Deep Dive into Anthropic’s Prospectus: Targeting $2 Trillion Valuation, Warns of “Potential Threat to Humanity”

Written by: Chaoxiang Research

On October 26, Anthropic filed an S-1 form with the U.S. Securities and Exchange Commission (SEC), officially launching its IPO process.

Intended listing venue: NASDAQ; Ticker: ANT.

The underwriting syndicate reads like an all-star team: led by Goldman Sachs, Morgan Stanley, JPMorgan Chase, and Citigroup, with Barclays, Bank of America, Deutsche Bank, and 18 other institutions participating. According to Reuters citing insiders, market expectations place the valuation above $2 trillion.

If this expectation materializes, Anthropic will surpass SpaceX’s $1.77 trillion record set in June this year, becoming the highest-valued IPO in human history.

But what truly stands out in the S-1 prospectus is how the company uses a legal document to lay bare the core contradictions of the AI industry before everyone.

Growth Curve: From $41 Million to $32.6 Billion

Let’s look at the financial data first.

Full-year revenue in 2023 was $41 million; in 2024, $386 million; in 2025, $4.585 billion; and in the first half of 2026, $16.286 billion.

In three and a half years, revenue has nearly increased 800-fold. By late July 2026, the annualized run rate had already surpassed $65 billion, with some media outlets estimating full-year revenue could reach $80 billion to $100 billion.

What does that mean? Netflix took 20 years to achieve over $30 billion in annual revenue, while Anthropic has only been established for five years.

Enterprise customer data is equally astonishing.

As of September 30, there were over 11,900 customers spending more than $100,000 annually, over 1,900 spending more than $1 million, and approximately 250 spending more than $10 million. The net revenue retention rate is around 900%, meaning existing customers' spend increases ninefold within a year.

Employee productivity is even more staggering. The S-1 discloses that as of late September, annual revenue per employee exceeded $6.6 million, up from $1.4 million a year earlier. Based on third-party estimates, Anthropic employs roughly 3,000 to 5,000 people. For comparison, Google had 32,000 employees when it reached $30 billion in annual revenue, and Salesforce had 79,000 employees at the same revenue level.

But revenue is only one side of the story. On the flip side, the company is simultaneously burning cash at an almost terrifying pace.

$518 Billion: A Decade-Long Compute Arms Race Bill

The number in the S-1 that sends shivers down Wall Street's spine is the infrastructure commitment.

The prospectus reveals Anthropic has committed to over $518 billion in compute and infrastructure purchase commitments and leases, mostly irrevocable contracts spanning the next decade. Approximately 80% of this amount operates on a "use-it-or-pay-for-it" basis.

Who is receiving this money?

Amazon Web Services: ~$110 billion, ten-year term, up to 5 gigawatts of compute capacity.

Google Cloud: ~$111.1 billion, until July 2033.

Broadcom: ~$161.2 billion, primarily equipment leasing, mostly non-cancellable. Microsoft Azure: ~$31.4 billion, until May 2033.

xAI (based on NVIDIA capacity): Up to ~$84.5 billion, until 2029, though most can be canceled with 90 days' notice.

AMD: Over $20 billion, including AMD's reverse investment of up to $5 billion.

$518 billion is close to Sweden's GDP. In the tech sector, Microsoft's capital expenditure in fiscal 2025 was approximately $80 billion, already one of the highest levels in history. Yet this startup has signed commitments more than six times Microsoft's annual capex.

As of June 30, 2026, Anthropic held approximately $131.7 billion in cash, cash equivalents, and short-term investments. While this sounds like a lot, against the $518 billion bill, it would only cover about a quarter.

This means Anthropic must ensure its revenue growth outpaces its compute expenditures, otherwise these irrevocable commitments will turn into a massive penalty.

The good news is that early signs show Anthropic is temporarily leading this race. In the first half of 2026, revenue grew 1,368% year-over-year, while compute and infrastructure expenses grew 300% year-over-year. Revenue is growing faster than costs, and gross margins are improving. Reports indicate that Q2 2026 already marked the first adjusted profitable quarter.

