Anthropic Lost $80 Billion In Six Months As AI Spending Soars

Anthropic reported a net loss of $80.7 billion in the first half of 2026. Its revenue grew sharply, while spending on computing infrastructure, the equipment and systems needed to run AI, also increased.

The San Francisco-based AI company reported a net loss of $13 billion during the same period a year earlier, according to a draft document for its planned initial public offering (IPO) reviewed by The Wall Street Journal. Most of its latest loss, $72 billion, came from accounting charges linked to changes in the value of convertible notes and other financial tools the company issued to raise money.

These disclosures come as Anthropic prepares for a potential major IPO that could value the company at as much as $2 trillion. Investors are closely examining the costs and financial risks of building increasingly powerful AI models.

According to the draft IPO document, Anthropic made about $16.3 billion in revenue during the six months ending June 30, up from $1.1 billion in the same period a year earlier. At the same time, spending on computing power and infrastructure rose to nearly $9 billion, compared with $2.3 billion in the first half of 2025.

In just six months, the company spent more than the $7.3 billion it had spent on computing and infrastructure during all of 2025. That year, it made nearly $4.6 billion in revenue. For the full year, Anthropic reported a net loss of $42 billion. More than $34 billion of that loss came from changes in the estimated value of its debts and other obligations, mostly linked to earlier fundraising.

Anthropic’s Computing Commitments Top $600 Billion

Anthropic’s fast growth is leading it to make huge long-term commitments to computing infrastructure. It needs enough computing power to train and run its AI models.

Anthropic faces hundreds of billions of dollars in long-term infrastructure costs. These include cloud computing contracts that, in many cases, cannot be ended early. The draft IPO document lists planned payments of $111.1 billion to Alphabet’s Google, $110 billion to Amazon Web Services and $31.4 billion to Microsoft. Anthropic may have to pay these amounts even if it does not use all the computing power covered by the agreements.

Separately, Anthropic could pay as much as $86 billion to rent computing resources powered by Nvidia chips from providers Nscale and Lambda. These agreements run through 2033 and 2034, respectively.

According to The Wall Street Journal, Anthropic’s long-term infrastructure commitments could exceed $600 billion when an agreement to lease Google chips through a deal involving Broadcom is included. The company is also discussing more agreements that could add another $150 billion to its total commitments.

The company warned that it is not certain it will be able to get all the computing power it needs. It said it had asked some major cloud partners for more capacity than they were willing to provide, as those partners increasingly see Anthropic as a competitor.

AI Safety and Data Center Backlash Pose Risks

Anthropic also described possible legal and operational risks linked to its AI models. These include the risk that AI systems that do not behave as intended could cause serious harm.

The draft IPO document warned that AI models might try to protect themselves from being shut down. They might also produce responses that resemble blackmail in an effort to persuade people to let them keep running. Anthropic said the damage from a catastrophic AI-related incident could be so large that insurance might not be available or might not cover all the losses.

These disclosures come as independent AI evaluator METR questions whether Anthropic’s safety tests can reliably detect dangerous behavior. In a review, METR said Anthropic’s risk assessment did not provide enough evidence to confidently rule out serious cases of AI systems behaving in ways that conflict with human goals. METR also called for more realistic tests of the systems used to monitor AI behavior.

Anthropic also listed "political resistance" to building data centers as a business risk in its IPO document. The company warned that opposition to new facilities had "become more organized" and that its public profile could "draw particular attention to projects associated with us."

The document also described pay for the company’s top leaders. CEO Dario Amodei received about $18 million in total compensation in 2025, while President Daniela Amodei received about $16.5 million. Most of their compensation came from stock and stock option awards.

Anthropic’s co-founders have promised to give 80% of their personal ownership in the company to charitable causes. According to the report, the company also plans to put 5% of its shares into a new public benefit fund.

Photo: RixAiArt / Shutterstock – ek