If they’re talking about unsupervised learning via STDP (which I assume they are based on their documents), unless they’ve had a major breakthrough since their last publication, I don’t think they’ll be able to scale up to any meaningful application that isn’t already dominated by Deep Learning (see IBM’s True North chip). Low-power Neuromorphics have been in development for ages, yet no one has managed to penetrate the machine-learning market. That’s because scaling up to intelligent, marketable behaviour is more than just recognising a pattern, which is all they seem to be doing. There’s no mention of developing explicit models, hierarchies or anything else.