According to a draft reviewed by Reuters, Anthropic reported a net loss of nearly $42 billion in 2025, while its revenue was around $4.6 billion. The company behind this Claude chatbot hopes to go public with a valuation of over $2 trillion.
Both of these numbers appear in the same document. The document also warns that the company's own AI model may pose a existential risk to humanity.
Anthropic Refused to comment on the prospectus, according to Reuters.
Where does the $42 billion loss come from?
A loss of approximately $34 billion came from non-cash accounting losses, which means that there was no actual outflow of funds. This portion of the loss originated from convertible bond financing—i.e., the funds invested by investors that can be converted into company shares in the future—and their valuation increased as the value of Anthropic itself rose.
In simple terms: the more valuable a company is, the higher the equity of its early supporters will be on paper, yet this portion is recorded as a loss in the financial statements.
After excluding this portion, the operating loss—i.e., the actual loss incurred in business operations—exceeded $8 billion, which is higher than the previous year's loss of about $3 billion. Revenue was close to $4.6 billion, 12 times that of the previous year.
The largest expense is computing power, which refers to the chips and servers required to train and run the AI model. Expenditures on computing power and infrastructure have tripled, reaching $7.33 billion, accounting for more than half of the total operating costs of $12.65 billion.
And the future expenses will be even greater. Anthropic plans to invest approximately $518 billion in cloud computing and infrastructure over the next few years, of which about 80% is non-cancellable.
Among them, Google accounts for at least $111 billion, and Amazon accounts for $110 billion. In comparison, Anthropic held $20.28 billion in cash by the end of 2025.
SpaceX is also one of the suppliers, while SpaceX is now held by Elon Musk's xAI. According to SpaceX's own IPO documents, Anthropic agrees to pay $1.25 billion to SpaceX every month until May 2029 in exchange for computing power.
Revenue growth is also accelerating. In the second quarter of 2026, revenue exceeded $11.5 billion, more than double the total revenue for the entire year of 2025. However, nearly a quarter of the revenue in 2025 came from just two customers, and most of these major customers did not have long-term contracts.
In May this year, Anthropic raised $65 billion with a valuation of $965 billion, exceeding the previous valuation of $852 billion disclosed by OpenAI. Now, supporters hope that its valuation in the public market will more than double.
Risk Warning
The risk section alone occupies nearly 80 of the 261 pages of the document, which is longer than the 48 pages dedicated to the business itself, according to Forbes. The company lists these risks so that no one can claim in the future that they were not warned.
The document warns that increasingly autonomous models may exhibit self-protective behaviors, including resisting shutdowns and engaging in tactics similar to extortion. It also mentions that during controlled tests, these models have been known to manipulate code and assist in fraudulent activities.
CEO Dario Amodei has previously expressed similar views in public. In an article in June, he called for the establishment of mandatory safety regulations similar to those in aviation regulation and advocated for third-party testing of advanced AI systems.
The company is expected to make its debut on the stock market after the mid-term elections in the United States in November, and this prospectus has not yet been publicly released.
