Anthropic Files IPO S-1: Founders to Sell Shares, No Primary Offering Confirmed

Original | Odaily Planet Daily (@OdailyChina)

Author | Sudi Xia @ Odaily Planet Daily

On the evening of October 26, Beijing time, artificial intelligence company Anthropic filed its initial public offering (IPO) application with the U.S. Securities and Exchange Commission (SEC). According to Anthropic’s S-1 filing (click to view), the company expects to begin trading on the Nasdaq Global Select Market under the ticker symbol “ANT.” Anthropic secretly submitted its IPO application in June. The prospectus indicates that the number of shares offered, price range, and underwriting discounts are left blank on the cover page; a total of 21 underwriters are involved in this IPO, with Goldman Sachs and Morgan Stanley serving as joint lead underwriters.

Given that the original document is hundreds of pages long and some of its data differs significantly from the earlier draft exposed by Reuters, Odaily Planet Daily has excerpted the key IPO details for readers to quickly review.

Operating loss of $693 million, not $8.7 billion

Since the draft prospectus content was reported by Reuters at the end of September, Chinese media have generally portrayed Anthropic's operating loss for the first half of 2026 as approximately $8.7 billion, while the operating loss line item in the prospectus’s income statement is -$693 million.

The $8.7 billion is calculated by subtracting the $72.388 billion fair value change loss from the net loss of $80.748 billion. The prospectus explains this change as: “These changes were primarily due to the increase in the fair value of our outstanding convertible notes, resulting from increases in enterprise value during each period.”

The net operating loss narrowed from $2.135 billion in the first half of 2025 to $693 million, a 67.5% decrease. According to the prospectus, revenue during the same period increased 14-fold year-over-year.

The following are the primary consolidated income statement data (in millions of U.S. dollars):

Cash and short-term investments amount to $131.651 billion.

The widely circulated figure of “cash of only $20.28 billion” corresponds to the point in time as of December 31, 2025. The prospectus’s balance sheet has been updated to June 30, 2026: cash and cash equivalents amount to $112.389 billion, short-term investments total $19.262 billion, for a combined total of $131.651 billion. The cash flow statement shows that net cash inflows from financing activities in the first half of 2026 were $110.774 billion, with cash and restricted cash at the end of the period totaling $114.918 billion.

The claim of "$518 billion in computing power commitments" is not supported by the document.

The previously leaked draft stated, "Anthropic plans to spend $518 billion on cloud computing, computing power, and infrastructure over the next year." However, a full search of the prospectus for "518 billion" and "450 billion" yields no results, and the contract commitment table does not list specific amounts at this preliminary prospectus stage. The contract amounts with each counterparty can be verified individually:

Microsoft: at least $44.9 billion, period from November 2026 to May 2033. Google: at least $114.5 billion, period from April 2026 to July 2033. Amazon: at least $110 billion, period from May 2026 to April 2036.

SpaceXAI up to $120.5 billion by 2029. Nscale up to $44.6 billion by 2033. Lambda $41.3 billion by 2034. Initial procurement of AMD Helios racks expected to exceed $20 billion.

An additional $125 billion to $150 billion in Google TPU purchases is planned, with delivery scheduled between 2028 and 2031. Broadcom assumes equipment leasing obligations of up to $42 billion.

Both the Microsoft and Amazon agreements include make-up clauses. The prospectus states that if actual consumption falls below the committed amount, a payment must be made to Google to cover the difference, with such difference settled via a true-up mechanism; the Amazon agreement has the same arrangement.

The figures for customer concentration are in the notes to the financial statements.

The main text of the prospectus states only that “a significant portion of our revenue comes from a limited number of customers,” but does not provide specific percentages; the exact figures are found in Note F-16: In 2025, Customer A and Customer B each accounted for 12% of revenue, Customer B accounted for 18% in the first half of 2025, Customer C accounted for 43% in 2023, and declined to 11% in the first half of 2026.

The same footnote also discloses that revenue generated through the cloud service provider (CSP) channel accounted for 11% of total revenue in 2023, 32% in 2024, 47% in 2025, and rose to 48% in the first half of 2026. The proportion of net accounts receivable generated through the same channel was 42% in 2024, 60% in 2025, and 58% as of June 2026.

The prospectus lists dependence on cloud providers as a risk factor: “We rely on a limited number of cloud service providers and suppliers, some of whom have strategic and investment relationships with us and are also our customers. This relationship may create complex situations that could lead to conflicts of interest and adversely affect our ability to secure computing power.”

Net income retention rate is 900%

The 500% figure cited in the market comes from CFO Krishna Rao’s podcast interview in May 2026. The prospectus discloses that, as of September 30, 2026, the net revenue retention rate (NRR) was approximately 900%.

