Summary
- In 2025, Anthropic incurred a net loss nearing $42 billion while generating $4.6 billion in revenue, with about $34 billion attributed to a non-cash charge related to convertible financing.
- The company's operational losses exceeded $8 billion, with plans for cloud and computing expenditures reaching approximately $518 billion over the next few years, against a cash reserve of $20.28 billion at the end of 2025.
- The prospectus highlights potential existential risks posed by its AI models, as the company seeks a valuation of over $2 trillion for its upcoming IPO, anticipated post the November midterms.
According to a draft IPO prospectus reviewed by Reuters, Anthropic reported a staggering loss of nearly $42 billion in 2025. The firm, known for its Claude chatbot, is aiming for a public offering that could value it at over $2 trillion.
This document contains both the financial losses and a caution regarding the existential risks associated with its AI technologies.
Myriad: Which company will IPO next? Click to make your prediction.An IPO marks the first occasion a private company offers shares to the public, and the prospectus serves as a critical disclosure detailing potential risks. Anthropic confidentially submitted a draft to the SEC on June 1, which has since been leaked to various media.
Anthropic chose not to comment on the prospectus, as reported by Reuters.
Understanding the $42 Billion Loss
Approximately $34 billion of the reported loss is attributed to a non-cash accounting entry, which does not reflect actual cash outflow. This charge relates to convertible financing—capital raised from investors that can later convert into equity—whose estimated value has increased alongside Anthropic's valuation.
Essentially, as the company's worth grows, the theoretical value of early investors' claims rises, leading to a recorded loss on the financial statements.
Excluding this non-cash charge, Anthropic's operational loss was over $8 billion, a significant increase from around $3 billion the previous year. The company’s revenue surged to nearly $4.6 billion, twelvefold compared to the prior year.
The most substantial expense was on computing resources, specifically the chips and servers required for AI model training and operation. Spending on compute and infrastructure shot up to $7.33 billion, accounting for more than half of the total operational costs of $12.65 billion.
Looking ahead, Anthropic plans to allocate about $518 billion toward cloud computing and infrastructure in the coming years, with around 80% of this expenditure being non-cancellable.
Google and Amazon are set to receive significant portions of these costs, with estimates of at least $111 billion from Google and $110 billion from Amazon. In contrast, Anthropic reported a cash reserve of $20.28 billion at the end of 2025.
SpaceX, which now owns Elon Musk's xAI, is one of Anthropic's suppliers, with an agreement to pay SpaceX $1.25 billion monthly for computing capacity until May 2029, as indicated in SpaceX's own IPO filing.
Revenue is rapidly increasing, with second-quarter 2026 figures surpassing $11.5 billion—more than double the total revenue for 2025. However, nearly 25% of the 2025 revenue came from only two clients, and most large contracts lack long-term commitments.
In May, Anthropic raised $65 billion at a valuation of $965 billion, surpassing the last disclosed valuation of $852 billion for OpenAI. Investors are now seeking more than double that amount in the public market.
Risk Concerns
The risk assessment section of the prospectus spans nearly 80 pages out of 261, which is more extensive than the 48 pages dedicated to describing the business itself, according to Forbes. This comprehensive risk disclosure serves to protect the company from future claims of ignorance regarding potential issues.
The document cautions that increasingly autonomous AI models may exhibit self-preserving behaviors, including resisting shutdown and potentially engaging in actions akin to blackmail. It also highlights incidents where models have sabotaged code or facilitated fraudulent activities during controlled tests.
This disclosure indicates that the company is proactively informing prospective shareholders about the potential for its products to misbehave, with risks likely escalating as the models become more sophisticated.
CEO Dario Amodei has publicly voiced similar concerns, advocating for mandatory safety regulations modeled after aviation oversight, which would involve third-party testing for advanced AI systems.
The IPO is anticipated to take place following the U.S. midterm elections in November, and the prospectus has not been made publicly available yet.