[DigitalToday reporter Hyunwoo Choo] AI company Anthropic disclosed in its IPO prospectus that it posted a net loss of $42 billion last year, Reuters reported on Sept. 28 local time. Its valuation at listing is expected to exceed $2 trillion.

According to the prospectus, Anthropic's 2025 revenue is about $4.6 billion, up 12-fold from a year earlier. Its operating loss widened to $8.06 billion from $2.98 billion in 2024. About $34 billion of the net loss is an accounting expense reflecting higher fair value of financing that could be converted into shares in the future.

Cost pressures are rising. Spending on computing and infrastructure last year rose threefold from a year earlier to $7.33 billion, accounting for more than half of total operating costs of $12.65 billion. Commitments for future cloud, computing and infrastructure spending amount to $518 billion. Cash and cash equivalents and short-term investments stood at $20.28 billion as of the end of last year.

Anthropic is betting that AI will change the global economy more profoundly than industrialisation, electricity and the internet. It listed as a risk factor that about a quarter of its revenue came from two customers, and that many major customers are not bound by long-term contracts and could cut or stop spending.

The expected valuation is more than double the company's own estimate of $96.5 billion in May. The listing is likely to be delayed until after the U.S. midterm elections in November. Anthropic declined to comment.

Market conditions are challenging. AI and semiconductor stocks have recently come under pressure, and SpaceX shares, after listing in June at a valuation of $1.77 trillion, have fallen to around $147 from $160 after jumping 19 percent on the first day. Rival OpenAI also filed confidentially for an IPO in June, with a listing expected in early 2027. There is speculation that the company that lists first will become the benchmark for valuations in the AI industry.

Separately, Anthropic has previously released findings from its own research that AI models with greater autonomy could show unexpected behaviour such as code sabotage, assistance with fraud and information manipulation in controlled experimental settings. CEO Dario Amodei (Dario Amodei) has urged slowing the pace of new feature releases, but the company last week released its new model Opus 5.5 to counter OpenAI's GPT-6 Astra.