Ask HN: You have 50k USD and want build an inference rig without GPUs. How?

This is more like a thought experiment and I am hoping to learn the other developments in the LLM inference space that are not strictly GPUs.<p>Conditions:<p>1. You want a solution for LLM inference and LLM inference only. You don&#x27;t care about any other general or special purpose computing<p>2. The solution can use any kind of hardware you want<p>3. Your only goal is to maximize the (inference speed) X (model size) for 70b+ models<p>4. You&#x27;re allowed to build this with tech mostly likely available by end of 2025.<p>How do you do it?