How Has Anthropic Made a Net Loss of $80.7 Billion?

2026-10-27

Key Takeaways

Anthropic recorded a net loss of $80.7 billion in the first half of 2026 (the six months to June 30), compared with about $13 billion a year earlier, according to its draft IPO filing as reported by the Wall Street Journal and Reuters.
Around $72 billion of the loss is an accounting charge: the estimated fair value of the company’s convertible notes and similar financing instruments rose as its valuation climbed. No cash left the company for this item.
Without that charge, the loss is roughly $8.7 billion, and most of it went to computing power. Anthropic spent about $9 billion on compute and infrastructure in six months, up from $2.3 billion a year earlier.
Revenue jumped to $16.3 billion in the first half from $1.1 billion, and third-quarter revenue reached $16.8 billion.
Anthropic has signed compute commitments worth at least $518 billion over about a decade, and some cloud contracts must be paid whether or not the capacity gets used.
The company is preparing to list as early as November, with a valuation range that reportedly starts near $1.5 trillion.

AI safety and robustness – conceptual image. Image source: Anthropic

Anthropic lost $80.7 billion in the first half of 2026 mostly because of an accounting rule, not because it spent that much money. About $72 billion of the figure comes from re-measuring convertible notes and other financing instruments at fair value. As investors priced the company higher, the shares those instruments may turn into became more valuable, and Anthropic had to book the increase as a loss. The remaining gap of roughly $8.7 billion is the real operating shortfall, and computing power accounts for most of it.

The numbers sit in Anthropic’s draft S-1 registration statement, filed confidentially ahead of a planned stock market debut. The Wall Street Journal first reported the details on October 20, and Reuters followed the next day. The same filing shows revenue growing almost 15-fold year on year, alongside compute obligations so large that they now dominate the company’s risk disclosures.

Where the $80.7 Billion Loss Comes From

The headline loss has two very different parts. One is a paper charge that moves with Anthropic’s valuation. The other is the cost of running and training Claude models faster than revenue can pay for them. Investors reading the filing will care far more about the second part.

Item	First half of 2025	First half of 2026
Revenue	$1.1 billion	$16.3 billion
Net loss	About $13 billion	$80.7 billion
Fair-value charge on financing instruments	Not broken out in reports	About $72 billion
Loss excluding that charge (calculated)	Not available	About $8.7 billion
Compute and infrastructure spending	$2.3 billion	About $9 billion

The $72 billion non-cash charge, explained simply

A convertible note starts life as a loan that can later turn into shares, usually at a future funding round or at an IPO. Under fair-value accounting, a company must re-price these instruments every time it reports. When the business becomes more valuable, the shares the notes would convert into are worth more, so the liability on the balance sheet grows. That growth flows through the income statement as a loss.

The result looks odd at first: the better Anthropic’s prospects appear to investors, the bigger this paper loss becomes. Nothing is paid out in cash. The charge mostly disappears from future results once the notes convert into equity, which typically happens when a company goes public.

This is not new for Anthropic. Its 2025 accounts show a net loss of about $42 billion, of which a similar fair-value charge made up roughly $34 billion. Capital Brief, which reviewed earlier figures, reported that the fair value of the notes roughly tripled during 2025. The 2025 operating loss was about $8.1 billion on revenue of around $4.6 billion, and the company ended that year with $20.28 billion in cash. For context, revenue in 2024 was only $386 million, with an operating loss of $2.98 billion.

The real operating loss underneath

Subtract the $72 billion charge from $80.7 billion and roughly $8.7 billion remains. That figure is calculated from the reported numbers rather than stated in the filing, but it lines up closely with what Anthropic spent on compute. The company put about $9 billion into computing power and infrastructure in the first half of 2026, nearly four times the $2.3 billion of a year earlier and more than its compute spending for the whole of 2025, which was about $7.3 billion.

Training frontier models and serving millions of paying users both consume enormous amounts of chips and power. Anthropic is paying for both at once, and it is buying capacity ahead of demand so that it does not run short during growth spurts.

Is Anthropic Really Losing That Much Money?

No. Anthropic did not lose $80.7 billion in cash. The bulk of the figure is a non-cash accounting adjustment linked to its rising valuation. The cash-relevant loss is closer to $8.7 billion for the half-year, which is still large, but it sits against revenue that is growing at a speed few software companies have ever reached.

