Not just Eastern users.
The big one has finally arrived.
On October 26, Anthropic publicly filed its S-1 registration statement with the U.S. Securities and Exchange Commission, preparing to list on the Nasdaq Global Select Market under the ticker symbol ANT. According to media reports, the company’s valuation has been discussed in a range of $1.5 trillion to $2 trillion, though the proposed price range has not yet been disclosed.
Within 24 hours, this more than 300-page prospectus has been thoroughly dissected by both the tech and finance communities, with everyone already scrutinizing the most eye-catching figures: revenue of $16.286 billion in the first half of 2026, compared to just $1.163 billion in the same period last year; operating profit of $974 million in the second quarter, marking the company’s first-ever quarterly operating profit; as for the net loss of $80.7 billion on the books, the bulk stems from $72.388 billion in changes to the fair value of convertible bonds, derivative liabilities, and SAFE agreements—non-cash accounting expenses.
But beyond these Wall Street narratives, domestic developers and everyday users experience a different kind of disconnect. In the eyes of mainland Chinese users, Claude is probably the AI product on the market today that has been banned most often, subject to the strictest risk controls, and most frustrating to use.
Why does a company that has developed cutting-edge large models always seem so indifferent when dealing with ordinary individual users?
After reviewing the financial notes and risk disclosures in the prospectus, the answer is actually right there in the numbers. In Anthropic’s business landscape, revenue generated from individual users accounts for a very small share.
Less than 20% of individual users
To understand who a company truly cares about, the most straightforward approach is to examine its revenue breakdown. In Note 2 of the financial statements in its prospectus, Anthropic divides its revenue into two main categories: usage-based revenue and subscription revenue.
The data comparison is striking. For the full year 2025, usage‑based revenue totaled $3.796 billion, while subscription revenue was only $789 million, a ratio of roughly 83% to 17%. By the first half of 2026, usage‑based revenue surged to $13.064 billion, with subscription revenue at $3.222 billion; the former still accounted for about 80%, while subscription revenue made up only around 20%.
The more critical details are hidden in the breakdown of subscription revenue. Within that 20%, there are both Claude Pro and Claude Max for individual users, as well as Team and Enterprise plans for enterprises; the prospectus does not separately itemize individual subscriptions. This means that the revenue contributed by individual users purchasing memberships certainly accounts for less than 20% of Anthropic’s total revenue—and may be much lower.
So, where does the majority of usage-based revenue come from? The answer is cloud vendor channels and tens of thousands of enterprises.
Revenue generated from selling to customers through the marketplaces of cloud providers such as Amazon, Google, and Microsoft has been soaring: approximately 11% in 2023, about 32% in 2024, roughly 47% in 2025, and around 48% in the first half of 2026. Nearly half of the market is supported by the distribution channels of these three major cloud giants.
Driving this money-printing machine is a rapidly expanding cohort of institutional clients. As of September 30, 2026, the number of customers generating over $100,000 in revenue over the past 12 months—counted by distinct organizations—has surged from just over 950 a year earlier to more than 11,900. The customer mix remains robust: in the first nine months of 2026, no single customer accounted for more than 10% of revenue, and the top five customers collectively represented approximately 25%.
Enterprise customer growth chart from the prospectus: Customers contributing over $100,000 in revenue over the past 12 months grew from approximately 20 at the end of 2023 to more than 11,900 as of September 30, 2026 | Image source: Anthropic
The prospectus uses a specific term for individual paying users: “prosumers,” defined as individual users who subscribe to AI tools for knowledge‑work tasks, such as researchers, developers, analysts, and writers. The title sounds very professional, but these users still occupy a marginal position when it comes to core financial metrics.
The entire 300-plus-page prospectus does not disclose daily or monthly active users, nor does it mention the number of paid subscribers.
The metric that best reflects this attitude is the net income retention rate.
As of September 30, 2026, Anthropic’s net revenue retention rate stands at approximately 900%, a figure that would make any software company envious. However, when calculating this key metric, individual subscribers to Claude Pro and Max are excluded, on the grounds that fixed-price subscriptions do not accurately reflect the expansion and contraction behaviors that this metric seeks to measure.
What a company cares most about is reflected in the metrics it reports to investors. When a company fails to disclose daily and monthly active users, and even excludes individual subscribers when calculating core retention rate, the weight it assigns to ordinary individual users speaks for itself.
Money is earned by Claude Code and the enterprise.
Since the bulk of revenue is billed on a usage basis, what exactly are enterprise customers paying for? The prospectus shines a spotlight on Claude Code, calling it “the product that most concretely defines what autonomous AI work looks like.”
An ordinary chat box doesn’t consume many tokens; what really drinks up computing power like water are agents that can write their own code, run tests, and fix bugs. The prospectus provides a comparison: in the third quarter of 2026, Claude Enterprise users who don’t use Claude Code will spend less than $75 per month on average, whereas the top 5% of Claude Enterprise users who do use Claude Code will spend roughly 20 times as much.
The prospectus categorizes the product forms into three generations: chatbots, Copilots embedded within individual applications, and “virtual collaborators” capable of taking on objectives and operating autonomously over extended periods | Image source: Anthropic
This is also why Anthropic has been able to generate tens of billions in revenue with a lean team. As of September 30, 2026, the company had approximately 5,300 employees, and over the past 12 months, its average revenue per employee exceeded $6.6 million—up from $1.4 million a year earlier.
