Anthropic Files for IPO with $1.5–2 Trillion Valuation Target

Author: Long Yue

On October 26, Anthropic officially filed its IPO application with the U.S. Securities and Exchange Commission, aiming to list on Nasdaq under the ticker symbol ANT, with a targeted valuation of $1.5 trillion to $2 trillion—potentially surpassing SpaceX’s historic record. Media reports indicate the IPO window is set for the week of November 16.

On that day, the U.S. semiconductor sector rallied—Nasdaq rose over 1% intraday, with Broadcom, AMD, and NVIDIA each climbing nearly 3%. Although Anthropic has not yet gone public, its suppliers were the first to surge. Anthropic has already signed power contracts totaling over $800 billion, extending as far as 2045. To the market, these long-term contracts may already represent guaranteed revenue firmly reflected on the balance sheets of major hardware giants.

This S-1, exceeding 300 pages and approximately 240,000 English words, gives the public its first opportunity to fully examine the financial foundation and governance logic of the world’s leading AI laboratory. The Financial Times of the UK has characterized it as a “referendum”—testing how much public market investors endorse Dario Amodei’s narrative of using mission-driven appeal to win over private investors. Based on current market sentiment, Amodei’s narrative appears to be resonating well.

Beyond the narrative, the financial data and details in the prospectus are equally striking.

In terms of revenue, Anthropic reported $376 million in revenue for the first quarter of 2025 and $11.555 billion in the second quarter of this year, representing a roughly 30-fold increase over five quarters. Regarding profitability, the company incurred a net loss of $80.7 billion in the first half of the year, but its operating profit turned positive in the second quarter at $974 million; the difference primarily stemmed from approximately $72.4 billion in fair value changes, with no cash outflow. In terms of control, Amazon and Google hold non-voting Class N shares, while 50.1% of the voting rights are held by a single Class F share, each purchased for $1 by the seven co-founders.

Meanwhile, Anthropic’s revenue growth from Q2 to Q3 declined to nearly 50%, with the top five customers accounting for a quarter of revenue over the first nine months. In over 90 pages of risk factors in the prospectus, the company stated that its technology could pose “catastrophic or existential risks.”

How dramatic is the growth? Anthropic’s S-1 provides its own data comparison.

According to the prospectus, Anthropic generated $410 million in revenue in 2023, increased to $3.86 billion in 2024, and further reached $45.85 billion in 2025. Growth accelerated further in 2026, with the company achieving $16.286 billion in revenue during the first half of the year. On a quarterly basis, Anthropic's revenue was $376 million in Q1 2025 and $11.555 billion in Q2 this year—an increase of approximately 30 times over five quarters. The intervening four quarters were $787 million, $1.382 billion, $2.04 billion, and $4.731 billion, respectively. This translates to approximately $127 million in daily revenue during the second quarter.

Through the first nine months of this year, media reports show that Anthropic's revenue increased from $4.6 billion last year to over $33 billion for the first nine months of this year, with per-person revenue of $6.6 million over the past year.

Sales and marketing expenses surged in tandem: $568 million for the full year of 2025, reaching $3.168 billion in the first half of this year alone.

However, the growth rate has slowed. Revenue more than doubled from the first quarter to the second quarter this year, and increased by nearly 50% from the second to the third quarter. Anthropic cautions in its prospectus that its rapid growth to date "may reflect early adoption" and could stall in the future.

Net loss of $80.748 billion in the first half of the year may seem alarming, but breaking down the income statement reveals that $72.388 billion stems from fair value changes in convertible notes, derivative liabilities, and SAFE instruments.

The mechanism is not complicated. Part of the money Amazon invested in Anthropic consists of convertible notes—essentially IOUs that state the holder can convert them into shares at a predetermined price in the future. Anthropic’s valuation rose from $380 billion in February to $965 billion in May, causing the value of these notes to increase significantly. Accounting standards require this appreciation to be recorded as a loss on the company’s books. The prospectus also notes that such changes are primarily driven by valuation increases and do not involve any cash outflows.

