On October 26, 2026, Anthropic publicly filed its Form S-1 registration statement with the Securities and Exchange Commission, setting up a listing on the Nasdaq Global Select Market under the ticker ANT. The filing landed with two numbers that appear to tell opposite stories.

The first: revenue of roughly $16.29 billion for the six months ended June 30, 2026, up from $1.16 billion in the same period a year earlier. That is a fourteenfold increase, and more than triple the $4.59 billion the company generated across the whole of 2025.

The second: a net loss of $80.75 billion for those same six months. A loss roughly five times the size of revenue is not a normal thing to see in a prospectus. But about $72.4 billion of it, close to 90 percent, is a non-cash accounting charge tied to the rising fair value of convertible notes, derivative liabilities and simple agreements for future equity, driven mostly by the increase in Anthropic's own valuation. The separate operating loss line for the half was $693 million, narrower than the $2.14 billion operating loss a year earlier; the distance between that figure and the headline net loss is made up of the fair-value charge plus other non-cash items of roughly $9.4 billion, partly offset by about $1.1 billion of interest income and about $0.7 billion of warrant gains.

In this article we explore what that $72.4 billion charge actually is, why a company can be punished in its accounts for becoming more valuable, what Anthropic's underlying operating numbers show, where the genuine risks sit, and how an ordinary investor should read loss figures in the coming wave of artificial intelligence listings.

What a convertible note is, in plain words

A convertible note is a loan that is designed to turn into shares rather than be repaid in cash. An investor hands over money, and instead of getting the money back with interest, they get equity when some future event happens, typically a funding round or a stock market listing. A SAFE, a simple agreement for future equity, is a stripped-down cousin of the same idea: money now, shares later, no interest and no maturity date.

Private technology companies use these instruments constantly because they let an investor put capital in without the two sides having to agree today on exactly what the company is worth. The price gets settled later.

The accounting consequence is where things get strange. Because the holder is owed shares rather than a fixed sum of cash, the obligation is not a simple fixed debt. Under the relevant accounting rules, these instruments frequently have to be carried on the balance sheet as a liability measured at fair value, and remeasured at the end of every reporting period.

Why success produces a loss

Fair value simply means the amount the obligation would be worth if it were settled or traded today. For an instrument that converts into equity, its value moves with the value of that equity. If the company's valuation rises, the shares the holder will eventually receive are worth more, so the liability on the balance sheet gets bigger.

An increase in a liability has to be matched by a charge in the income statement. So the better the company does, and the higher its valuation climbs, the larger the paper loss it reports.

The pattern in Anthropic's own filing makes the mechanism unmistakable. The fair-value charge was $1.0 billion in 2023, $5.4 billion in 2024, $34.3 billion for full-year 2025, and $10.9 billion in the first half of 2025 before reaching $72.4 billion in the first half of 2026. That curve does not track compute bills or payroll. It tracks the company's valuation, which went from a $380 billion post-money figure in its February 2026 Series G round to a reported figure of around $965 billion in a further private round in May 2026, and then to a $1.32 trillion mark on Nasdaq Private Market in late September 2026.

The critical point for a reader trying to assess the business: no money left the building. Not a dollar of the $72.4 billion was paid to anyone. It is a revaluation of a promise to issue shares.

What happens to the charge

These instruments exist precisely because the company was private. Once a company lists, convertible notes and SAFEs generally convert into ordinary shares, the liability moves off the balance sheet and into equity, and the recurring remeasurement charge stops being a feature of the income statement. The economic cost is real, but it is borne by existing shareholders through dilution, not by the company through cash. Dilution and a cash loss are different things, and conflating them is the single most common error in reading a filing like this one.

The operating business underneath

With the accounting noise removed, a very different company appears.

  • Operating loss for the first half of 2026 was $693 million, down from $2.14 billion in the first half of 2025, even as revenue grew roughly fourteenfold.
  • The second quarter of 2026 produced $974 million of operating income, Anthropic's first quarterly operating profit, after a $1.7 billion operating loss in the quarter immediately before.
  • On an adjusted basis the company reported $1.62 billion of operating income for the half, a 10 percent adjusted operating margin.
  • Operating cash flow for the half was approximately $3.93 billion.

Compute and infrastructure expenses, the cost of the chips and data centres that run the Claude models, reached $8.95 billion in the first half of 2026, up from $2.33 billion a year earlier. That is a near fourfold increase. But revenue grew faster, which is why the margin flipped. For a business whose main variable cost is compute, the question of whether revenue grows faster than inference costs is close to the whole investment case.

