Ask HN: What is the black swan event that could bring down GPU costs?

Re:Ask HN: What is the black swan event that could bring dow... by kbesplv on Thursday October 15, 2026 @05:33PM (#50110631)

Training is 10-100x more compute than inference. The current rush is all training. When it tips over to mostly inference, a lot of that hardware is just going to sit there.

[ Parent | on HN ]

Re:Ask HN: What is the black swan event that could bring dow... by enjoyyourlife on Thursday October 15, 2026 @09:08PM (#50112451)

Isn't it the opposite? It seems like the AI labs are using up way more compute for people to use their LLMs than to train them

[ Parent | on HN ]

Re:Ask HN: What is the black swan event that could bring dow... by verdverm on Thursday October 15, 2026 @10:16PM (#50113069)

MiMo-2.6 RL'd for $3.5M too, training costs are coming down super fast.

The build out is for inference, Jevon's Paradox is what we are doing. Some of this will play into training, but it's definitely more about inference.

[ Parent | on HN ]