WSJ 曝光 Anthropic 招股书草案：计划七至十年投入 7500 亿美元算力
WSJ 看到了 Anthropic 的招股书草案，7500 亿美元算力预算、大客户集中度、AI 灾难保险风险都写在里面，细节很猛。
据 WSJ 披露的 Anthropic 招股书草案，公司计划在未来七至十年内支出约 7500 亿美元用于算力，其中包括与 Google 的 1110 亿美元、AWS 的 1100 亿美元、Microsoft 的 310 亿美元合同，以及与 Broadcom 的 1600 亿美元。另有 1500 亿美元的算力交易正在谈判中，Nscale 和 Lambda 占 860 亿美元。2026 年上半年约五分之一的营收来自两家未具名客户，公司预期未来将长期依赖少数客户。招股书还警告，若发生灾难性 AI 事件，损失可能大到无法获得保险覆盖。
The WSJ has seen Anthropic's new draft prospectus. Some numbers:
They plan to spend three-quarters of a trillion dollars on compute over the next seven to ten years.
- $111 billion on Google cloud computing contracts - $110 billion with Amazon Web Services - $31 billion on Microsoft - $86 billion on Nscale and Lambda - $160 billion on Broadcom - $150 billion in additional compute deals under negotiation
I don't see Elon/Colossus listed here? So unless they put that under future negotiations (which seems odd) I think that's another $85 billion through 2029. Unless it's been been cancelled? It only requires 90 days notice to do so.
Around one-fifth of Anths revenue in the first half of 2026 came from just two unnamed customers. They expect a 'significant portion' of their revenue will come from a 'limited number of customers for the foreseeable future.'
They said they bear liability for certain harmful model behaviors, and warn that the 'harm that could result from a catastrophic AI-related incident' might be so large that it would be unable to obtain insurance to cover the potential damages.
