Behind the $80.7 Billion Paper Loss: Anthropic’s Real Cash Burn Rate and the Logic of a $2 Trillion IPO

In October 2026, a draft IPO roadshow document cited by The Wall Street Journal exposed the financial reality of the AI industry with rare clarity: Anthropic recorded a net loss of $80.7 billion in the first half of 2026, more than six times the $13 billion loss in the same period a year earlier. The figure quickly became a headline, but it is an almost misleading signal.

Accounting Magic: A $72 Billion Paper Game

According to The Wall Street Journal, $72 billion of the $80.7 billion loss came from accounting adjustments arising from changes in the valuation of convertible notes and other financing instruments, which are non-cash items. The mechanism is not unfamiliar: when the convertible bonds issued in Anthropic’s various financing rounds increase in value—meaning the higher the company’s valuation, the larger these “paper losses” become. In other words, the more the market recognizes Anthropic’s valuation, the worse its GAAP book loss appears.

After stripping out the $72 billion in non-cash adjustments, the company’s actual operating loss was about $8.7 billion. That is still a huge sum, but compared with $80.7 billion, it is different in order of magnitude and entirely different in nature. The former is an accounting consequence of the financing structure; the latter is the actual rate at which the company consumes cash to sustain operations.

Anthropic’s compute and infrastructure spending was reportedly $7.3 billion for all of 2025, while in the first half of 2026 alone, the figure already reached nearly $9 billion, exceeding the full-year total. The acceleration in infrastructure spending is the real cash-burn signal.

Revenue Explosion: The Substance of 15x Growth

At the same time, Anthropic delivered a revenue report card. According to The Wall Street Journal, revenue reached $16.3 billion in the first half of 2026, compared with just $1.1 billion in the same period of 2025, representing year-over-year growth of about 15x. That pace exceeds most historical precedents in the SaaS industry.

Q3 data further reinforced the trend: revenue was $16.8 billion, up 45% quarter over quarter, and the company disclosed that it had achieved profitability on an adjusted basis. It is worth noting that there is an essential difference between “adjusted profitability” and GAAP profitability—the former typically excludes items such as stock-based compensation, depreciation, and amortization, and is an internally defined metric used by tech companies; it is not equivalent to positive cash flow.

However, there is one data point investors should not overlook: about one-fifth of Anthropic’s 2026 revenue reportedly comes from just two customers. Such high customer concentration means that budget changes or strategic shifts at top customers could directly deal a severe blow to the company’s revenue—this is the fragility hidden behind the $16.3 billion.

A $750 Billion Compute Gamble: Structural Dependence Has Formed

The longer-term challenge lies in the scale of capital requirements. According to The Wall Street Journal, Anthropic disclosed that it may spend another $750 billion over the coming years to purchase compute, with some agreements still under negotiation. Amazon has reportedly made a 10-year commitment for AWS compute capacity to provide compute support for the Claude model.

This means Anthropic is already deeply tied to cloud infrastructure suppliers, and its competitiveness depends to a large extent on whether it can continuously secure scarce compute and on its bargaining relationship with major cloud providers. The $750 billion commitment—even if some of it has not yet been signed—reveals a harsh reality: the moat of top AI models is currently composed mainly of capital intensity rather than technical secrets. Only those who can keep burning money to buy compute can stay at the table.

A $2 Trillion Valuation: Has the Market Lost Its Judgment?

According to Reuters and multiple reports, Anthropic plans to go public around November 2026, with a target valuation of about $2 trillion. Some reports cite slightly different figures (ranging from $1.5 trillion to $2 trillion), but the order of magnitude is broadly consistent.

Based on H1 2026 revenue of $16.3 billion, a linear annualization would imply about $32.6 billion for the full year, corresponding to a price-to-sales (P/S) ratio of about 60x at the current valuation. Even using the company’s own target revenue of $34.5 billion for 2027, the corresponding P/S ratio is around 58x. That valuation level would be astronomical in any traditional industry, but AI infrastructure companies have their own special logic: when the growth rate itself is exponential and the market is not yet saturated, the P/S metric has limited meaning.

The real question is: how long can this growth rate be sustained? Anthropic’s own risk disclosures have already provided part of the answer—high customer concentration, reliance on external compute, and competitors (OpenAI and Google Gemini) accelerating in parallel—these risk factors also appear in the roadshow materials.

Independent Judgment: Reading the Real Bet Behind the Numbers

The $80.7 billion net loss is an accounting number, not an operating number. The metrics that truly need watching are: where operating losses narrow to from $8.1 billion for full-year 2025, whether quarterly revenue can maintain quarter-over-quarter growth of more than 40%, and the contract renewal situation with the two top customers.

Anthropic’s current financial structure displays a clear internal logic: offset massive infrastructure investment with ultra-fast revenue growth, while using the accounting characteristics of financing instruments to make the GAAP loss figure look more “dramatic,” thereby reinforcing the narrative that “we are the most important bet in history” and creating a psychological anchor for an IPO at a $2 trillion valuation.

This logic is not without historical precedent. But history also tells us: when one-fifth of a company’s revenue depends on two customers and future compute spending commitments are as high as $750 billion, any statement about “having turned profitable” requires carefully reading the notes in parentheses. Adjusted profit is not profit, and a paper loss is not cash burn; the distance between the two is the entire suspense over the reasonableness of this company’s valuation.