[anigbrowl]

$10 million is cheap. True, it only does discrete optimization problems, but I think you could make your money back in a few years years renting it out at $5k/hour and consulting on translating problems to that domain.

Will some clients be wildly overpaying for something they could do equally well on regular computers? Sure, and they'll love every $ of it because of the bragging rights. Is this an efficient use of the hardware? Certainly not, it'll probably be exploiting <1% of the system's potential. Doesn't matter. People will frequently pay more for novelty than actual utility. If their vanity subsidizes the tiny subset of research computation that would have serious economic benefits, I call that a win-win.