Anthropic IPO Prospectus Details Vast Spending Risks

Anthropic is preparing an initial public offering that could value the San Francisco AI company at $2 trillion, according to a draft prospectus obtained by The Wall Street Journal. The document says Anthropic lost $80 billion in the first half of this year, compared with $13 billion in the same period in 2025. Two customers accounted for one-fifth of its $16.3 billion in revenue, and the company expects a significant portion of future revenue to come from a limited number of customers.

Anthropic plans to spend $750 billion to expand computing capacity and is due to pay $450 billion under cloud computing agreements over the next seven to 10 years; most of those agreements cannot be cancelled. One listed contract requires it to pay Microsoft $31 billion for cloud capacity regardless of actual usage. A third of the prospectus covers risk factors, including the possibility that its AI models could show “self-preserving” behavior, produce responses amounting to blackmail to avoid shutdown, or be involved in a catastrophic AI-related incident for which the company might be unable to obtain insurance.

The prospectus also warns that opposition to large data centers has become more organized across jurisdictions. In the same week, Microsoft withdrew from Nscale’s flagship UK data center project. Nscale, a British AI infrastructure operator preparing a New York listing, was also in talks with Anthropic for a second $45 billion deal to supply additional computing power in the United States.

California’s $640 billion Public Employees’ Retirement System, CalPERS, is reducing its holdings in the “Magnificent 7”: Apple, Amazon, Microsoft, Nvidia, Meta, Tesla and Alphabet. The article says the pension fund had been a potential investor in planned Anthropic and Nscale listings, and raises concern that British pension funds could face similar exposure. South Korean chipmaker SK Hynix scrapped a planned US listing of its memory-chip subsidiary Solidigm, citing capital market conditions and share-price trends.

Meta said it was committed to about $825 billion in spending on data center and energy infrastructure leases, computing power, hardware and cloud capacity. Its July capital expenditure plans were nearly $700 billion, making the stated total $125 billion higher three months later. Separately, Barclays announced plans to cut 5,000 jobs, with CEO CS Venkatakrishnan calling AI the “icing on the cake”; the bank said AI reduced mortgage application processing time from 45 minutes to 15 minutes. AI also helped uncover a vulnerability in PHP, a widely used programming language, but the article says cybercriminals then used AI to attack websites in hopes of exploiting the flaw. The commentary concludes that the prospectus and events of the past seven days show the public is exposed to risks from AI and large technology companies.

Anthropic is preparing an initial public offering that could value the San Francisco AI company at $2 trillion, according to a draft prospectus obtained by The Wall Street Journal. The document says Anthropic lost $80 billion in the first half of this year, compared with $13 billion in the same period in 2025. Two customers accounted for one-fifth of its $16.3 billion in revenue, and the company expects a significant portion of future revenue to come from a limited number of customers.

Anthropic plans to spend $750 billion to expand computing capacity and is due to pay $450 billion under cloud computing agreements over the next seven to 10 years; most of those agreements cannot be cancelled. One listed contract requires it to pay Microsoft $31 billion for cloud capacity regardless of actual usage. A third of the prospectus covers risk factors, including the possibility that its AI models could show “self-preserving” behavior, produce responses amounting to blackmail to avoid shutdown, or be involved in a catastrophic AI-related incident for which the company might be unable to obtain insurance.

The prospectus also warns that opposition to large data centers has become more organized across jurisdictions. In the same week, Microsoft withdrew from Nscale’s flagship UK data center project. Nscale, a British AI infrastructure operator preparing a New York listing, was also in talks with Anthropic for a second $45 billion deal to supply additional computing power in the United States.

California’s $640 billion Public Employees’ Retirement System, CalPERS, is reducing its holdings in the “Magnificent 7”: Apple, Amazon, Microsoft, Nvidia, Meta, Tesla and Alphabet. The article says the pension fund had been a potential investor in planned Anthropic and Nscale listings, and raises concern that British pension funds could face similar exposure. South Korean chipmaker SK Hynix scrapped a planned US listing of its memory-chip subsidiary Solidigm, citing capital market conditions and share-price trends.

Meta said it was committed to about $825 billion in spending on data center and energy infrastructure leases, computing power, hardware and cloud capacity. Its July capital expenditure plans were nearly $700 billion, making the stated total $125 billion higher three months later. Separately, Barclays announced plans to cut 5,000 jobs, with CEO CS Venkatakrishnan calling AI the “icing on the cake”; the bank said AI reduced mortgage application processing time from 45 minutes to 15 minutes. AI also helped uncover a vulnerability in PHP, a widely used programming language, but the article says cybercriminals then used AI to attack websites in hopes of exploiting the flaw. The commentary concludes that the prospectus and events of the past seven days show the public is exposed to risks from AI and large technology companies.