The prospectus also specifies the calculation scope for this metric: “The base period is the three months ending prior to the measurement date, and the numerator is the revenue for the three months ending on the measurement date; customers with revenue below $1 during the base period, as well as Claude Pro and Claude Max fixed-subscription users, are excluded from the calculation.”

The prospectus defines Run-Rate Revenue in the terminology section as: the annualized estimated value of our revenue-generating activities as of a given measurement date, calculated as the sum of (i) the consumer fees incurred over the past 28 days ending on that measurement date multiplied by 13, and (ii) the total monthly subscription fees as of that measurement date multiplied by 12.

The consumption fee is calculated by multiplying the past 28 days by 13, not the industry standard of 12.

The specific value of this metric is left blank in the prospectus, disclosed as: "As of October 31, 2026, our Run-Rate Revenue was ___ USD."

The risk factors section of the prospectus qualifies this metric as follows: “Therefore, these metrics may not accurately reflect our actual performance or future results. Metrics such as Run-Rate Revenue, net revenue retention rate, number of customers, and Claude success rate are based on numerous assumptions and limitations.”

The quarterly revenue disclosed in the prospectus is: $4.7 billion in the first quarter of 2026, $11.6 billion in the second quarter, and $16.8 billion in the third quarter. The sequential growth rate for the second quarter was 132%, and for the third quarter, it was 45%.

Monthly annualized recurring revenue (ARR) trajectory: $9 billion by end of 2025, $30 billion in February 2026, $47 billion in May, $65 billion in July, and $67 billion by end of Q3.

The company provided three explanations for the slowdown in growth: competitors such as OpenAI launched GPT-5.6 and GPT-6 Astra while lowering prices to compete in the agent market; the company’s own anti-distillation measures resulted in account suspensions and customer attrition; and approximately 5% to 7% of revenue came from model distillation activities, which are expected to decline further in the fourth quarter.

The prospectus also discloses a difference in reporting methodology: revenue figures include sales through the cloud platform channel without deducting platform commissions, whereas OpenAI’s disclosed annualized revenue reflects net revenue after deducting commissions, making the two figures not directly comparable.

The capital structure disclosed on the prospectus cover page includes five categories: 1 share of Class T common stock, 1 share of Class F common stock, and 0 shares of Class E common stock, as shown below:

Class T is held by the Long-Term Benefit Trust (LTBT). The prospectus describes the trust’s mandate as: “Operating as an independent entity governed by trustees with no financial interest, who possess expertise in AI safety, national security, public policy, and social enterprise.”

The trust may elect and remove Class T directors, and the Class T directors elected thereby shall constitute a majority of the authorized board seats. Class T common stock has no other voting rights or economic rights.

Class F shares will be issued prior to this offering to Founder LLC, composed of seven co-founders, holding 50.1% of the total voting power. Class N common shares have no voting rights and are held by key strategic business partners. Anthropic considers itself a "controlled company" under Nasdaq rules.

The risk factors section of the prospectus states: “Our capital structure includes several recently designed provisions not commonly found in other companies with multiple-class share structures, which may substantially eliminate the ability of Class A common stockholders to influence significant decisions, including changes in control. Our capital structure will, for the foreseeable future, concentrate voting power in LTBT and Founder LLC, resulting in these entities’ voting rights not being derived from or aligned with their economic interests in the company.”

The governance structure section further states: "Therefore, holders of Class A common stock will typically be unable to influence the outcome of any matter submitted to shareholders for a vote or the election of directors, regardless of how they vote—unless a majority of the continuing founders cannot agree on how to vote the Class F common stock."

The company has donated 846,840,960 shares of Class N common stock to its wholly owned subsidiary, the Anthropic Public Benefit Fund, representing 5% of the fully diluted shares as of June 30, 2026. The board has also been authorized to issue additional Class N common stock, up to 2% of the total outstanding shares, unexercised options, unvested RSUs, and other equivalent securities at any time the stock price increases by 50% from the previous price milestone, with the IPO offering price serving as the starting point.

This authorization is exercised at the discretion of the board. The seven co-founders have separately pledged to dedicate 80% of their personal Anthropic equity to charitable purposes.

The list of directors and senior management disclosed in the prospectus (as of September 30, 2026):

The company has four committees: Audit, Compensation, Nominating and Governance (NGC), and Security and Risk. The Audit and Compensation committees are composed entirely of independent directors. Jay Kreps resigned as a director in May 2026, Christopher Liddell joined in February 2026, and Vas Narasimhan joined in April 2026.

The prospectus's section "Major Shareholders and Selling Shareholders" states that the five co-founders, as selling shareholders, are reducing their Class A ordinary shares:

The company explicitly states: “We are selling Class A ordinary shares, and the selling shareholders listed in this prospectus are separately selling certain Class A ordinary shares. The company will not receive any proceeds from the sale of our Class A ordinary shares by any selling shareholder.”