The bad news is that this is just the beginning. Training costs for frontier models continue to climb, with industry estimates suggesting a single training run could exceed $30 billion by 2028.

The Secret Behind the Losses: Operations Are Already Near Break-Even

Flipping to the income statement, the picture looks terrifying.

Net losses were $42 billion for all of 2025, and $80.7 billion for the first half of 2026.

But there is a major trap behind these two numbers: the vast majority of the losses stem from non-cash fair value adjustments under accounting standards, which have nothing to do with actual cash burn.

Specifically, during early financing rounds, Anthropic issued numerous convertible notes, derivatives, and SAFEs. These financial instruments require fair value measurement. As Anthropic's corporate valuation skyrocketed, the fair value of these liabilities surged correspondingly, reflecting on the income statement as massive non-operating losses.

In 2025, this fair value adjustment loss amounted to $34.3 billion. In the first half of 2026, it soared to $72.4 billion.

In other words, the faster the company's valuation rises, the larger the book "losses" become. This is an intuitive-counteracting accounting phenomenon: the more valuable it gets, the more it appears to lose money.

Stripping out these paper losses and looking at operations: operating losses were $8.06 billion in 2025, narrowing significantly to $693 million in the first half of 2026. Given that revenue for these six months was $16.3 billion, the operating loss ratio has compressed from 176% in 2025 to 4.3%.

After deducting an additional $2.2 billion in stock-based compensation (also a non-cash expense), the operational level is already near break-even, and may even have turned positive.

However, another time bomb is buried on the income statement.

In 2025, Anthropic recorded a $1.5 billion settlement expense related to Bartz. This resulted from a copyright class-action lawsuit: authors accused Anthropic of using books downloaded from piracy sites to train Claude. The two parties reached a settlement in September 2025, involving approximately 500,000 works at an average of $3,000 per work. It is one of the largest copyright settlements in U.S. history.

This $1.5 billion is a one-time charge, but the risk it represents is long-term: the legal boundaries of training data for AI companies remain unsettled to date.

Money Passing from Left Hand to Right: The Dual Identity of Amazon and Google

One of the most intriguing sections of the S-1 is the disclosure regarding the relationship between Amazon and Google.

These two companies simultaneously play four roles: investors, cloud computing providers, distribution channels, and competitors.

Amazon holds significant equity in Anthropic through convertible notes, which will convert to voting-less Class N shares upon the IPO. Google disclosed via court filings in early 2025 that it holds roughly 14% stake, also subject to a maximum ownership cap. Together, they accounted for approximately 47% of Anthropic's 2025 revenue (by distributing Claude through their cloud platforms).

This forms a perfect closed-loop funding cycle: Amazon and Google invest in Anthropic → Anthropic spends that money on AWS and GCP to buy compute → Customers use Claude via AWS and GCP, generating revenue that flows back to Anthropic → Anthropic continues buying compute.

This flywheel operates beautifully, but the Risk Factors section of the S-1 bluntly states that these relationships "may create incentives that are not entirely aligned with Anthropic's own interests."

To translate: your largest supplier is also your largest competitor. Amazon has its own AI team and the Titan model, while Google has Gemini. While helping Anthropic sell Claude, they are simultaneously marketing their own competing products.

How long can this delicate balance of cooperation and competition last?

Equity Structure: A Constitutional Design for an AI Company

The financial data reveals Anthropic's commercial face. The governance architecture reveals the company's soul, or rather, the soul Dario Amodei wants to showcase.

Anthropic's equity is divided into five classes:

Class A: Publicly traded float, one share one vote.

Class N: Held by strategic partners (primarily Amazon), carries no voting rights, votes only when legally mandated, and automatically converts to Class A upon transfer to non-affiliates.

Class F: Only one share, held by the Founder LLC, with seven co-founders as members. This single share carries 50.1% of the ordinary voting rights.

Class T: Also only one share, held by the Long-Term Benefit Trust (LTBT), granting the right to elect a majority of the board seats.

Class E: Fractional share issuance, reserved for employees (excluding co-founders), with restricted voting rights.

