Anthropic IPO Prospectus: First-Half Loss Hits $80.7 Billion, $750 Billion Computing Spend Draws Attention

TradingKey - Artificial intelligence company Anthropic is accelerating its IPO plans, but newly disclosed financial information is also prompting investors to reassess its growth model.

According to a draft S-1 prospectus reviewed by The Wall Street Journal, the Claude developer generated $16.3 billion in revenue in the first half of 2026, but its net loss reached $80.7 billion over the same period, while it could also incur up to $750 billion in computing and infrastructure expenditures over the coming years.

These figures demonstrate that commercial demand for generative AI is growing rapidly, while also exposing Anthropic's heavy reliance on a handful of customers, cloud service providers, and external financing. As the relevant prospectus filing remains a draft and the company declined to comment, the financial data and contract sizes in the final public version may still be subject to adjustment.

Where Did the $80.7 Billion Loss Come From Amid Rapid Revenue Growth?

Anthropic's revenue reached $16.3 billion in the first half of this year, far exceeding $1.1 billion in the same period last year. However, its net loss for the period also widened from about $13 billion in the first half of last year to $80.7 billion, standing in sharp contrast to its revenue growth on the surface.

The main reason for the sharp widening of losses did not stem entirely from daily operations. The draft prospectus showed that about $72 billion of this was an accounting charge related to changes in the valuation of convertible notes and other financing instruments. Such adjustments may fluctuate significantly alongside changes in the company's valuation and the value of related financial instruments, and do not mean that an equivalent cash outflow occurred during the period.

However, even after excluding this accounting impact, the company's losses remain worthy of attention. In the first half of 2026, Anthropic spent nearly $9 billion on computing power and infrastructure, almost four times the $2.3 billion in the same period last year, and also exceeding the related expenditure of about $7.3 billion for the full year of 2025.

This reflects that the commercialization of AI models still requires substantial upfront investment. Even if revenue increases rapidly, the costs of training models, running inference services, and expanding data centers may continue to squeeze operating margins.

Notably, Anthropic recently disclosed to potential investors that its third-quarter revenue reached $16.8 billion, up about 45% from the second quarter, and stated that it achieved adjusted profitability and positive free cash flow for the quarter. Although this shows that its operational performance is improving, adjusted profit is not the same as net profit under GAAP, and investors still need to pay attention to the specific calculation criteria and subsequent cash flow performance.

Computing Power Commitments Up to $750 Billion

Computing power supply is the most prominent constraint in Anthropic's expansion. According to the draft prospectus, the company faces approximately $450 billion in cloud computing payment obligations over the next seven to ten years, with most of these contracts being non-cancelable. This means that even if actual computing power usage or customer demand falls below expectations, Anthropic may still be required to pay the relevant fees contractually.

According to the arrangements disclosed in the filing, Anthropic plans to deploy funds to Google Cloud under Google parent Alphabet (GOOGL) and Amazon Web Services under Amazon (AMZN) totaling approximately $111 billion and $110 billion, respectively, while committing approximately $31 billion to Microsoft's (MSFT) cloud platform. The company has also committed to pay a combined total of up to $86 billion to Nscale and Lambda by 2033–2034 to lease computing resources powered by Nvidia (NVDA) chips.

Factoring in more than $160 billion in obligations arising from arrangements via Broadcom (AVGO) to use Google's custom chips, Anthropic's locked-in computing power and infrastructure commitments exceed $600 billion. The company is still in talks for up to roughly $150 billion in new agreements, and if all these deals materialize, the total scale could further rise to $750 billion.

While these long-term contracts ensure Anthropic gains access to the resources needed to train and run its models, they also convert future growth expectations into fixed costs in advance. If AI demand continues to expand and computing resource utilization remains high, these investments will serve as an important barrier for the company to expand its market share. However, if model prices decline, customers shift to open-source solutions, or industry demand falls short of expectations, these non-cancelable payment obligations could continuously drain cash flow.

Customer Concentration and AI Safety Risks Test IPO Valuations

Although Anthropic's revenue is growing rapidly, its customer base remains concentrated. Of its $16.3 billion in revenue in the first half of 2026, about 20% came from two unnamed customers, representing a combined contribution of approximately $3.26 billion. The company expects that for the foreseeable future, a significant portion of its revenue will continue to depend on a limited number of major customers.

While this structure can rapidly amplify revenue during growth phases, it also means that a single customer cutting usage, pushing down prices, or switching to competitors could significantly impact performance. Some enterprise customers have not signed long-term procurement agreements, leaving Anthropic to bear non-cancellable long-term computing costs while facing the asymmetric risk of customer demand changing at any time.

Anthropic also listed AI safety and potential legal liabilities as major risks. According to its draft prospectus, the company is concerned that its models could exhibit behavior that does not align with human intent, leading to data breaches, cybersecurity incidents, or other damages. If losses from such events exceed insurance coverage, the company may have to bear unpredictable financial liabilities.

Meanwhile, data center expansion could also encounter growing local resistance. Factors such as power supply, land use, environmental impact, and community opposition could delay construction progress, further increasing uncertainty around computing power supply.

These risks are influencing how investors assess Anthropic's IPO valuation. While previous market discussions reached valuations as high as $2 trillion, Reuters reported, citing people familiar with the matter, that after the latest financial data was disclosed, some investors began considering a valuation range of around $1.5 trillion. These figures remain subject to market discussion and do not represent final IPO pricing.