Sure, that's fair. If you're aiming for state of the art performance. Otherwise, you can get close and do it on reasonably priced hardware by using smaller distilled and/or quantized variants of llama/r1.

Really though I just meant "it's a no-brainer that they are popular here on HN".

I pay 78 cents an hour to host Llama.

Vast? Specs?

Runpod, 2xA40.

Not sure why you think buying an entire inference server is a necessity to run these models.