Anthropic IPO Prospectus Revealed: Revenue of $4.59 Billion, Net Loss of $42 Billion

From the revenue side, Anthropic's growth rate has been exceptionally rapid.

In 2025, the company's revenue grew from less than $400 million in the previous year to $4.59 billion, an increase of 1,088%. The rapid penetration of its Claude model in the enterprise market served as a vital foundation for revenue expansion. Prior market research also indicated that Anthropic's share in enterprise AI procurement expanded rapidly, demonstrating that its monetization capabilities have gradually extended from technical products to real-world enterprise demand.

However, revenue growth has not translated into profits in tandem.

Anthropic recorded a net loss of approximately $42 billion in 2025, of which about $34 billion resulted from accounting charges caused by valuation changes in convertible financing instruments; therefore, this figure cannot simply be interpreted as the company actually burning through $42 billion in cash.

Even excluding this impact, however, the company's operating loss still exceeded $8 billion, indicating that its core business remains a long way from achieving stable profitability.

More notably, Anthropic's cost structure is heavily reliant on computing resources. Prospectus materials show that the company's spending on computing and infrastructure reached approximately $7.33 billion in 2025, a sharp year-over-year increase, becoming one of the most significant components of operating costs. This spending model differs substantially from traditional software companies: AI models require continuous purchasing of GPU compute power, cloud services, and data center resources, and as model capabilities strengthen and user scale expands, inference and training costs may escalate accordingly.

This means that Anthropic needs to prove two things simultaneously in the future. On one hand, the company must continue to maintain rapid revenue growth; on the other hand, it must also lower the unit cost of AI services fast enough, otherwise a larger revenue scale may come with even more massive computing expenditures.

This is also a major distinction between public and private market evaluations of AI companies. In the past, investors focused more on model capabilities and user growth, but now they need to further observe whether revenue growth can ultimately outpace infrastructure costs.

Another key piece of information disclosed in Anthropic's prospectus is that the company's future infrastructure needs remain massive.

Relevant materials show that the company faces approximately $518 billion in commitments related to cloud computing, compute power, and infrastructure over the coming years. This figure is not a lump-sum cash payment over the next year; rather, it reflects Anthropic's long-term spending arrangements with infrastructure providers to support future AI model training and commercial deployment.

This scale demonstrates that Anthropic's growth trajectory differs significantly from that of typical software startups.

Traditional SaaS companies can typically add users on top of existing servers, whereas frontier AI companies must continuously expand computing resources. Growth in model parameters, expansion of context windows, deployment of AI agents, and an increasing number of enterprise clients will all continue to drive up demand for compute power.

For this reason, Anthropic relies on major partners such as Amazon, Google, and Nvidia for computing and financial support. Reuters previously reported that Anthropic has established deep partnerships with several tech giants and made large-scale arrangements centered around its compute power needs for the coming years.

However, heavy investment also entails higher execution risk.

If AI demand continues to grow rapidly, these massive infrastructure commitments can help Anthropic lock in compute power in advance, providing a safeguard for future revenue growth. Conversely, if market demand cools, or if the price of model services drops faster than compute costs, the company may face underutilization of its infrastructure.

In this IPO prospectus, the company devoted a considerable amount of space to disclosing AI safety risks. Reuters reported, citing the prospectus, that roughly 80 pages of the 261-page filing are dedicated to risk factors, far exceeding the length of the risk sections in many traditional corporate IPOs.

Anthropic specifically warned that highly advanced AI models may exhibit unpredictable behaviors, such as resisting shutdown, manipulating information, and even simulating extortion in testing environments. The company also emphasized that as model capabilities and scope of application continue to expand, the risk of models causing potential harm may also increase.

This risk disclosure also reflects a tension in Anthropic's business model. The company has long treated AI safety as a key competitive differentiator against OpenAI, Google, and Meta, but the fierce model race requires it to continuously increase release speeds. If safety testing extends product cycles, Anthropic could come under pressure in the competition for customers.

This content was translated using AI and reviewed for clarity. It is for informational purposes only.