Anthropic Lost $80 Billion, but Mostly on Paper
Set Trending Topics as a preferred source on Google.
Going public means opening the books: According to its I.P.O. prospectus, Anthropic posted a net loss of $80.7 billion (about €72 billion) in the first half of the year. That is shown in a draft of the prospectus viewed by The Wall Street Journal, Reuters reports. Most of it, however, is a pure accounting loss.
The Numbers at a Glance
In the first half of the year, the maker of Claude generated $16.3 billion in revenue (about €14.5 billion), up from $1.1 billion a year earlier. For all of 2025, revenue was $4.6 billion. By comparison, the net loss of $80.7 billion looks enormous; in the first half of 2025, it stood at about $13 billion.
About $72 billion of the loss, though, involves no cash leaving the company. It stems from the revaluation of convertible notes and other financing instruments: When Anthropic’s value rises, so does the value of these instruments, and the company has to book that as a loss. Real money went mainly into computing power and infrastructure, nearly $9 billion in the first half, more than the $7.3 billion spent in all of 2025.
$450 Billion for Computing Power
The contracts in the prospectus show how hungry Anthropic is for computing power. The company has signed noncancelable cloud agreements worth about $450 billion over the next decade:
- Google Cloud: $111 billion
- Amazon Web Services: $110 billion
- Microsoft: $31 billion
Including ongoing negotiations, for instance with the British data center operator Nscale, the total could reach as much as $750 billion. According to the prospectus, Anthropic sought more capacity than the big cloud providers were able to supply. That is why it also leases the Colossus 1 and Colossus 2 data centers in Memphis from Elon Musk’s SpaceX, for as much as $84.5 billion, according to Reuters. Thanks in part to such leases, SpaceX has just reported a quarterly profit of $4.4 billion.
Profitable on an Adjusted Basis
When courting investors, Anthropic paints a very different picture. For the third quarter, the company reported $16.8 billion in revenue and an adjusted operating margin of more than 11 percent. That would make Anthropic profitable on an adjusted basis for the second consecutive quarter. However, this metric excludes training costs for new models and stock-based compensation, two of the biggest cost items for any A.I. lab.
The prospectus also lists the risks openly. Two customers accounted for about 20 percent of revenue in the first half, and Anthropic expects a small number of large customers to keep contributing a significant share. Then there are the huge fixed costs for computing power, which remain even if growth slows. Anthropic also warns that damages from severe A.I. safety incidents could exceed its insurance coverage.
A Race With OpenAI
Anthropic is not the only A.I. lab heading for the stock market. OpenAI has also confidentially filed for an I.P.O. and was most recently hoping for a valuation of about $1 trillion. SpaceX, which now also includes Mr. Musk’s A.I. business xAI, has been listed since June.
Anthropic is aiming for a valuation of up to $2 trillion (about €1.8 trillion) in its planned I.P.O. in the coming weeks. According to the prospectus, the chief executive, Dario Amodei, and the president, Daniela Amodei, each received a base salary of about $655,000 in 2025 and have pledged to give away 80 percent of their stakes. In addition, Anthropic plans to put 5 percent of its shares into a public benefit fund.
