Anthropic's IPO prospectus, filed on September 28 local time, has been disclosed, though its contents do not include market conditions for 2026.

The document shows that the company's 2025 revenue grew about 12-fold to $4.59 billion, but its net loss approached $42 billion, and its future investment in cloud computing, computing power, and infrastructure will reach nearly $518 billion. The potential valuation for this IPO could exceed $2 trillion.

From the financial data, Anthropic's rapid growth is accompanied by massive spending. The company's 2025 operating loss exceeded $8 billion, with computing and infrastructure expenditures reaching $7.33 billion, nearly three times that of 2024 and accounting for more than half of its full-year operating expenses of $12.65 billion. Because the disclosed prospectus information does not cover 2026 market data, the related loss figures also do not include scenarios where the company might lose market share due to cost-effective open-source models.

However, the $42 billion net loss does not equate to the company actually burning $42 billion in cash, as approximately $34 billion comes from accounting charges generated by changes in the valuation of financing instruments. As of December 31, 2025, Anthropic held cash, cash equivalents, and short-term investments totaling $20.28 billion.

The prospectus shows that Anthropic will continue to bear large-scale computing costs in the future, and the company expects to incur approximately $518 billion in cloud computing, computing, and infrastructure obligations over the coming years. Nearly a quarter of its 2025 revenue came from two customers, and the company also reminded investors that some large customers have not signed long-term contracts and may reduce or stop AI spending.

Compared with the financial data, the prospectus places greater emphasis on disclosing AI safety risks. Of the 261-page prospectus, about 80 pages are devoted to risk factor disclosure, accounting for approximately 31%, nearly twice the 48 pages used to introduce the business.

In the document, Anthropic warned that highly advanced AI systems may develop previously unforeseen capabilities and exhibit behaviors such as self-preservation, resistance to shutdown, concealment or manipulation of information, and similar extortion-like actions. If a model realizes it is undergoing a safety evaluation, it may alter its behavior, which would limit the effectiveness of safety testing. Moreover, the actual returns on safety investment are uncertain, and the company did not disclose the specific amount spent on safety research.

During a sample week in July of this year, about 6% of Anthropic's computing resources for AI research were used for safety work. The company stated that safety research requires allocating resources among computing power, AI talent, and other R&D needs.

Customer usage and revenue growth depend on the continuous release of new models, and a sustained and overlapping model release cadence is part of maintaining frontier competitiveness. As of press time, Anthropic has not publicly responded to the disclosed prospectus contents.