Imagine building a multi billion dollar computing cluster solely for research.

I guess the main reason it hasn't been done is that the deprecation is still huge due to chip advancements.