Anthropic IPO Filing Raises Mission Questions

Anthropic’s preliminary prospectus for its planned IPO, made public on Monday, says the company aims to develop advanced AI “for the benefit of humanity,” while warning it cannot assure investors that this purpose or its expected positive impact will be achieved. The phrase “Company Purpose” appears 42 times in the 318-page filing. The prospectus also warns that sufficiently advanced AI could create catastrophic or existential risks if it is not developed and deployed responsibly, and says Anthropic cannot assure its efforts to reduce those risks will succeed.

Investors are being asked to weigh that mission against large expenses and ambitious growth expectations. Anthropic’s annualized costs already run into tens of billions of dollars, largely for chips and computing resources. Market expectations put the potential IPO fundraising at about $100 billion and the company’s valuation as high as $2 trillion; the offering is expected to launch sometime next month. For the first six months of 2026, revenue reached $16.3 billion, up from $1.2 billion in the same period a year earlier. Operating expenses, including computing and infrastructure, were nearly $9 billion, compared with $2.3 billion a year earlier. The operating loss narrowed to $693 million from $2.1 billion. Under generally accepted accounting standards, however, the first-half loss was $80.7 billion, including a $72.4-billion provision for convertible notes issued to cover startup expenses.

The prospectus lists roughly $500 billion in non-cancellable commitments for computing chips and data center capacity. The amounts include $110 billion payable to Amazon, $114.5 billion to Google, $44.9 billion to Microsoft and $120.5 billion to SpaceX.AI, a SpaceX subsidiary; some deals run for 10 years. The company must pay even if it does not use the contracted resources. Hiltzik notes that the commitments could weigh on Anthropic if customer growth slows. The filing says five customers accounted for about 25% of revenue in the first nine months of 2026, though no single customer represented more than 10%. Many customers have no long-term contracts and can cut or stop spending at any time.

Public shareholders would have limited voting power under Anthropic’s five-class share structure. Class A shares, planned for Nasdaq trading, carry one vote each but will never hold more than 49.9% of votes on company policy or director elections. A single Class F share, jointly controlled by the seven co-founders, including CEO Dario Amodei and President and board chair Daniela Amodei, will retain 50.1%. That arrangement is set to continue until about three years after fewer than three co-founders remain at the company. A single Class T share belongs to the independent Long-Term Benefit Trust, overseen by AI safety and public policy experts, which can elect a majority of the board. The columnist questions what happens if the trust’s trustees believe the founders have weakened their commitment to safety.

Hiltzik compares Anthropic’s stated purpose with Google founders’ “Don’t be evil” motto around its IPO. He cites Federal Trade Commission charges in 2011 and 2012 that Google deceptively breached consumers’ privacy and users of Apple’s Safari browser, respectively; Google settled both cases without admitting wrongdoing and paid a then-record $22.5 million civil penalty in the second case. Google has also faced criticism over extremist content on YouTube and said it tried to adjust recommendation algorithms to suppress it. Hiltzik says these precedents do not establish that Anthropic will follow the same path, but argues that recording worthy intentions in a prospectus does not guarantee they will prevail over commercial pressures. Anthropic declined his request for comment, citing federal IPO quiet-period rules.

Anthropic’s preliminary prospectus for its planned IPO, made public on Monday, says the company aims to develop advanced AI “for the benefit of humanity,” while warning it cannot assure investors that this purpose or its expected positive impact will be achieved. The phrase “Company Purpose” appears 42 times in the 318-page filing. The prospectus also warns that sufficiently advanced AI could create catastrophic or existential risks if it is not developed and deployed responsibly, and says Anthropic cannot assure its efforts to reduce those risks will succeed.

Investors are being asked to weigh that mission against large expenses and ambitious growth expectations. Anthropic’s annualized costs already run into tens of billions of dollars, largely for chips and computing resources. Market expectations put the potential IPO fundraising at about $100 billion and the company’s valuation as high as $2 trillion; the offering is expected to launch sometime next month. For the first six months of 2026, revenue reached $16.3 billion, up from $1.2 billion in the same period a year earlier. Operating expenses, including computing and infrastructure, were nearly $9 billion, compared with $2.3 billion a year earlier. The operating loss narrowed to $693 million from $2.1 billion. Under generally accepted accounting standards, however, the first-half loss was $80.7 billion, including a $72.4-billion provision for convertible notes issued to cover startup expenses.

The prospectus lists roughly $500 billion in non-cancellable commitments for computing chips and data center capacity. The amounts include $110 billion payable to Amazon, $114.5 billion to Google, $44.9 billion to Microsoft and $120.5 billion to SpaceX.AI, a SpaceX subsidiary; some deals run for 10 years. The company must pay even if it does not use the contracted resources. Hiltzik notes that the commitments could weigh on Anthropic if customer growth slows. The filing says five customers accounted for about 25% of revenue in the first nine months of 2026, though no single customer represented more than 10%. Many customers have no long-term contracts and can cut or stop spending at any time.

Public shareholders would have limited voting power under Anthropic’s five-class share structure. Class A shares, planned for Nasdaq trading, carry one vote each but will never hold more than 49.9% of votes on company policy or director elections. A single Class F share, jointly controlled by the seven co-founders, including CEO Dario Amodei and President and board chair Daniela Amodei, will retain 50.1%. That arrangement is set to continue until about three years after fewer than three co-founders remain at the company. A single Class T share belongs to the independent Long-Term Benefit Trust, overseen by AI safety and public policy experts, which can elect a majority of the board. The columnist questions what happens if the trust’s trustees believe the founders have weakened their commitment to safety.

Hiltzik compares Anthropic’s stated purpose with Google founders’ “Don’t be evil” motto around its IPO. He cites Federal Trade Commission charges in 2011 and 2012 that Google deceptively breached consumers’ privacy and users of Apple’s Safari browser, respectively; Google settled both cases without admitting wrongdoing and paid a then-record $22.5 million civil penalty in the second case. Google has also faced criticism over extremist content on YouTube and said it tried to adjust recommendation algorithms to suppress it. Hiltzik says these precedents do not establish that Anthropic will follow the same path, but argues that recording worthy intentions in a prospectus does not guarantee they will prevail over commercial pressures. Anthropic declined his request for comment, citing federal IPO quiet-period rules.