He's worked on hardware-accelerated conv nets (http://yann.lecun.com/exdb/publis/pdf/farabet-iscas-10.pdf) before and is working on new hardware for deep learning. It's become obvious enough that there's a potential market that multiple companies are working on this problem. Dedicated hardware might allow even bigger deep nets, as well as using deep learning in small devices like cellphones and machines around the home.