You don't buy expensive, power-hungry [hard]ware that's going to cost an arm and a leg to store, power, and cool for the next year when you only need its brute force for a few hours.
But he is planning to do conversions over and over, one after another, handling problems as they occur. As such, one of his goals is that the conversion be as speedy as possible, and he specifically said that he doesn't want to share a CPU with other cloud users. He wants one fast CPU devoted 100% to his project.
It would make sense to at least test the whole concept in the cloud before buying hardware. That costs almost nothing to do, and then it can point you to just what you need in a server.
It seems like most of the proposals on that website are MORE MORE MORE of everything. What about RAM? 64GB, ECC! What about disk? 4 250GB SSDs in RAID10! What about multi-thread? Gotta have 16 cores! What about single-thread? Better make that dual Xeons so that we can use the ECC RAM and since those are the best! Does it need fancy graphics? Nope, so we better build a second system to use as a console for the big one so that it can be put in another room so that you don't hear the hurricane of fans! Wait, noise? Ok, scratch that, let's buy some big fancy cooling rigs even though we aren't overclocking so that it is as quiet as a mouse, but let's still build that second console!
Why not at least profile the thing in the cloud and figure out what you really need?