According to the company's adjusted (non-GAAP) metrics, the adjusted operating profit for the first half of the year was $1.618 billion, primarily excluding stock-based compensation and non-cash charitable donation expenses; the adjusted net profit was $2.671 billion.

According to the Financial Times, in the first quarter of 2025, Anthropic spent more than $2.50 on compute and infrastructure for every dollar of revenue; by the second quarter of 2026, this figure had dropped to less than 50 cents.

The prospectus provides annual figures: in 2023, computing power expenditures exceeded revenue by more than 10 times; in 2024, they were 653%; in 2025, 160%; and in the first half of 2026, 55%.

Profit turned positive. For the three months ended June 30, Anthropic achieved an operating profit margin of 8% and an operating profit of $1 billion, compared to an operating loss of $1.2 billion in the same period last year. Specifically, the second-quarter operating profit was $974 million; after adding back non-cash expenses such as stock-based compensation, adjusted operating profit reached $2.235 billion, with a margin of 19%. Net cash inflow from operating activities for the first half of the year was $3.9 billion.

The prospectus summarizes four reasons:

• Improved training efficiency. Claude is increasingly involved in the development of next-generation models, an internal company process referred to as "closing the loop";
• Improved inference efficiency reduces the computing power required for the same level of performance;
• Train and run models simultaneously on Amazon Trainium, Google TPU, NVIDIA GPU, and AMD GPU chips, while building our own scheduling system;
• New model price reductions. After the release of Opus 4.5 in November 2025, prices and per-task costs decreased, causing quarterly revenue growth to rise from 48% in the previous quarter to 132% in the first quarter of 2026; in September this year, the release of Opus 5.5 and Fable 5.1 further reduced per-task costs by 25% to 45%.

At the same time, the company warned that performance may experience "significant period-to-period fluctuations" due to hash rate expenses.

In the first nine months of this year, five customers accounted for 25% of Anthropic’s revenue, and only 250 customers spent more than $10 million annually.

From a quarterly trend perspective, the share of top clients has been declining. The revenue contribution from the top five customers decreased from approximately 36% in Q1 to about 21% in Q3, with 79% of Q3 revenue coming from clients outside the top five, and no single customer accounting for more than 10%. During the same period, the absolute revenue from the top five customers increased from approximately $1.7 billion to about $3.5 billion. The prospectus also shows that approximately 48% of revenue comes from cloud service providers’ distribution channels.

Change in customer count: As of the end of 2023, there were approximately 20 customers spending over $100,000 annually; by September 2025, this number had grown to about 950, and by September 2026, it exceeded 11,900. Additionally, there are approximately 40 million customers spending less than $100,000 annually. More than half of the Fortune 500 companies are now customers spending over $100,000 per year, but their spending on Anthropic averages less than 0.1% of their total expenses, with the top 10 users still spending less than 1%.

The net revenue retention rate is approximately 900%, meaning that customers who spent $1 a year ago are now spending an average of $9 during the same period this year. This figure excludes subscription users.

On the revenue side, pay-as-you-go revenue amounted to $3.8 billion in 2025, accounting for 83%, and $13.1 billion in the first half of 2026, accounting for 80%; subscription revenue in the first half was $3.2 billion, compared to $180 million in the same period last year. Over 80% of revenue in the third quarter came from model versions released within the past six months.

The prospectus disclosed that, for the first nine months ended September 30, approximately 5% to 7% of Anthropic’s revenue came from customers later found to have violated usage policies or terms of service. This occurred due to a time lag between detection, investigation, and account suspension; revenue was still recognized for services provided prior to suspension.

The company also noted that increased efforts to identify and ban such activities in August and September are expected to reduce the proportion of this revenue, potentially lowering future income and growth rates. Based on approximately $33 billion in revenue for the first nine months, 5% to 7% equates to roughly $1.6 billion to $2.3 billion.

Anthropic refers to computing power as a "critical input" and has committed to agreements totaling over $800 billion, extending as far as 2045. The founders acknowledge that these investments may "appear recklessly large at first glance," but describe their approach as "bold but rational."

