(This article is authored by Gao Heng, and published on Titanium Media with authorization.)
Article by Gao Heng
Revenue for the first half of the year exceeded three times last year’s full-year total, yet the net loss reached as high as $80.748 billion. On October 26, the U.S.-based artificial intelligence company Anthropic, PBC (hereinafter referred to as “Anthropic”) publicly filed its initial public offering (IPO) registration documents with the U.S. Securities and Exchange Commission (SEC), revealing this starkly contrasting financial performance. According to the prospectus, in the first half of 2026, the company generated $16.286 billion in revenue—approximately 14 times that of the same period last year—while its operating loss narrowed from $2.135 billion to $693 million.
The substantial net loss primarily stems from changes in the fair value of financing instruments and does not directly reflect the company’s actual operating performance. As its enterprise customer base grows and demand for model usage expands, Anthropic’s profitability is improving. However, with competitors such as OpenAI and Google continuously launching new models and adjusting pricing, coupled with sizable long-term compute‑power procurement contracts already in place, whether the company can sustain revenue growth and convert it into stable profits remains a critical challenge for this IPO.
01 Surge in enterprise customers, with half-year revenue exceeding last year’s full-year total by threefold
Anthropic’s revenue has grown rapidly over the past three years.
The prospectus shows that in 2023, Anthropic generated revenue of $41 million, which grew to $386 million in 2024 and further reached $4.585 billion in 2025.
Entering 2026, this growth momentum has accelerated further. In the first half of the year, the company reported revenue of $16.286 billion, roughly 14 times the $1.163 billion recorded in the same period of 2025 and more than three times last year’s full-year revenue. Specifically, first-quarter revenue stood at $4.731 billion, while second-quarter revenue surged to $11.555 billion.
Enterprise customers’ procurement and use of AI models are a key driver of Anthropic’s revenue growth.
Unlike pricing models that rely solely on fixed subscription fees, Anthropic offers both pay-as-you-go and subscription options. Enterprises can access the Claude model via an application programming interface (API) or a cloud computing platform and pay based on actual usage.
The prospectus shows that in the first half of 2026, Anthropic’s usage-based revenue reached $13.064 billion, accounting for approximately 80% of total revenue, compared to $983 million in the same period of 2025; subscription revenue grew from $180 million in the prior-year period to $3.222 billion, representing about 20% of total revenue.
Both revenue streams posted substantial growth, but usage-based billing accounted for roughly 80% of the year-over-year revenue increase. This type of business is primarily priced based on metrics such as the volume of data processed by the model; the more frequently customers invoke the model, the higher their typical fees tend to be.
As companies apply Claude to more real-world business scenarios, Anthropic’s number of major clients is also growing rapidly.
The prospectus discloses that, as of September 30, 2026, more than 11,900 customers had generated over $100,000 in revenue for Anthropic over the past 12 months, compared to just over 950 during the same period in 2025. Among them, more than 1,900 customers contributed over $1 million in revenue, and approximately 250 customers accounted for revenues exceeding $10 million.
In terms of revenue scale and customer base, Anthropic’s commercialization has progressed from early-stage market expansion to a phase in which enterprise customers are steadily increasing their procurement. Notably, code development is a key area where it has built a competitive advantage.
Anthropic’s Claude Code can assist developers with tasks such as code generation, modification, and debugging, gradually integrating into enterprise software development workflows. Compared to general-purpose conversational applications, the use cases for code development are more specific, and enterprises can assess the practical value of related tools through development efficiency metrics.
The report “2025: The State of Generative AI in the Enterprise,” published in December 2025 by the U.S. venture capital firm Menlo Ventures, indicates that, based on surveys of U.S. companies’ AI usage and market modeling estimates, Anthropic holds a 40% share of the enterprise large-model API spending market, surpassing OpenAI’s 27% and Google’s 21%.
From a market perspective, Anthropic’s estimated share has risen from 12% in 2023 to 40% in 2025, while OpenAI’s has declined from 50% to 27% over the same period. In the market for code-generation models, Anthropic’s estimated share stands at 54%, compared with 21% for OpenAI. Menlo Ventures believes that the widespread adoption of Claude Code is one of the key factors driving Anthropic’s increased market share.
Payment data from another U.S. corporate expense management platform, Ramp, also reflects the shifting competitive landscape between the two companies in the enterprise market.
The Ramp AI Index, released by Ramp on October 14, shows that as of September 2026, 43.8% of its platform’s enterprise customers had paid fees to Anthropic, surpassing OpenAI’s 40.6%.
However, looking at the growth for the month, the two companies differ. Compared with August, Anthropic’s enterprise paid adoption rate increased by 0.25 percentage points, while OpenAI’s rose by 1.12 percentage points, narrowing the gap between them.
Ramp’s Chief Economist, Ara Kharazian, noted in the report that as corporate adoption of AI-powered solutions increases, new customer growth is slowing. Going forward, market expansion will rely more on existing customers scaling their usage rather than simply adding more purchasing organizations.