The principal parties to the underwriting agreement are also three: “we, the selling shareholders, and the following underwriters will enter into an underwriting agreement regarding the offering of Class A ordinary shares.”

According to the prospectus, the primary objectives of this offering include "facilitating an orderly distribution of shares by selling shareholders," meaning it primarily serves the selling shareholders'减持 arrangements.

It should be noted that Anthropic has consistently had significant computational power funding needs, and the market has long anticipated a large-scale primary (new share) financing round. The publicly filed S-1 has not yet finalized the specific ratio between primary (new shares) and secondary (existing shares), so this offering cannot be definitively characterized as solely a secondary exit.

As of September 30, 2026, the company had approximately 5,300 full-time employees, the majority of whom were employed in research, engineering, and product development. Revenue per employee for the trailing twelve months (TTM) increased from $1.4 million to over $6.6 million, while headcount grew by approximately 170% during the same period, corresponding to a TTM revenue growth of approximately 1,200%.

The business section discloses that by September 2026, over 90% of the code contributions in the company’s codebase were written by Claude, and the average daily code output per person in the third quarter exceeded 45 times the full-year average of 2024.

In terms of customer structure, the number of customers contributing over $100,000 in TTM revenue increased from more than 950 as of September 30, 2025, to over 11,900; those contributing over $1 million reached more than 1,900, and those contributing over $10 million totaled approximately 250. According to the prospectus, the former group includes all of the top 10 companies on the Fortune list, more than half of the Fortune 500, and over one-third of the Global 2000.

The prospectus also discloses that revenue is concentrated among high-usage customers: in the third quarter of 2026, Claude Enterprise users who did not use Claude Code spent less than $75 per month on average, while the top 5% of users who used Claude Code spent approximately 20 times that amount.

The risk factors section accounts for approximately one-third of the main text. The prospectus states regarding the potential consequences of advanced AI systems: “Advanced AI systems, if developed, deployed, or operated without appropriate safeguards, oversight, and security measures, may pose catastrophic or existential risks to humanity, particularly when such systems are misused, replicated, weaponized, or permitted to operate with high levels of autonomy.”

Regarding model behavior, the prospectus states that the models, including Anthropic’s proprietary models, have exhibited and may in the future exhibit unexpected strategic or self-preserving behaviors: “attempting to resist shutdown, conceal or manipulate information, or generate outputs that can be interpreted as coercive, deceptive, or manipulative, including behaviors similar to extortion, or attempting to influence human decision-makers to maintain the model’s continued operation or access.”

Regarding the consequences of autonomous agent execution, the prospectus warns: "Errors, goal misalignment, or security vulnerabilities may lead to real-world consequences—including those that are difficult or impossible to reverse, such as submitting code changes, deleting or modifying data, or completing financial transactions—where effective intervention may not occur in time."

The company also disclosed the relationship between its own development speed and security investment: “We have observed that our models are accelerating the pace of our own research. We must allocate limited resources, capital, and talent between security-related work and other initiatives, and our investments in security often lack clear returns.”

Regarding product trade-offs, the prospectus states: “We chose not to develop certain commercially attractive products, such as image and video generation models, in order to allocate our computing resources toward our research and security priorities.”

Full-text search shows 12 references related to safety research, with no specific expenditure amounts disclosed.

The business section "Powerful AI" lists five features:

Cognitive abilities reach or exceed those of top human experts in most professional and scientific domains; capable of interacting with all interfaces accessible to human virtual work, including text, audio, video, and direct computer control; able to autonomously reason through complex tasks over extended periods—lasting hours, days, or even weeks—and proactively seek clarification and feedback when needed; capable of coordinating across multiple instances and collaborating with humans and other AI systems; able to interact with the physical world through digital connections, controlling laboratory equipment, robotic systems, and other machines.

The risk factors section of the same document states that advanced AI may pose an existential risk to humanity.

This preliminary prospectus still contains some blanks that will be filled during the roadshow phase: the number of shares to be issued and the price range (cover page), the specific Run-Rate Revenue figures, and the amounts for each item in the contract commitment table.

There are a total of 21 underwriters. Goldman Sachs and Morgan Stanley serve as joint representatives and have the option to purchase additional shares within 30 days.

Other underwriters include JPMorgan Chase, Citigroup, Barclays, Wells Fargo, BofA Securities, Deutsche Bank, RBC Capital Markets, UBS, BMO, BNP Paribas, Credit Agricole, Jefferies, Mizuho, Morgan Mitsubishi, Société Générale, TD Securities, BTIG, Cantor Fitzgerald, Capital One, and Macquarie.

The offering timeline disclosed in the prospectus indicates that, under SEC rules, companies using confidential filing must publicly submit their registration documents at least 15 calendar days before initiating the roadshow. Based on this calculation, launching the roadshow during the week of November 9 is procedurally feasible.