Anthropic has finally released its prospectus.

On October 26 (local time in the United States), Anthropic officially submitted its S-1 prospectus to the U.S. SEC, initiating its Initial Public Offering (IPO), with plans to list on Nasdaq under the ticker symbol 'ANT'.

As a leading global large model company, Anthropic's prospectus offers the outside world its first opportunity to see the true picture of the large model business through complete financial data.

Several figures in it are particularly noteworthy.

First is the astonishing revenue growth rate.

In the first half of 2026, Anthropic achieved revenue of $16.286 billion, a year-on-year increase of approximately 1300%. The revenue for just six months has reached 3.55 times that of the entire previous year.

What's even more noteworthy is that this growth is backed by a strong ability to expand existing clients.

As of the end of September this year, the number of Anthropic's clients contributing over $100,000 in revenue over the past 12 months surged from just over 950 a year ago to 11,900.

During the same period, the Net Revenue Retention Rate (NRR) reached approximately 900%. This means that the revenue from the same group of clients a year later was nine times that of the previous year.

At the same time, the company boasts extremely high per capita revenue generation among its employees.

As of the end of September this year, the past 12-month revenue (TTM) per employee at Anthropic exceeded $6.6 million, compared to just $1.4 million a year ago.

As revenue scales up, Anthropic is beginning to demonstrate significant operational leverage.

In the second quarter of this year, Anthropic's revenue grew by 1368% YoY, while computing and infrastructure expenses increased by about 300% during the same period. The revenue growth rate has significantly surpassed that of computing expenses.

Based on adjusted figures, Anthropic's operating profit for the first half of the year reached $1.618 billion, with a profit margin of around 10%.

This has also led to an interesting phenomenon.

On one hand, Anthropic exhibits the business characteristics of a top-tier software company: strong client expansion capabilities, astonishing per capita revenue efficiency, and gradually emerging operational leverage as revenue scales up.

On the other hand, it is also a company highly reliant on cutting-edge research and computing infrastructure. To maintain model competitiveness, Anthropic still needs to invest substantial funds in R&D, model training, and computing procurement.

The characteristics of research institutions, software companies, and capital-intensive infrastructure enterprises are simultaneously appearing in the same large model company.

This unique business model also raises a question worthy of reconsideration by investors: When large model revenue begins to grow at a rate far exceeding computing costs, what level of long-term profitability can this business achieve?

Through Anthropic's prospectus, we may, for the first time, gain a relatively complete understanding of the growth logic, cost structure, and future commercial value of large model companies.

/ 01 /Revenue Surges 13x, Per Capita Revenue Crushes Meta

In the first half of 2026, Anthropic achieved revenue of $16.286 billion, a year-on-year increase of 1300%, equivalent to 3.55 times that of the entire previous year.

If we shorten the time frame to a single quarter, the growth becomes even more exaggerated.

The prospectus shows that in the four quarters of 2025, Anthropic's revenues were $376 million, $787 million, $1.382 billion, and $2.040 billion, respectively.

In the first quarter of 2026, revenue jumped to $4.731 billion, a 132% increase from the previous quarter and a 1158% increase YoY.

In the second quarter, it further surged to $11.555 billion, a 144% increase from the previous quarter and a 1368% YoY increase.

In other words, Anthropic's revenue in the latest quarter has reached 2.52 times that of the entire year in 2025.

Moreover, this rapid growth is continuing. According to information disclosed at an investor conference on October 14, Anthropic's revenue in the third quarter further reached $16.8 billion, a roughly 45% increase from the previous quarter.

A key factor supporting this growth is the rapid expansion of enterprise clients and the continuous increase in AI usage depth among existing clients.

At the end of 2025, Anthropic had only 1,500 clients contributing annual revenue exceeding $100,000. By the end of September 2026, this number had surpassed 11,900, nearly eight times the previous figure in just nine months.

The growth among large clients is even more astonishing. As of the end of September 2026, Anthropic had over 1,900 clients contributing annual revenue exceeding $1 million, with about 250 clients contributing over $10 million annually.

In other words, among clients contributing annual revenue exceeding $100,000, about 16% have crossed the $1 million threshold.

At the same time, Anthropic's penetration among large enterprises is also rapidly increasing.