This design makes Anthropic a legally defined "controlled company." The seven founders control 50.1% of the voting power through the Founder LLC, while five independent trustees of the LTBT control board composition.

The LTBT trustee lineup is noteworthy enough to highlight separately: Chair Neil Buddy Shah is the CEO of the Clinton Health Access Initiative; Richard Fontaine is the CEO of the Center for a New American Security; and newly joined in July 2026 is Ben Bernanke, former Chair of the Federal Reserve.

These individuals share a common trait: none hold equity in Anthropic. Their role is to ensure that Anthropic, as a Public Benefit Corporation (PBC), does not deviate from its AI safety mission for the sake of short-term profits.

As part of its PBC commitment, Anthropic will initially issue approximately 847 million Class N shares to the "Anthropic Charitable Foundation" at the IPO, representing 5% of the fully diluted share count. Thereafter, each time the stock price rises 50% beyond the previous milestone, the board may issue up to an additional 2% to the foundation. The seven co-founders have also individually pledged to donate 80% of their personal equity to charitable causes.

All of this combined, Anthropic attempts to answer an unprecedented question: how to install a moral pressure relief valve that can never be dismantled on a company listed on a publicly traded market obsessed with profit maximization?

This stands in stark contrast to OpenAI's path. OpenAI started as a nonprofit, but gradually turned the nonprofit shell into mere decoration, with Sam Altman himself admitting the governance structure was "not great." Anthropic's approach is to bake public benefit obligations into the company's DNA from the start, then hand the keys to a group with no financial ties to the money.

Will this design work? It is an experiment that will take ten or even twenty years to validate, but at least on paper, it raises the banner of ethics and charity.

80 Pages of Risk Factors

Every S-1 contains a dedicated chapter on risk factors, and for the vast majority of companies, it reads like a lawyer-written disclaimer: lengthy, abstract, and monotonous.

Anthropic's is different. The risk factors section of this prospectus is approximately 80 pages long, taking up nearly a third of the entire 261-page document. An AI company, in fact, uses a volume far exceeding its business description to tell, or perhaps intimidate, investors that this technology could destroy humanity.

Wording in the S-1 includes that AI models may "act in unexpected, misaligned, or autonomous ways," may "take harmful actions without instruction from users or third parties, and the rate of progress in model capabilities may outpace the development of safety measures."

Anthropic also warns inside that in the event of a catastrophic AI-related incident, the scale of losses could be so vast that insurance cannot cover them.

Such wording would be unimaginable in the S-1 of any other industry. But in the AI sector, it might precisely be the core of Anthropic's brand narrative: because we take this risk seriously, you should trust us to manage it.

This is a highly sophisticated form of "advancing by retreating," laying the most extreme risks on the table, which paradoxically becomes a brand asset.

What Is This IPO Actually Pricing?

Zooming out.

Anthropic's IPO is ostensibly pricing an AI company, but what is truly being priced is a much larger question: exactly how much are frontier AI large model companies worth?

Until now, all frontier AI labs have lived behind the curtain of the private markets. Private market pricing is a game among a select few institutional investors: opaque information, poor liquidity, and valuations open to arbitrary speculation. Anthropic's S-1 is the first to expose a frontier AI company's complete financial statements, customer data, cost structure, and governance architecture to the sunlight of the public markets.

This means that public market investors will finally have the opportunity to collectively assess the following questions:

What should be the risk premium for a company whose revenue grew 800-fold while simultaneously signing $518 billion in irrevocable commitments?

Will the market assign a premium or discount to an "safety-first" AI company compared to a "growth-first" AI company?

When your two largest suppliers are simultaneously your competitors and investors, how should this relationship be reflected in valuation?

Does a governance structure that hands board control to an external trust increase or decrease investor confidence?

The answers to these questions will directly impact the valuation anchors for OpenAI (reportedly planning to list in 2027) and all other AI companies.

After telling the story for so long, it's time to hand in the test paper.

Disclaimer: This article is for informational purposes only and does not constitute investment advice.