Looks like panic sell off, but main question is:

do models with DeepSeek architecture still scale up?

If yes, then bigger clusters will outperform in near future. NVidia wins as tide rises all boats, and them first.

If not, then it's still possible to run several models in parallel to do the same, potentially big, job. Just like humans team. All we need is to learn how to do it efficiently. This way bigger clusters win again.