The prospectus shows that as of the end of September this year, over 50% of Global Fortune 500 companies have adopted Anthropic's products or services; among Global 2000 companies, about 30% have contributed over $100,000 in revenue to Anthropic over the past 12 months.

However, even within these large enterprises, Anthropic's share of budget remains limited.

According to the prospectus, the average spending by Fortune 500 clients on Anthropic accounts for less than 0.1% of their total enterprise spending.

This means that as AI application scenarios continue to expand within enterprises, there is still significant room for growth in revenue contribution from individual clients.

What's even more noteworthy than the growth in client numbers is Anthropic's extremely high Net Revenue Retention Rate (NRR).

NRR is one of the core metrics for measuring the growth quality of SaaS and enterprise software companies. Simply put, it observes how much revenue a group of existing clients can still contribute a year later.

For example, if a group of enterprise clients contributed $1 million in revenue last year and the same group contributed $1.2 million this year, the NRR would be 120%.

This figure reflects the combined effects of client churn, spending contraction, and additional purchases from existing clients. According to Anthropic's disclosure, as of the end of September 2026, its NRR was approximately 900%.

This means that for the group of existing clients under the statistical scope, the revenue scale has expanded to about nine times that of the same period last year.

Finally, let's look at an interesting metric: per capita revenue among employees.

As of the end of September 2026, Anthropic had approximately 5,300 employees.

During the same period, the past 12-month revenue (TTM) per employee exceeded $6.6 million. At the end of September 2025, this figure was only $1.4 million. In just one year, Anthropic's per capita revenue has increased to more than 4.7 times its previous level.

How exaggerated is this level? According to Meta's 2025 financial report, calculating by dividing full-year revenue by the number of employees at year-end: Meta's per capita revenue was about $2.55 million, Alphabet's about $2.11 million, and Microsoft's about $1.24 million.

In other words, Anthropic's latest disclosed per capita TTM revenue is approximately 2.6 times that of Meta, 3.1 times that of Alphabet, and 5.3 times that of Microsoft.

All signs indicate that Anthropic is demonstrating a very unique business model: extremely strong capabilities in expanding existing clients and generating per capita revenue.

/ 02 /Behind the Whopping $80.7 Billion Loss in Six Months

Even more astonishing than the revenue explosion is Anthropic's rate of losing money.

In 2025, Anthropic reported a full-year net loss of $41.965 billion. By 2026, it had lost $80.748 billion in just the first half, nearly double the entire previous year's loss, averaging over $400 million in losses per day.

The vast majority of these losses came from changes in fair value. The prospectus shows that of the $80.748 billion net loss in the first half, $72.388 billion came from changes in the fair value of financing instruments such as convertible notes, derivative liabilities, and SAFEs, accounting for nearly 90%.

Why did this category incur such massive losses?

The reason is simple. Over the past few years, Anthropic has raised funds through instruments like convertible notes. As the company's valuation rapidly increased, the potential equity value corresponding to these instruments also rose.

For example, suppose a startup previously issued a convertible note worth $1 billion to investors. Later, the company's valuation significantly increased, and the future equity value corresponding to this note also rose.

If, after revaluation, the fair value of the note liability increases to $3 billion, the company may need to recognize a $2 billion book loss under relevant accounting treatments.

Therefore, the $72.388 billion primarily represents book losses due to valuation changes and does not mean Anthropic actually spent that much cash.

From an operational perspective, Anthropic is approaching a true profitability inflection point.

In the first half of 2025, Anthropic achieved revenue of $1.163 billion, with an operating loss of $2.135 billion and an operating profit margin of -184%.

By the first half of 2026, revenue had surged to $16.286 billion, while the operating loss narrowed to $693 million, a year-on-year reduction of about 68%, with the operating profit margin improving to -4%.

If we break it down by quarter, the changes are even more pronounced.

In the fourth quarter of 2025, Anthropic's operating loss was $2.836 billion. In the first quarter of 2026, the loss narrowed to $1.667 billion.

By the second quarter, the company achieved revenue of $11.555 billion and an operating profit of $974 million, with an operating profit margin of 8.4%, marking its first quarterly operating profit under U.S. GAAP. Based on adjusted figures, Anthropic's operating profit for the first half of the year reached $1.618 billion, with a profit margin of around 10%.

Why did profitability change so dramatically in just six months?

The key lies in the fact that revenue growth began to far outpace the growth rate of computing expenses.