The balance sheet is correspondingly unusual for a loss-making company. As of June 30, 2026, Anthropic held $112.4 billion in cash and cash equivalents plus a further $19.3 billion in short-term investments.

The risks that are not accounting artefacts

Dismissing the headline loss does not mean the filing is free of alarming disclosures. Several are substantive.

  • Future payment obligations to cloud providers run to roughly $450 billion over the next seven to ten years, with total computing-related spending commitments potentially reaching a ceiling of around $750 billion. Earlier reporting on a draft version of the prospectus cited a $518 billion commitment figure. These are contractual obligations that do require cash, and they are enormous relative to any revenue run rate disclosed so far.
  • Customer concentration is high. About one fifth of revenue comes from just two customers. Losing either would be material.
  • The prospectus itself warns that Anthropic's models could pose catastrophic or even existential risk to humanity if misused, replicated, weaponised or granted high autonomy, and that resulting harm could exceed the company's insurance coverage. That is an unusual thing to read in a risk factors section, and it is there because the company is a public benefit corporation founded in January 2021 with safety as part of its stated purpose.

The disagreement among professionals is sharp. David Trainer of New Constructs, commenting on an earlier draft prospectus, called the planned listing the most ridiculous IPO of the year, pointing to the valuation relative to losses and the $518 billion of contractual commitments. Chamath Palihapitiya of Social Capital, after the S-1 disclosed the swing to operating profitability, called it an amazing business. Both are looking at the same document.

The size of the deal

The offering is reportedly being run by 22 underwriting banks led by Goldman Sachs and Morgan Stanley, and could raise as much as $100 billion, which would exceed the $86 billion raised by SpaceX in its own record 2026 listing. Reported target valuations have ranged from $1.5 trillion to $1.75 trillion in investor discussions, with some reports citing figures up to $2 trillion. Pricing and listing have been reported as possible as early as the week of November 16, 2026.

None of those valuation or timing figures are confirmed in the filing itself. The S-1 as filed did not set a price range or share count, and reported dates come from unnamed sources rather than company statements. A reader should treat them as expectations, not facts.

How to read a loss line in an AI prospectus

The general lesson travels well beyond this one company. When a fast-growing private business lists, three distinct things can all be called a loss, and they mean entirely different things.

  • Cash burn. Money actually leaving the business. Found by looking at operating cash flow, not net income. Anthropic's was positive at roughly $3.93 billion for the half, the company's first positive operating cash flow, with roughly $1.4 billion of that attributed in reporting to customer prepayments.
  • Operating loss. Revenue minus the cost of running the business. This is the honest measure of whether the model works. Anthropic's was $693 million and narrowing.
  • Non-cash fair-value and share-based charges. Revaluations of instruments like convertible notes and SAFEs, plus stock compensation. Large, volatile, and frequently driven by the company's valuation rather than its performance.

A headline net loss simply adds all three together. For a company whose valuation has multiplied several times in eighteen months, the third category can swamp the first two by an order of magnitude and tell a reader almost nothing about the business.

Key Takeaways for Investors

  • Roughly $72.4 billion of Anthropic's $80.75 billion first-half net loss is a non-cash charge on convertible notes, derivatives and SAFEs, driven by the company's rising valuation rather than by spending.
  • The charge grows as the valuation grows, which is why it went from $1.0 billion in 2023 to $34.3 billion in 2025 to $72.4 billion in half of 2026.
  • The underlying operating loss was $693 million for the half, down from $2.14 billion, and the second quarter produced $974 million of operating income, the company's first quarterly operating profit.
  • Compute costs rose sharply to $8.95 billion but revenue rose faster. Whether that relationship holds is the central question for any AI model developer.
  • The genuine risks in the filing are the multi-hundred-billion-dollar cloud commitments stretching over seven to ten years, the concentration of roughly a fifth of revenue in two customers, and the company's own disclosed catastrophic-risk language.
  • Reported valuations between $1.5 trillion and $2 trillion are not in the filing. Pricing terms had not been set as of the filing date.

Conclusion

The $80.7 billion figure will be quoted for months, often by people arguing that the AI boom is built on nothing. It is worth being precise about why that argument fails in this instance, and also about where it does not fail. The loss is largely an artefact of how private companies finance themselves, and it will largely vanish on listing. The commitments to spend hundreds of billions on compute over the next decade will not vanish, and they sit against a revenue base that, however fast it is growing, is still measured in tens of billions.

The interesting test is not whether Anthropic can report a profit. It already has, for one quarter. The test is whether a business whose principal cost is renting computation can keep its revenue growing faster than the price of that computation, for long enough to cover contracts that run to the mid-2030s. The accounting answers nothing about that. It only determines how loud the headline is while the question stays open.