limited. The prospectus also admits that this structure may lead to certain 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."

Among the seven founders, CEO Dario Amodei and his sister, company president and chair Daniela Amodei, are the core figures.

In 2025, Dario's compensation is approximately $18 million, while Daniela's is $16.4 million, primarily in stock and option rewards. Both have committed to using 80% of their personal holdings in Anthropic for charitable purposes.

Anthropic's board also includes four other members, who will be elected by a "long-term benefit trust" established by the company, with current trustees including former Federal Reserve Chairman Ben Bernanke and national security expert Richard Fontaine.

Trillion-Dollar AI Arms Race: OpenAI is the Biggest Competitor

Anthropic faces fierce competition in the AI field from multiple fronts, with OpenAI being its main competitor.

The two companies are engaged in a comprehensive battle for enterprise clients, top talent, and policy influence in Washington. According to media reports, OpenAI secretly submitted its IPO application in June this year, expecting to complete the listing by early 2027 at the latest.

Anthropic was founded about five years ago by researchers who left OpenAI due to differences in corporate governance and AI safety philosophies, and it released its first large language model in March 2023, directly competing with OpenAI. Additionally, the company is also competing with SpaceX's xAI, Alphabet's Google, and Meta in the AI infrastructure space.

Amazon and Google are Anthropic's two early strategic partners, both injecting billions of dollars in investments while providing cloud computing infrastructure support for the training and deployment of the Claude model.

Analysts believe that the first pure target in the AI field to complete an IPO will set a benchmark for the entire industry's valuation system and provide a direct entry channel for investors who have been waiting for years to enter the AI race.