AI giant Anthropic loses $80 billion, but it's not as bad as it looks

One of the biggest losses ever is in the name of AI company Anthropic. This is evident from documents that leaked this week, a month before the planned IPO of the Claude maker. The loss mainly exists on paper, but there is a much larger bill underneath.

Anthropic, the creator of chatbot Claude, posted a net loss of $80.7 billion in the first half of 2026. Over the whole of 2025, that was still 42 billion dollars.

These figures can be found in a draft prospectus, the document with which a company introduces itself to investors. The Wall Street Journal was able to see the piece and wrote about it yesterday.

Yet most of the money has never been spent. Anthropic previously borrowed money with the agreement that the lenders would later receive shares. Because the company rose considerably in value, those shares are now worth much more.

According to the accounting rules, this increase in value must be entered as a cost in the books, although no money leaves the company. This is about 72 billion dollars. Without that item, there would be a loss of about $8.7 billion.

Meanwhile, sales grew rapidly: from 1.1 billion dollars to 16.3 billion dollars in a year. That's almost fifteen times as much.

AI models like Claude run on thousands of powerful computers in large data centers. Anthropic largely does not own those computers, but rents them from tech companies such as Google, Amazon and Microsoft.

That rent goes up a lot. In the first half of the year, Anthropic spent nearly $9 billion on computing power, more than in all of 2025. For the coming years, the company has already signed more than 600 billion dollars in rental agreements.

The tricky part: most of it can't be cancelled by Anthropic. Just like a rented house, you pay the rent even if you do not live there. If AI becomes less popular, the bill will simply stay in place.

In addition, Anthropic relies on a small group of customers. Two customers accounted for about one-fifth of sales, some $3.3 billion. According to the newspaper, who they are is not listed.

Not the new shareholders. Anthropic is controlled by a small group of independent regulators, a kind of council of sages without own shares in the company.

That group may appoint four of the seven directors and also send them away again. For example, Anthropic should continue to focus on secure AI, not just profit. One of the members is Ben Bernanke, former chairman of the US central bank.

Ann Lipton, professor of law at the University of Colorado, sees little room for new investors. Without major changes, “public shareholders will have very few governance rights,” she told The Wall Street Journal.

Whether investors are awake to this remains to be seen. According to the Reuters news agency, investors are now talking about a value of around 1.5 trillion dollars. Previously, 2 trillion dollars were mentioned.

At $1.5 trillion, investors pay more than 22 times what Anthropic now makes in a year. The company wants to go public in November and can raise up to $100 billion, according to the newspaper, more than SpaceX's $86 billion in June.