With US$114.5 billion of Google compute alone committed regardless of use, Anthropic's margins depend on holding prices against cheaper open-weight rivals, including Chinese ones.
Anthropic, the US developer of the Claude AI models, has told prospective shareholders that competition from China, the concentration of advanced chipmaking in Taiwan and US export controls on its own models all pose risks to its business. The warnings sit in the Form S-1 that Anthropic filed with the US Securities and Exchange Commission on October 26, ahead of a planned Nasdaq listing under the ticker ANT. The risk factors run from page 26 to page 119, according to The Next Web.
The filing also gives the first full public view of the company’s finances. Anthropic reported first-half revenue of US$16.3 billion, up from US$1.16 billion a year earlier, and a net loss of US$80.7 billion. The Next Web reported that most of that loss came from a US$72.4 billion non-cash revaluation of convertible notes, and that the second quarter brought Anthropic’s first operating profit, of US$974 million.
The prospectus is a rare document in which a leading US lab sets out, under securities-law liability, how it sees Chinese competition affecting its revenue and costs. The company has not given a valuation in the filing.
Anthropic links China’s lighter rules to faster rivals
Yahoo Finance reported that the filing warns growing US-China strategic competition could lead to different regulatory systems in each country, limiting American developers’ access to Chinese markets while letting rivals under other rules gain ground. Anthropic wrote that "increasingly restrictive AI regulation" in the United States and other Western countries could let competitors in less regulated places develop and deploy AI more quickly, Yahoo Finance reported.
The cost pressure is already visible. Yahoo Finance noted that DeepSeek and Moonshot AI have released open-weight models, meaning models whose trained parameters anyone can download and run, that compete with proprietary US systems on a growing range of tasks at much lower prices. Forbes contributor Drew Bernstein wrote that the S-1 lists open-weight models released by state actors among its competitive risks, and quotes the filing’s warning that customers could move to smaller, cheaper or open-source alternatives.
Anthropic has accused DeepSeek, Moonshot AI, Alibaba and other Chinese companies of distillation, using a model’s answers to train a rival model, and called their conduct an "industrial-scale, covert campaign" to copy its models’ capabilities, Yahoo Finance reported. China has rejected US distillation allegations. The Next Web separately reported that customers who broke Anthropic’s usage policy or terms accounted for 5% to 7% of its revenue.
Taiwan and Washington are named alongside Beijing
The supply risk is concentrated in one place. In a separate report, The Next Web said the filing describes efforts to build enough advanced chip manufacturing capacity outside Taiwan, and, citing Politico, that it counts threats from China among the factors that could disrupt that supply. The exposure is large because Anthropic has committed to buy computing capacity worth at least US$114.5 billion from Google, US$110 billion from Amazon and US$44.9 billion from Microsoft, which it must pay for regardless of actual use, according to Forbes.
The filing also records how US policy has already touched Anthropic directly. In June, the Department of Commerce imposed a worldwide export restriction on Anthropic’s most advanced models, Fable 5 and Mythos 5, which was lifted after the company temporarily suspended customer access and updated its safeguards, Yahoo Finance reported. Anthropic warned that other models may face these or similar restrictions in the future.
Security is the third China thread. The S-1 discloses that a Chinese state-sponsored group used Anthropic’s Claude Code tool in September 2025 to try to break into about 30 targets, Bernstein wrote. The filing calls the episode "the first documented case of a large-scale cyberattack executed without substantial human intervention."
The roadshow will test how investors price these risks
Forbes reported a target valuation of about US$1.5 trillion, an offering of US$60 billion to US$100 billion and pricing as soon as November 19. Yahoo Finance’s live coverage reported that a range of US$1.5 trillion to US$2 trillion was under discussion. Goldman Sachs and Morgan Stanley lead the underwriting group.
The Next Web reported that the top five customers accounted for about 25% of revenue in the nine months to September 30.
The amended S-1 that typically follows the first round of SEC comments, and the price range Anthropic sets for the roadshow, will show whether the China risks change the terms investors demand.