Anthropic’s valuation can’t secure its moat

Artificial intelligence (AI) company Anthropic, best known as the maker of the Claude AI model, is preparing to go public at a valuation of $1.4 trillion, which could make the IPO one of the largest in history. However, this particular refrain has of late seemed more ominous than cause for optimism, and the Anthropic IPO could be no exception.

The company has said it will offer 825 million shares at $70-78 apiece, to raise up to $64.4 billion. The valuation includes shares to be issued through employee options and other arrangements, and its market capitalisation based on outstanding shares is likely to be near $1 trillion. (It was valued at $965 billion in a private funding round in May.)

Anthropic’s financial prospects depend on continuing to spend extraordinary sums of money. Its amended prospectus filed with the U.S. Securities and Exchange Commission on November 10 revealed roughly $900 billion in spending commitments, almost entirely for computing infrastructure, until 2045. Of this, the company expects to incur half by 2030 and $70.7 billion in the twelve months from September 30.

The filing also mentions another $150-175 billion in potential payments under arrangements sans fixed dollar commitments.

These figures clarify a colossal exposure to future demand that neither Anthropic nor other Big AI companies have thus far been able to guarantee.

And while Anthropic’s revenue has also been growing, from $2.5 billion in the first three quarters of 2025 to $33.2 billion in the corresponding slice of 2026, the proposed fully diluted valuation exceeds annualised sales 17-fold.

The current business model is, largely, to charge customers according to how many tokens they use to run their tasks on its AI models. From January to September 2026, this revenue constituted 79% of the total revenue, giving Anthropic a reason to encourage users to maximise token use, irrespective of changes in the value of the output generated — a tendency famously called tokenmaxxing. But three factors complicate this picture. First, Anthropic has said AI models released within the last six months accounted for more than 80% of its third quarter revenue, which means its engineers need to keep building progressively more capable models. Second, it has not established that it will be able to sustain its profitability after honouring its infrastructure commitments.

The third factor is the most important, however: neither Anthropic nor its competitors, including OpenAI, currently has a lasting competitive advantage, a.k.a. a moat.

In fact, the difficulties of securing a moat are better illustrated by OpenAI CEO Sam Altman’s position of late, in favour of new regulatory and compliance demands. These demands could impose costs that larger AI companies will bear more easily than smaller ones.

Legitimate reasons

Legitimate reasons to regulate AI include security failures and the risks posed by autonomous systems but governments should endeavour overtly to prevent such regulatory capture given the perverse incentives facing AI companies.

Overall, neither Anthropic nor its competitors have presented evidence of a sturdy moat. Instead, thanks to capable open-source development, models are becoming cheaper and competitors are catching up, and efficiency gains threaten the pricing power on which their valuations depend.