Anthropic 拟可终止与 Lambda 的 349 亿美元 GPU 云合同
S-1 文件里挖出的细节很有意思：Lambda 没赶上场地租约截止日，Anthropic 拿到了终止 349 亿合同的权利，但费用按交付才起算，等着反而零成本。做算力采购或关注 AI 基建的人值得看看合同条款怎么设计的。
Anthropic 修改后的 S-1 上市文件显示，Lambda 未能在约定日期前落实数据中心场地租约，触发 2026 年 8 月合同中的退出条款，Anthropic 因此有权终止总额 349 亿美元的 GPU 云合同。Anthropic 分别在 2026 年 5 月、6 月和 8 月与 Lambda 签约租用 NVIDIA GPU 算力，按交付并验收的容量付费，费用在交付后才起算，等待期间不产生成本。
合同的退出机制依托场地租约：由于 GPU 需要部署在 Lambda 负责租赁的数据中心内，Lambda 错过锁定场地的截止日期即构成违约条件。终止合同可免除 349 亿美元的承诺支出，但代价是失去短期内难以替代的算力，只能转向成本明显更高的短期算力采购。目前双方均未发出终止通知，Anthropic 表示“目前无意”退出覆盖其约 413 亿美元 Lambda 承诺中大部分的合同。
这组数字反映出头部 AI 公司算力承诺规模与执行风险并存：交付挂钩付费和退出条款为买方保留了灵活性，但场地、电力等基础设施环节仍可能拖累整个算力链条。
Anthropic can now cancel a $34.9B GPU cloud contract because Lambda missed its deadline to lease a facility.
From Anthropic's amended its S-1 filing today.
Canceling would erase $34.9B of commitments, but it would lose capacity it can’t quickly replace and lean harder on short-term compute that costs significantly more.
The lease covers the data center site Lambda needs for those GPUs, and because fees start only on delivery, waiting costs Anthropic nothing in fees and keeps its exit open.
It signed deals with Lambda in May, June and August 2026 to rent NVIDIA GPU capacity, paying only as that capacity is delivered and accepted.
The August deal let either side walk away if Lambda failed to secure the facility lease by a set date, and Anthropic’s amended IPO filing confirms it didn’t.
Neither side has sent notice, and Anthropic doesn’t “currently intend” to exit a deal covering most of its roughly $41.3B Lambda commitment.
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Anthropic, Lambda, NVIDIA, GPU云算力.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai). Cross-source corroboration: 11 independent sources have covered this development (including Decoder, IT之家, Xianbao Qian), confirming industry-wide signal strength.