Main contracts include:

• Broadcom: Broadcom partnered with Google to develop TPUs and leased them to Anthropic under a five-year contract totaling $161 billion; starting in 2028, Anthropic is expected to spend up to an additional $150 billion on TPUs;
• AMD: Expected expenditures exceed $20 billion, with additional options to roughly double the scale; AMD will invest up to $5 billion in Anthropic upon chip deliveries;
• Google, Amazon, Microsoft: Together, they will provide at least $269.4 billion in cloud computing power over the next decade;
• Nscale, Lambda: Contract amount up to $86 billion.

The contract durations vary among the three major cloud providers: Google has a minimum commitment of $114.5 billion, spanning from April 2026 to July 2033; Amazon has a minimum commitment of $110 billion, extending to April 2036; and Microsoft has a minimum commitment of $44.9 billion, extending to May 2033. All three contracts include "regardless of actual usage" payment terms, and Google’s and Amazon’s contracts further stipulate make-up payments for underutilization. Such "take-or-pay" clauses are common in long-term agreements for energy sources such as natural gas and electricity.

According to the financial statement notes in the prospectus: As of June 30, the company had signed fixed or guaranteed purchase commitments totaling $265.6 billion, data center and computing power leases that were contracted but not yet commenced amounting to $321.4 billion, and new or amended long-term contracts signed after June 30 totaling $245 billion, for a combined total of over $830 billion. As of June 30, the company held approximately $131.6 billion in cash and short-term investments, and raised approximately $110.4 billion during the first half of the year through its G and H rounds of financing.

The mismatch between cost and revenue timing is also a key focus for investors. The prospectus shows that approximately 80% of revenue is settled based on actual usage, meaning revenue depends on customer consumption, while computing power costs are often locked in more than a year in advance.

The total contract value between Anthropic and SpaceX is up to $120.5 billion, lasting until 2029. After all computing power comes online in December, monthly costs are expected to be approximately $3.1 billion; starting early next year, either party may terminate with 90 days' notice.

Looking back on this relationship, there have been several turning points. In January, Anthropic revoked xAI employees’ access to Claude via Cursor. In February, after Anthropic announced its funding round, Musk publicly criticized its AI. In May, Anthropic announced it was leasing xAI’s Colossus 1 data center in Memphis, after which Musk reversed his stance, praising Anthropic’s team for their competence and commitment to doing the right thing. The contracting party in the prospectus is SpaceX.AI LLC, formerly known as X.AI LLC, and now a wholly owned subsidiary of SpaceX.

According to SpaceX’s prospectus, its entire AI business generated $818 million in revenue in the first quarter of this year. Anthropic’s monthly compute costs are nearly four times this amount.

The Financial Times of London estimates that if Anthropic were to go public at a $2 trillion valuation, Amazon would hold approximately one-fifth, or $400 billion, after all convertible notes are converted. Amazon’s ownership cap is 12%, with any excess held in note form and gradually converted as capacity becomes available; however, Amazon may waive this cap.

Google holds approximately 7% and may reinvest up to $20 billion under an agreement earlier this year, with $10 billion available for subscription at the offering price; Google may also serve as a cornerstone investor in the IPO, subject to mutual agreement.

Among financial investors, documents show that Spark Capital invested less than $1 billion, with its stake valued at $14.5 billion; according to a person familiar with the matter, Coatue invested $4.75 billion, with a stake size comparable to that amount.

The lock-up schedule is also unusual. Similar to SpaceX, Anthropic has five lock-up expiration dates within 76 days after its IPO, allowing non-founder directors, executives, and certain current and former employees to sell a portion of their shares on the first day of trading. Typically, early investors must wait 180 days after the IPO before selling.

On the employee side, the prospectus simulated a listing as of June 30, estimating total incentive expenses of approximately $29.7 billion, averaging over $5.6 million per employee based on roughly 5,300 employees. In an internal secondary share transfer in April, insider shareholders, including the Amodei siblings and the CFO, cashed out $4.66 billion.

