Anthropic提交IPO文件：巨额亏损、快速增长，并担忧AI可能终结人类

一份泄露的IPO招股书让外界首次详细看到Anthropic的财务状况。这家Claude背后的AI公司去年营收46亿美元，但亏损高达420亿美元，烧钱速度惊人。招股书显示，Anthropic在2025年营收同比增长1088%，但运营亏损从上一年的29.8亿美元扩大到80.6亿美元。为了跟上AI军备竞赛，公司去年在算力和基础设施上花了73.3亿美元，是2024年的三倍。Claude Code等产品带来用户激增后，Anthropic一度遭遇算力紧张，随后加速锁定计算资源。目前公司已承诺未来在云服务、算力和基础设施上的支出高达5180亿美元，亚马逊和谷歌既是其重要投资方，也是主要云基础设施供应商。截至2025年12月，Anthropic持有202.8亿美元现金及短期投资，但招股书也披露，两家未具名客户贡献了2025年近四分之一的营收，且许多大客户并未签订长期合同，存在减少或停止支出的风险。 Anthropic reportedly aims for a valuation of more than $2 trillion in its IPO, which could happen after the U.S. midterm elections in November following several delays. That would be more than double its estimated $965 billion valuation in May and rank among the largest public debuts in history. For public investors, the listing would offer one of the first direct ways to bet on a leading frontier AI lab. But the filing also lays out Anthropic's own warnings about the technology it builds: more autonomous AI systems could behave unpredictably, create security problems, enable fraud or manipulate information, and pose "existential risks to humanity." Founded in 2021 by former OpenAI employees including CEO Dario Amodei, Anthropic has positioned itself as a safety-conscious alternative to win enterprise customers. Recently it has even lobbied for an AI slowdown, with executives and engineers publicly warning AI could lead to human extinction—while still shipping new products, including a more powerful Opus model last week.