Anthropic宣布与SpaceX建立合作伙伴关系,以显著提升其计算能力。公司将获得Colossus 1的全部容量,并在一个月内增加超过300兆瓦的电力及22万块以上NVIDIA GPU。此举直接导致Claude各项服务的用量上限大幅提高:Claude Code的5小时限制翻倍,Pro和Max版本取消高峰时段限制,Opus API的速率限制也急剧提升。其中,Opus API第四层级现已支持每分钟1000万输入令牌和80万输出令牌,表明其目标已转向处理大型智能体工作负载。Anthropic正在构建一个跨NVIDIA、AWS Trainium和Google TPU的多供应商计算堆栈。
Anthropic ties up with SpaceX to get more comput
Taking all capacity at Colossus 1, adding 300MW+ and 220K+ NVIDIA GPUs within a month.
The immediate change is higher ceilings: Claude Code gets doubled 5-hour limits, Pro and Max lose peak-hour cuts, and Opus API limits rise sharply.
Opus API tier 4 now reaches 10M input tokens/min and 800K output tokens/min, which points at large-agent workloads rather than chat-sized requests.
Anthropic is building a multi-supplier stack across NVIDIA GPUs, AWS Trainium, and Google TPUs, backed by Amazon, Google, Broadcom, Microsoft, NVIDIA, Fluidstack, and now SpaceX.