Anthropic acknowledged in its prospectus that its "novel" governance structure differs significantly from that of most public companies, with common shareholders having little to no control over corporate decisions.

The structure of the media-organized entity is as follows: At the top is the Long-Term Benefit Trust (LTBT), whose members appoint themselves and typically have the authority to elect and remove four of the seven directors; Dario Amodei and Daniela Amodei cannot be removed by the trust. The remaining three directors are nominally elected by shareholders but are in fact selected by the two siblings and five other co-founders, as they hold 50.1% of the voting rights through a special class of shares. Control over the fourth seat by the trust requires approval from the nominating committee, which includes Daniela Amodei; in the event of a deadlock, a tie-breaking vote is cast by a representative of the employees.

The details of this special stock are as follows: Prior to the listing, the company will issue a single Class F share to Founder LLC, composed of seven co-founders. This share has no economic rights but holds 50.1% of all voting rights, including the power to elect all directors except Class T directors.

Conversely, N shares held by major shareholders such as Amazon and Google have no voting rights. The prospectus also states that the company qualifies as a "controlled company" under Nasdaq rules and is exempt from certain corporate governance requirements.

Prior to its listing, Anthropic allocated 846,840,960 Class N shares to its wholly owned subsidiary, the Anthropic Public Benefit Foundation, representing 5% of the fully diluted shares.

The board also approved a framework: for every 50% increase in stock price above the IPO offering price, an additional 2% of shares may be contributed to the fund. Each of the seven co-founders has committed to dedicating 80% of their personal equity to philanthropy.

The risk factors in the prospectus are about three times longer than those in SpaceX’s similar disclosures and differ significantly from the generic, boilerplate disclosures common in corporate filings. The document warns that its technology could pose "catastrophic or existential risks" to humanity and might "substantially assist malicious actors seeking to develop chemical, biological, or other dangerous weapons."

The失控 scenarios described in the document include the model potentially developing "unintended... self-preservation behaviors" in the future, such as attempting to resist shutdown, concealing or manipulating information, and generating coercive, deceptive outputs similar to "extortion." Regarding whether these risks are covered by insurance, the prospectus states that no single insurer, nor the entire insurance industry combined, may be able to provide coverage.

The risk factors also note that the company is increasingly using AI to accelerate its own research and development, which may cause improvements in model capabilities to outpace the development and implementation of safety measures; the company has previously delayed model releases for this reason, such as the limited release of Mythos through Project Glasswing.

At the policy level, the document acknowledges that the U.S. government may increase costs, restrict operations, or introduce new regulations affecting demand for the model. Earlier this year, Anthropic was designated as a "supply chain risk" by the U.S. Department of Defense, and government agency contract revenue accounts for approximately 1% of the company’s total revenue. The conclusion of this section in the prospectus states: "Many of the risks we face are interconnected and may compound or trigger cascading adverse effects."

The seven founders advised potential investors in the document to "not view your purchase purely from a trading perspective," acknowledging that the company’s decisions "may contradict investor expectations," as profit maximization must be balanced against the "public interest," and some business bets "may seem strange or unconventional in isolation." Media noted that the founding team’s emphasis on still being in senior leadership positions is "uncommon," and described it as an "oblique jab" at OpenAI, which has lost most of its founding team.

The prospectus describes the company culture as "low-ego." According to leaks, Anthropic employees internally refer to themselves as "Ants," aligning with the ticker symbol.

The prospectus defines Run-Rate Revenue as: the usage-based fees from the last 28 days multiplied by 13, plus the monthly subscription fees multiplied by 12. The column for October is currently blank in the document and is expected to be updated in the S-1 filing around mid-next week. In market discussions, a figure above $73 billion is considered a reasonable benchmark; if it exceeds $78 billion, the topic of reaching $100 billion by December will regain momentum.

Matt Kennedy, Senior Strategist at Renaissance Capital, which specializes in IPO research, told Reuters that no company wants to drag things out too long at the critical juncture before going public; although they might achieve a higher valuation next year, they decided to "seize the moment."