Written by | Duoke

Source | Beiduo Business & Beiduo Finance

Anthropic, PBC (hereinafter "Anthropic"), the parent company of the Claude large language model, recently made public its prospectus filed with the U.S. Securities and Exchange Commission (SEC) on October 26, 2026, planning to list on Nasdaq under the ticker "ANT."

According to Anthropic's disclosure, its post-money valuation in the H round stood at $965 billion, while U.S. capital market expectations for its listing valuation have already reached $1.5 trillion to $2 trillion. If priced at the upper limit, its fundraising would surpass the $86.25 billion record (including the over-allotment option) set by SpaceX in June this year, making it the largest IPO in global history.

The operating data disclosed in Anthropic's prospectus presents a stark duality. On the one hand, the rapid growth of the company's revenue confirms the authenticity of enterprise-level AI demand, and the inflection signal of quarterly profitability at the operating level also proves that its commercialization capability is beginning to materialize.

On the other hand, however, massive book losses, deep dependence on cloud vendor channels, and irrevocable compute commitments in the hundreds of billions of dollars together constitute a risk exposure that cannot be ignored in Anthropic's valuation narrative, reminding investors of potential cracks beneath its high-growth story.

For Anthropic, which plans to list on Nasdaq, how to absorb the above structural pressures while delivering growth will become a key variable affecting its IPO pricing and subsequent valuation.

I. Leading Market Position, but Channel Structure Raises Concerns

According to Tianyancha and the prospectus, Anthropic was co-founded in 2021 by former OpenAI Vice President of Research Dario Amodei, Vice President of Safety Policy Daniela Amodei, and several former core OpenAI members.

Anthropic's core business is the research and development of the Claude series of large models, delivered externally mainly through two methods: model API billing by usage volume and subscription products such as Claude Pro/Max, while expanding enterprise scenarios through application tools such as Claude Code and Claude Cowork.

Anthropic's enterprise strategy aims to drive durable customer adoption through paths including embedding Claude directly into customers' critical business processes and supporting customers in developing their own products based on Claude. As an agentic workflow, Claude can drive growth, improve efficiency, and bring considerable returns to enterprise customers.

This verifiable value is being translated into rapid customer expansion. As of the end of September 2026, Anthropic's net revenue retention rate was as high as 900%; it had more than 11,900 customers with annualized spending above $100,000, more than 1,900 customers above $1 million, and about 250 customers above $10 million.

In terms of customer composition, Anthropic's customers with annualized spending of $100,000 or more already cover all Fortune 10 companies, more than 50% of Fortune 500 companies, and more than 30% of Global 2000 companies. Top enterprise customers have become an important part of its revenue base.

In addition to direct sales, Anthropic also uses distribution partners to expand customer reach. Claude can be accessed through the three mainstream cloud platforms AWS, Google Cloud, and Microsoft Azure. Anthropic jointly promotes with these cloud service providers (CSPs), signs contracts directly with customers through their platforms, and pays each CSP based on transactions.

But distribution channels are both a source of revenue and a cost burden. On the revenue side, about 48% of Anthropic's revenue in the first half of 2026 was completed through Amazon AWS, Google Cloud, and Microsoft Azure, compared with only 11% in 2023; on the cost side, fees paid to channel partners also increased by $2.7 billion year over year.

The deeper problem is that Anthropic's channel structure implies conflicts of interest. In addition to serving as Anthropic's distribution channels, the three cloud vendors also play the triple roles of investors, compute suppliers, and competitors with their own models.

Beiduo Business & Beiduo Finance has learned that Anthropic admitted in its prospectus that the compute it needs has exceeded the upper limits the three cloud vendors are willing to provide, partly because these vendors increasingly regard Anthropic as a direct competitor.

II. Strong Growth Momentum, Profitability Inflection Yet to Be Proven

From 2023 to 2025, Anthropic's revenue was $41 million, $386 million, and $4.585 billion, respectively, with a compound annual growth rate of 957.5%, growing more than a hundredfold in three years; in the first half of 2026, its revenue reached $16.286 billion, about 14 times that of the same period in 2025, and the high-growth trend continues.

From the revenue structure, Anthropic's API usage-based revenue in the first half of 2026 reached $13.1 billion, accounting for 80%. This API-dominated structure directly links its revenue sources to customers' token consumption, without the need to expand sales and delivery teams headcount-wise, and offers strong marginal scalability.

But the other side of rapid scale growth is persistently huge losses. In 2023, 2024, 2025, and the first half of 2026 (the "reporting periods"), Anthropic recorded net losses of $1.531 billion, $8.309 billion, $41.965 billion, and $80.748 billion, respectively.

Anthropic explained that the above losses mainly stem from fair value changes in convertible notes, derivative liabilities, and SAFEs, with related losses amounting to $72.388 billion. The reason is that the company's valuation continued to rise during each reporting period, driving the fair value of liability instruments such as preferred shares and convertible bonds to expand simultaneously.

If the above non-cash accounting losses are excluded, Anthropic's adjusted net profit has turned from -$1.849 billion in the first half of 2025 to positive $1.618 billion in the first half of 2026, with an adjusted operating margin of about 10%; in the second quarter alone, adjusted net profit reached $2.235 billion, with a margin of about 19%.

In addition, Anthropic's operating cash flow turned positive in the first half of 2026, achieving a net inflow of $3.9 billion, compared with a net outflow of $1.5 billion in the same period last year. However, about $2.2 billion of that came from non-cash items such as equity incentives, and the actual sustainability of cash flow generation remains to be observed.

The bigger uncertainty lies in Anthropic's list of compute commitments. The company has signed compute procurement agreements totaling more than $400 billion, involving rigid procurement commitments from Google, Amazon, and Microsoft, as well as equipment finance leases with Broadcom, with contract terms extending into the 2030s, and most contracts stating that payment must be made "regardless of actual usage."

In addition, Anthropic's agreement with SpaceX.AI has a cap of $120.5 billion, and there are also GPU service agreements with Nscale and Lambda, at approximately $44.6 billion and $41.3 billion, respectively. The above agreements are all billed on the premise of capacity delivery and acceptance, and are subject to conditions such as financing and delivery milestones.

Costs have been locked in by contract, while revenue depends on customer usage and market growth. Once demand falls short of expectations, Anthropic's compute spending, which cannot be scaled back, will directly squeeze its profits and cash flow. The time gap between cost rigidity and revenue elasticity is a variable that cannot be ignored in the company's valuation narrative.

Admittedly, what Anthropic tells the outside world is a story of "high growth + high investment," and its initial public offering will also set a valuation benchmark for leading AI labs. But how high a premium capital markets will be willing to pay for it ultimately still depends on whether its commercialization speed can catch up with the pace of fulfilling its compute commitments.

In this sense, this largest IPO in global history is both the first capital market pricing test for an AI unicorn and a prolonged test of investors' patience against the compute bill.