Anthropic’s Reported $637 Billion in Compute Commitments
TMTPost, citing Anthropic’s October 25 IPO filing, reports that the company has about $637 billion in framework agreements with seven compute suppliers, including Google, Amazon, Microsoft, Broadcom, SpaceX AI, Nscale and Lambda. The total combines agreements with different levels of commitment and is not entirely unconditional spending. The filing’s broadest figure, which exceeds $800 billion, is a theoretical maximum including all options rather than an expected expenditure. The report analyzes how financing, packaging and power constraints could affect AI infrastructure financing.
The scale and duration of the agreements make Anthropic’s cash flow and ability to finance infrastructure central considerations for understanding the risks surrounding AI compute commitments.
（本文作者为 韩工观察，钛媒体经授权发布）
文 | 韩工观察
2026年10月25日，Anthropic把一份335页的招股书摊在华尔街面前。数字漂亮得晃眼：2026年上半年营收162.86亿美元，同比增长13倍；第二季度经营利润9.74亿美元，是史上第一次单季转正。
但真正让投行失眠的是附注里那串数字：未来十年，它向谷歌、亚马逊、博通、SpaceX等七家供应商承诺了约6370亿美元算力支出，其中绝大部分"照付不议"——用不用，钱都得付。
一家刚证明"大模型能赚钱"的公司，同时签下了一张相当于其年收入24倍的长期账单。
这不是悖论。这是AI算力行业第一次把底牌摊在桌面上：技术已经跑通了当期的小账，但全行业没人跑通过远期的大账。市场正在争论"AI有没有泡沫"——这个问题问错了。
#Anthropic#Google#Amazon#Microsoft#Broadcom#SpaceX AI#Nscale#Lambda#OpenAI