Musk Allocates 220,000 GPUs to Claude, Doubling 5‑Hour Limits and Building Space‑Based Compute

Colossus 1 Supercomputer

SpaceX AI built the Colossus 1 cluster from scratch at record speed. It contains more than 220 000 Nvidia GPUs—including H100, H200 and GB200 accelerators—distributed across data‑center sites that draw over 300 MW of power, comparable to the electricity consumption of a small city.

Anthropic Compute Partnership

All Colossus 1 capacity has been allocated to Anthropic for running its Claude model. The allocation instantly doubled Claude’s five‑hour usage quota while leaving the weekly quota unchanged, resulting in faster overall consumption of the compute pool.

Claude paid‑user reports indicate higher limits and reduced latency, whereas Grok users on the same infrastructure receive only a single “?” response, illustrating the performance gap created by the compute boost.

Anthropic’s Existing Supplier Landscape

Anthropic already holds multi‑year compute agreements worth billions of dollars with Google, Amazon, Microsoft and Nvidia, so the partnership does not eliminate other options.

Service Reliability Issues

Claude’s rapid user growth has caused frequent rate‑limit throttling and service interruptions. The official Claude status page shows that in the past 90 days more than half of the time the service experienced partial (yellow) or major (red) outages.

Response Strategy to Compute and Power Scarcity

Short‑term: Reduce user throttling to keep customers from being blocked.

Mid‑term: Avoid reliance on a single supplier that can become a bottleneck.

Long‑term: Diversify across hardware generations, geographic locations, power sources and deployment paths.

Strategic Context

SpaceX AI (merged with xAI) is simultaneously developing its own Grok model to compete with Claude while renting out the massive Colossus 1 capacity to Anthropic. This “enemy‑of‑my‑enemy” arrangement coincides with upcoming IPO plans for SpaceX, OpenAI and Anthropic in 2026, a factor noted by analysts.

References

https://www.anthropic.com/news/higher-limits-spacex

https://x.ai/news/anthropic-compute-partnership

Code example

往
期
推
荐
1、
高德二面：线上慢SQL导致CPU飙升，怎么解决？
2、
十万个why：有了 Spring Cloud Gateway，为什么前面还得挡一层 Nginx？
3、
Spring Boot 4.0.6 发布，紧急修复了 8 个安全漏洞，赶紧看看自己的 Spring Boot 版本！！
4、
Spring Boot 3 的 EasyExcel 快速入门教程（已验证）
5、
极简落地！SpringBoot3 注解式敏感词高效处理
6、
正斜杠 vs 反斜杠，让全球程序员多写了 100 亿行代码
点
分
享
点
收
藏
点
点
赞
点在看