Enterprise customer expansion has driven rapid revenue growth at Anthropic, but providing model services also requires ongoing investment in computing resources. Despite the company’s steadily expanding revenue scale, its profitability tells a different story.
02 Net loss exceeded $80 billion, yet operating losses narrowed significantly.
The prospectus shows that from 2023 to 2025, Anthropic reported net losses of $1.531 billion, $8.309 billion, and $41.965 billion, respectively. In the first half of 2026, its net loss widened further to $80.748 billion, far exceeding its revenue of $16.286 billion for the same period.
The primary reason for the substantial losses was not model development or day-to-day operating expenses, but rather changes in the accounting valuation of financing instruments.
In the first half of 2026, Anthropic recognized a loss of $72.388 billion due to changes in the fair value of convertible notes, derivative liabilities, and future equity simple agreements (SAFEs), accounting for nearly 90% of its net loss for the same period. In addition, the company recorded other net expenses totaling $8.757 billion, primarily related to the initial fair-value recognition and subsequent valuation adjustments of financing-related agreements.
These losses are closely tied to Anthropic’s financing structure. Previously, the company raised capital through convertible notes and other instruments, with the value of some of these instruments adjusting in line with changes in the company’s valuation. When Anthropic’s valuation increases, the fair value of the corresponding financial liabilities may also rise, resulting in losses on the income statement—though this does not necessarily mean the company paid an equivalent amount in cash during the same period.
Despite the massive paper losses, Anthropic’s operating profit has improved significantly.
In the first half of 2025, Anthropic reported an operating loss of $2.135 billion, with an operating loss margin of approximately 183.6%. In the same period of 2026, the operating loss narrowed to $693 million, and the operating loss margin declined to 4.3%.
On a quarterly basis, in the first quarter of this year, Anthropic still reported an operating loss of $1.667 billion, while in the second quarter it posted an operating profit of $974 million—the first time its operating profit turned positive in a single quarter during the period disclosed in its prospectus.
This change is primarily driven by revenue growth outpacing the rate of increase in operating expenses.
The prospectus shows that in the first half of 2026, Anthropic’s total operating expenses amounted to $16.979 billion, compared with $3.298 billion in the same period of 2025—a roughly 4.1-fold increase—while revenue grew by about 13 times over the same period. With revenue rising by $15.123 billion and operating expenses increasing by $13.681 billion, the difference between the two resulted in an operating loss that narrowed by $1.442 billion year over year.
Among these, computing power and infrastructure remain Anthropic’s largest expense category.
In the first half of 2025, the company’s computing power and infrastructure expenses totaled $2.334 billion, rising to $8.945 billion in the same period of 2026—a year-over-year increase of 283%, surpassing even the full-year figure of $7.326 billion for 2025.
However, as revenue has expanded rapidly, this expense as a percentage of revenue has declined from approximately 200.7% in the first half of 2025 to 54.9%.
Anthropic explained in its prospectus that the increase in computing‑power expenses was primarily driven by the gradual delivery of computational resources under previously signed cloud‑service contracts. Such costs are typically recognized when the computing power becomes available, rather than at the time the contract is executed. As revenue grows, the share of pre‑locked computing‑power costs in total revenue has correspondingly declined.
According to the functional classification of operating expenses in Note 8 of the prospectus, Anthropic’s gross margin has also improved significantly.
In the first half of 2026, the company’s cost of revenue was $2.38 billion, resulting in a gross margin of approximately 85.4%, compared to about 12.4% in the same period of 2025. During the same period, Anthropic’s R&D expenses reached $9.741 billion.
Cost of revenue primarily comprises the computing resources directly used to serve customers, along with related expenses, while R&D expenses cover investments in model research, development, and other activities. These two categories differ from the previously mentioned cost-of-compute and infrastructure expenses, which are classified by expense nature, and therefore should not be double-counted.
In addition to computing power, personnel and sales expenses are also growing rapidly.
In the first half of 2026, Anthropic’s personnel expenses increased from $538 million in the same period last year to $3.152 billion, with approximately $2.2 billion attributable to equity-based compensation; other operating and administrative expenses rose from $306 million to $1.714 billion.
More pronounced growth was seen in sales, marketing, and partner expenses, which rose from $120 million in the same period last year to $3.168 billion. In its prospectus, the company attributed approximately $2.7 billion of this year-over-year increase to higher channel partner expenses.
This change is related to Anthropic’s sales model. In the first half of 2026, approximately 48% of the company’s revenue will be generated through cloud service provider market channels. These platforms help Anthropic reach enterprise customers, and the company is required to pay partnership fees in accordance with the agreements.
In terms of cash flow, in the first half of 2026, Anthropic’s net cash flow from operating activities reached $3.929 billion, compared to a net outflow of $1.513 billion in the same period last year.