Once you understand the operating logic of this money-making machine, take a look at how it treats its free users.
In financial accounting, the computing power consumed by free users is classified under “Sales and Marketing” expenses. The financial statement notes clearly state that this expense category includes channel partner fees, marketing and advertising, personnel costs, as well as “the cost of computing power and infrastructure provided for free access.” In the first half of 2026, sales and marketing expenses totaled $3.808 billion. In other words, free users represent a customer acquisition cost on the books.
The risk factors section makes the point more explicitly. The company acknowledges that offering free access is largely intended to build awareness and usage habits, reaching individuals who perform knowledge work outside corporate organizations, with the hope of gradually converting them into paying customers; however, this conversion may not succeed, and the free tier could also cannibalize paid subscriptions.
The prospectus also emphasizes twice that the company “has been operating in an environment where computing power is constrained, with customer demand exceeding available capacity.” The prospectus does not specify how computing power is allocated among different customers, but given the perpetual shortage of computing resources, it’s easy to see the math: on one side are individual subscribers with fixed monthly fees, and on the other are enterprise users of Claude Code who spend over 20 times as much.
The “Revenue per $1 of Computing Power” chart in the prospectus: rising from $0.40 in the first quarter of 2025 to $2.08 in the second quarter of 2026 | Image source: Anthropic
Pro and Max users worldwide are experiencing changes to quotas and rate limits. In response, Anthropic openly acknowledges in its risk factors: “We have altered, and may in the future alter, usage caps, rate limits, or the capabilities of our models and products, which could harm our reputation and the perceived value of our offerings among users. Some of these changes have already triggered, and may continue to trigger, accusations that our disclosures regarding model access or capabilities are inadequate or misleading.”
Knowing that revising the quota would invite criticism and damage its reputation, Anthropic still included this provision in its prospectus, proactively warning investors.
As a user of “Fireseed”
Since revenue from individual users is so low and it still consumes limited computing resources, why doesn’t Anthropic simply shut down its consumer-facing platform and become a pure backend API provider?
This is precisely the nuance in the prospectus. While individual users contribute only a small share to revenue, they hold irreplaceable strategic value in the chain of commercial expansion.
They are the spark that helped Anthropic break into the enterprise.
The prospectus reviews this growth flywheel. So far, Claude’s adoption has been largely organic, driven by software developers and professional individual users. These users first adopt Claude themselves and then promote its broader use within their organizations, bypassing the need to build a large direct-sales team. Officially, this approach enables them to achieve scale with a smaller team than other major enterprise technology companies.
Software developers are the most central group. The prospectus argues that they understand how systems work, have credibility in evaluating technical tools, and frequently advocate for the adoption of new technologies internally; gaining their trust is the foundation for Claude’s wider dissemination among enterprise employees.
Once you understand this logic, you can see why free computing power is recorded under the “Sales and Marketing” account. Providing services to individual users is, at its core, Anthropic’s customer acquisition strategy: a developer first uses it themselves, then returns to the company to persuade their team to purchase the Enterprise plan, ultimately integrating the entire R&D workflow into Claude Code, which is billed on a pay-as-you-go basis.
However, this model has a vulnerability. The prospectus acknowledges that it assumes individual adoption will continue to translate into organizational-level adoption. In its risk factors section, Anthropic warns investors that companies with stronger consumer businesses, larger user bases, and more established brands may increase their investments in enterprise customers; if these competitors capture “the starting point of our enterprise adoption”—namely, individual users—through consumer-facing applications and other means, the company’s business could be adversely affected.
This creates a rather awkward situation. Anthropic relies on individual users to build its reputation and serve as internal advocates, fearing that competitors might poach this group; at the same time, its prospectus explicitly states that the company’s research priorities, product roadmap, and marketing team are all aligned around enterprise use cases.
It requires you to be the spark, but its core resources are all dedicated to serving enterprises. When large enterprise orders pour in and computing power becomes severely constrained, it’s precisely the spark itself that ends up having its oxygen supply cut off first.
A B2B company cloaked in a consumer-grade guise
With this prospectus now closed, Anthropic’s true nature has become quite clear: it is, at its core, a B2B company. To acquire customers and build an ecosystem, it has turned its product into a chat interface that anyone can easily use; the real business lies in selling pay-per-use computing power and autonomous agents to tens of thousands of enterprises.
Product architecture diagram from the prospectus: At the bottom are the four model series—Mythos, Opus, Sonnet, and Haiku; in the middle is the developer platform; and at the top is the value delivered to enterprise customers. | Image source: Anthropic
As for the issue of account bans, which is most sensitive to domestic users, the prospectus also includes a formal statement: the company “complies with regional policies” and implements IP‑based geographic blocking to restrict access from certain regions to its models, platforms, and applications. Mainland China was never included in its service‑eligible regions to begin with.
When a company valued at two trillion dollars stakes everything on enterprise workflows and code-based agents, individual users will only be relegated further down its priority list.
It’s just unclear whether Anthropic can still maintain its proud position as a starting point for enterprise adoption when more and more developers, having been turned away, switch to other competitors.
*Featured image source: Anthropic
This article is an original piece by GeekPark. For reprints, please contact Geekjun via WeChat at geekparkGO.