Anthropic's confidential IPO filing reveals USD 80.7 billion H1 loss, massive computing costs

Leading super intelligence company Anthropic reported a net loss of USD 80.7 billion in the first half of 2026, while outlining significant computing costs, customer concentration and potential AI-related liabilities ahead of a planned initial public offering (IPO), according to a draft prospectus reviewed by The Wall Street Journal.

Leading super intelligence company Anthropic reported a net loss of USD 80.7 billion in the first half of 2026, while outlining significant computing costs, customer concentration and potential AI-related liabilities ahead of a planned initial public offering (IPO), according to a draft prospectus reviewed by The Wall Street Journal. The company generated USD 16.3 billion in revenue during the six months ended June 30, compared with USD 1.1 billion in the corresponding period a year earlier, the newspaper reported. The loss largely reflected accounting adjustments related to financing agreements, including USD 72 billion in charges linked to changes in the value of convertible notes and other financial instruments.

According to The Wall Street Journal, Anthropic's computing and infrastructure expenditure reached nearly USD 9 billion in the first half of 2026, exceeding the USD 7.3 billion spent during the whole of 2025. The company generated USD 4.6 billion in revenue in 2025. Anthropic's draft prospectus also indicated that its future infrastructure commitments could exceed USD 600 billion, with the company considering additional computing deals worth up to USD 150 billion, The Wall Street Journal reported.

The newspaper reported that Anthropic has committed to spending around USD 450 billion on cloud computing agreements over the next seven to 10 years, with most of these arrangements being non-cancellable. These include commitments of USD 111 billion to Google Cloud, USD 110 billion to Amazon Web Services and USD 31 billion to Microsoft, regardless of actual usage. The company has also agreed to spend up to USD 86 billion renting Nvidia chips through AI cloud providers Nscale and Lambda, alongside a commitment exceeding USD 160 billion for Google chips through an arrangement with Broadcom, according to the report.

Anthropic warned that access to computing capacity remained uncertain, saying its requirements had exceeded what some cloud partners were willing to provide. The company also identified the growing resistance to data centre construction as a potential business risk. The Wall Street Journal reported that around one-fifth of Anthropic's USD 16.3 billion in first-half revenue came from just two customers, highlighting its dependence on a limited customer base.

The draft prospectus also outlined risks associated with potentially harmful AI behaviour, including models generating responses resembling blackmail or displaying self-preserving tendencies to avoid shutdown. Anthropic warned that the financial consequences of a catastrophic AI-related incident could exceed available insurance coverage. The disclosures come as Anthropic prepares for a potential blockbuster IPO that could value the company at USD 2 trillion, according to The Wall Street Journal. (ANI)