The prospectus indicates that, in addition to improved operating performance, this change was also influenced by factors such as an increase in accounts payable, a decrease in prepaid expenses, and growth in deferred revenue. During the same period, the company’s accounts receivable likewise rose significantly alongside business expansion.
Overall, revenue growth outpaced the increase in operating expenses, leading to a significant narrowing of Anthropic’s operating loss and an improvement in operating cash flow. However, as competitors continue to launch new models and customer demand and pricing evolve, the sustainability of this operational improvement will be directly affected.
03 Model competition intensifies; long-term investment in computing power remains a key challenge.
The prospectus indicates that Anthropic lists OpenAI and Google as its primary competitors in the developer- and enterprise‑customer market. The company also notes that some competitors possess stronger financial resources, more extensive technological capabilities, and a greater supply of computing power.
Among them, OpenAI is continuously expanding its enterprise customer base and the code development market, while Google simultaneously boasts the Gemini model, a cloud computing platform, and its own custom-built computing chips. For Anthropic, competition extends beyond model capabilities and product pricing to encompass customer channels, computing resources, and sustained R&D funding.
This competition has already been reflected in companies’ decisions to switch model suppliers.
A study released by Ramp on October 6 shows that, as of September 2026, the proportion of companies in its sample switching AI model providers—calculated as a three-month moving average—has reached 8%, marking a new high for this metric.
The report notes that after Anthropic launched Claude Opus 4 and Sonnet 4 in May 2025, more enterprises were drawn to its models. Since then, the number of vendor switches has gradually declined, but in June 2026, OpenAI introduced new models and lowered prices, once again reshaping the competitive landscape.
Ramp’s research, published on October 14, further shows that, as of the week ending October 11, token usage in its token spending management sample increased by 52.4% compared to the week of September 13, while spending during the same period rose by only 23%.
In this sample, the number of tokens used by enterprises has grown significantly faster than actual spending. Changes in model pricing, usage patterns, and computational efficiency mean that large-model companies may not be able to generate revenue that scales proportionally with usage.
Enterprises can adjust suppliers based on model performance and pricing, which also requires Anthropic to continuously launch new, competitive products.
The prospectus discloses that in the third quarter of 2026, more than 80% of the company’s revenue will come from model versions released within the preceding six months. Anthropic states that, following the release of new models, most customers will gradually migrate to higher-performance versions.
New models are constantly being released, requiring ongoing investment in R&D and training resources, while computing power procurement typically needs to be planned well in advance.
The prospectus shows that, as of June 30, 2026, Anthropic’s future fixed or minimum procurement commitments total approximately $265.6 billion, with contract durations ranging from 1 to 20 years. The company further disclosed that its minimum commitment under the computing‑power procurement agreement with Amazon Web Services is at least $110 billion, while its related procurement commitments with Google Cloud amount to at least approximately $114.5 billion. These agreements fall under separate disclosure categories, so the amounts cannot be simply added together.
Long-term procurement can secure the computing resources needed for future model development and customer use, but some contracts include minimum payment obligations that do not decrease in tandem with reduced customer usage.
In contrast, most of Anthropic’s revenue comes from model services billed on a pay-as-you-go basis, with some enterprise customers having no long-term procurement commitments and the ability to switch vendors based on model performance, pricing, and their own needs.
If market competition causes customers to switch to other models, or if model pricing declines leading to a slowdown in revenue growth, Anthropic will still be required to bear the portion of compute costs that has already been agreed upon.
This competitive yet collaborative relationship is particularly evident among the major cloud service providers.
Amazon, Google, and Microsoft both provide computing resources to Anthropic and sell Claude services through their respective cloud platforms, while also competing with Anthropic in certain AI-related businesses.
Take Google as an example: it both owns the Gemini model and cloud computing infrastructure, enabling it to provide computing power to Anthropic, while also directly selling its own AI models to enterprise customers. For Anthropic, cloud platforms serve as a crucial channel for expanding its customer base and securing computing resources, with procurement prices and partnership fees directly impacting the company’s operating costs.
To improve the supply of computing resources and long-term costs, Anthropic is advancing its “Anthropic Direct” infrastructure model, enhancing its ability to allocate computing power by directly leasing compute chips and deploying dedicated computing resources.
This model can reduce reliance on traditional cloud service procurement methods, but it also means that companies must bear equipment leasing and other infrastructure costs and arrange the corresponding funding in advance.
From expanding its enterprise customer base to turning operating profit positive, Anthropic has demonstrated the profitability improvements brought by scaled operations. However, as competitors such as OpenAI and Google continue to launch new models, adjust pricing, and see long-term compute contracts come into effect, the company must maintain strong customer demand and improve the efficiency of its compute resource utilization.
For Anthropic, which is poised to enter the capital markets, revenue growth is already backed by concrete performance metrics. Whether this revenue can translate into stable profits amid ongoing model competition and substantial investments in computing power will be the more critical test in the next phase.
For more exciting content, follow TMTPost on WeChat (ID: taimeiti), or download the TMTPost App.