the architecture they've designed works in a similar way to neurons . . . 15X-300X lower computation energy . . . ideally suited for the kind of processing tasks that brains do rather well: analog-data-sensing, cognitive-computing, associative memory and so on

Servers tend to be doing the same thing repetitively, possibly many times over simultaneously (e.g. get web page, send web page, get web page, send web page). I'm not sure a 'neural' chip is best suited to that job given the description. From the description it'll be much more useful in 'workstation' level tasks like image processing and trend spotting in datasets. You have a 'dumb' server with lots of throughput sending you piles of data then you have a 'neural' chip in your workstation that's sifting and processing it as you define exactly what you're looking for.

Possibly also if we can make a piece of electronics that works more like a nerve cell, it's easier to get it to work with the output a nerve cell gives out (and to give output a nerve cell can process) so it might be a step nearer to honest-to-goodness cybernetics.

Such a chip could be very handy for cybernetic connections. It might start out for aiding disabled people but how long before it becomes used as a way of installing information so that people can take exams? This technology might make exams a waste of time, as you could program all the relevant skills to someone.

You're talking about William Gibson's biosofts or the skills downloading from The Matrix. Be hard to do for anything that requires motor training but purely intellectual skills like languages would seem a possible candidate. There's a world more to do though, before you get 'downloadable skills'. Just having a chip that can talk to a neuron is like having a wheel and you're talking about building a space shuttle.