In the first half of 2025, Anthropic's computing and infrastructure expenses were $2.334 billion. By the first half of 2026, these expenses had grown to $8.945 billion, a 283% YoY increase.

While this seems like rapid growth, revenue exploded by 1300% during the same period. This means that computing investments began to generate increasingly higher revenue returns.

In the first half of 2025, for every $100 in revenue, Anthropic incurred $201 in computing and infrastructure expenses.

By the first half of 2026, this figure had dropped to $55.

Looking at it quarterly, the changes are even more pronounced. In the second quarter of 2025, every $100 in revenue corresponded to about $177 in computing expenses. By the second quarter of 2026, this had dropped to $48.

In just one year, the proportion of computing expenses in revenue has decreased by nearly 129 percentage points.

Several reasons lie behind this. First, the continuous increase in enterprise clients and the rapid growth in model invocation volumes expanded revenue scale. At the same time, ongoing optimizations in model training and inference efficiency, along with improved computing utilization and the use of multiple chip platforms, provided room for cost optimization.

These changes ultimately reflected on the income statement. In the second quarter of 2026, Anthropic's revenue grew by 144% QoQ, while total operating expenses increased by only about 65% during the same period.

Revenue growth began to release significant operational leverage, driving the company from a $1.667 billion loss in the first quarter to a $974 million profit in the second quarter.

This at least validates an important profitability path for the large model business model: As paid demand continues to grow, revenue growth surpasses that of computing and other operating expenses, and scale effects begin to truly translate into profits.

/ 03 /Large Model Companies Are Evolving into a New Type of Business Organization

AI large model companies may be forming a new type of business organization that differs significantly from traditional internet and software companies.

They simultaneously possess the characteristics of research institutions, software companies, and capital-intensive infrastructure enterprises.

First, research capabilities simultaneously determine product competitiveness and production costs.

This is a distinctive feature of Anthropic's business model. The prospectus mentions that from the beginning, Anthropic designed model research and computing efficiency in tandem, with researchers and infrastructure engineers jointly optimizing model architecture and inference systems.

This R&D model brings benefits in two ways.

On one hand, improved model capabilities can attract more clients and drive revenue growth; on the other hand, enhanced algorithms and inference efficiency reduce the computing resources required to complete the same tasks, thereby lowering costs.

In other words, technological innovation at large model companies can simultaneously impact both revenue and cost ends.

This is also why Anthropic places such emphasis on talent density. The prospectus discloses that over the past 12 months leading up to the end of September 2026, the company's regrettable employee turnover rate was only about 1%.

Second, computing resource management is becoming a core competitive advantage.

Anthropic devotes significant space in its prospectus to introducing its multi-chip computing architecture.

Currently, the company can simultaneously use AWS Trainium, Google TPU, NVIDIA GPU, and AMD GPU, allocating computing resources based on the cost and performance requirements of different tasks through an internal scheduling system.

This architecture not only reduces dependence on a single chip supplier but also improves the overall utilization efficiency of computing resources.

Anthropic even believes that its ability to train and run models across multiple chip architectures has, after years of investment, formed a competitive advantage that is difficult to replicate on a large scale.

This means that future competition among large model companies may largely hinge on who can provide services at lower costs while maintaining the same model performance.

Finally, enterprise workflows are becoming a new commercialization barrier.

As products like Claude Code and Claude Cowork improve in capabilities, Claude is gradually entering enterprise business scenarios such as programming, finance, healthcare, law, and cybersecurity.

When enterprises deeply integrate AI into actual workflows, the long-term accumulation of system integration, usage habits, and business experience increases the cost of switching products.

At the same time, as enterprises continuously discover new scenarios during usage and feed their needs back to Anthropic, they drive ongoing improvements in models and products.

This creates a virtuous cycle: improved model capabilities lead to more application scenarios; deeper enterprise usage enhances client stickiness, generating more revenue and product feedback.

From this perspective, Anthropic's prospectus actually reveals a set of business logic that is taking shape in the large model industry.

Scientific research and innovation determine product capabilities and cost curves, computing power infrastructure determines scalability efficiency, and corporate workflows determine the sustainability of revenue growth.

These three elements reinforce each other, ultimately enabling leading large model companies to achieve increasingly strong economies of scale.

Text/Xiaobai