# AI 模型瓶颈在算力而非速度

- 来源：Haider. (@haider1)
- 发布时间：2026-09-13 19:45
- AIHOT 分数：31
- AIHOT 链接：https://aihot.news/items/cmtzrk5u90539ro1btizhw3cf
- 原文链接：https://x.com/haider1/status/2099102127922823674

## AI 摘要

AI 模型不需要放慢，因为事情已经极端了

算力才是瓶颈

OpenAI 花了数百万算力，解决了一个千禧年大奖难题，所以很多事可能已经可行。只是算力不够

归根结底取决于超大规模厂商能多快上线更多算力

## 正文

AI models don't need to slow down because things are already extreme

compute is the bottleneck

openai spent millions on compute and cleared a Millennium Prize problem, so a lot may already be possible. there just isn't enough compute

it comes down to how fast hyperscalers can bring more compute online