Analysts at FourWeekMBA made the same point in their breakdown of the filing: once the financing charge is removed, the story becomes one of compute spending racing revenue, not of a business collapsing.

How Fast Is Anthropic’s Revenue Growing?

Revenue for the first six months of 2026 reached $16.3 billion, up from $1.1 billion a year before. Growth accelerated within the half: second-quarter revenue of about $11.6 billion was more than double the roughly $4.7 billion booked in the first quarter.

Period	Revenue	Change
Q1 2026	About $4.7 billion	—
Q2 2026	About $11.6 billion	More than doubled from Q1
Q3 2026	$16.8 billion	Up about 45% from Q2

Anthropic told investors on October 14 that third-quarter revenue came in at $16.8 billion. That is still a steep climb, yet some investors were reportedly disappointed that quarterly growth slowed from the pace seen in spring. Bloomberg reported that the company’s annualized revenue run rate reached about $65 billion by the end of July. Claude Code, the company’s coding agent, is one of the main engines behind that growth, as developers and enterprises pay for heavy agentic workloads.

The filing also exposes customer concentration. Two customers produced about one-fifth of first-half revenue. In 2025, two customers accounted for nearly a quarter of revenue, at roughly 12% each. At the end of 2025, Amazon and Google together represented 60% of the money customers still owed Anthropic. Both companies are also major investors and cloud suppliers, so the relationships run in several directions at once.

Why Anthropic Spends So Much on Compute

Running out of capacity is the risk Anthropic seems most determined to avoid. Reuters reported that the company has committed to at least $518 billion of compute spending over roughly a decade. Other readings of the draft filing put total obligations even higher once chip leases are counted.

Partner	Reported commitment
Google Cloud	$111 billion
Amazon Web Services	$110 billion
Microsoft	$31 billion
Nscale and Lambda	Up to $86 billion (running to 2033–2034)
Broadcom and Google chip lease	More than $160 billion

The cloud agreements alone come to about $450 billion over seven to ten years. Adding the chip lease takes the total above $600 billion, according to investingLive’s review of the filing, and deals still under negotiation could add up to $150 billion more. Several of the contracts with Google, Amazon and Microsoft must be paid regardless of actual usage. The filing also says partners could not supply all the compute Anthropic wanted, which pushed it to lease its own data centers and to work with newer operators.

Chief executive Dario Amodei has spoken openly about the danger in this approach. In a February interview with Dwarkesh Patel, he said a lab that buys compute far ahead of its revenue could go bankrupt if growth falls short of plan. The S-1 shows Anthropic making exactly that bet, at a scale that makes it one of the largest buyers in the AI infrastructure market.

What the Risk Factors Warn About

Risk disclosures take up roughly a third of the draft filing. Beyond the compute obligations and customer concentration, Anthropic warns that damages from a catastrophic incident involving its models could exceed its insurance coverage. It also flags organized local opposition to data center construction, which could delay the capacity it has already agreed to pay for.

Taken together, the disclosures describe a company whose revenue curve looks exceptional while its cost commitments assume that curve keeps rising for years. A sharp slowdown would leave fixed payments in place with less income to cover them.

What the Loss Means for the Anthropic IPO

Anthropic is expected to go public as early as November, with Morgan Stanley, Goldman Sachs and JPMorgan working on the listing. Reports put the starting valuation range around $1.5 trillion, below earlier hopes of a $2 trillion valuation. The gap between those numbers is where the S-1 matters most: buyers will weigh the cash loss and the compute bills, not the headline accounting figure.

The listing will also invite direct comparison with rivals. Anthropic’s long valuation race with OpenAI continues, and OpenAI’s own IPO plans will give investors a second reference point once its numbers become public. For a broader view of how Anthropic’s listing compares with other AI players, revenue growth, gross margins and compute obligations will be the metrics to watch.

The fair-value charge should shrink in importance after the IPO, since conversion of the notes ends the need to keep re-pricing them. What remains is a simpler question: can revenue keep growing fast enough to cover hundreds of billions of dollars in compute contracts? The first-half numbers show strong momentum. The third-quarter slowdown shows why investors want proof that it continues.

Disclaimer: This article is for information only and does not constitute investment advice. The figures come from media reports on a draft registration statement and may change in the final prospectus.

If you are interested in this topic, we suggest you check our articles:

Sources: Reuters via ARY News, Al Jazeera, VKTR, investingLive, FourWeekMBA, The Standard (Bloomberg)

Written by Alius Noreika