Here is what is going on: http://www.forbes.com/sites/anthonykosner/2014/12/29/tech-2015-deep-learning-and-machine-intelligence-will-eat-the-world/
It’s really interesting what you do. It seems it’s a kind of cognitive computing.
I agree with your point: deep learning is perhaps the most difficult technique from ml, and there are still tons of practical problems there including scalability, but there are start-ups that already master it: AlchemyAPI (now part of IBM) for image and text, MetaMind for image and Twitter sentiment analysis, etc.
Deep learning is only part of this revolution; it includes also other ml techniques, text mining / nlp, cognitive computing, etc. My point is: the most sophisticated engineering gives us a really good chance to compete on the current market.
Google's voice recognition is already in production and is based on deep learning. They even extended their capacity in DL by acquiring DeepMind the last year. Beyond Socher's start-up are resources of Stanford. So I think the results would be very impressive as well.
About AlchemyAPI: it seems IBM is going to extent its cognitive framework based on IBM Watson (mostly nlp primitives + automated reasoning) with AlchemyAPI's feature engineering based on deep learning. The game is going from computational primitives to higher levels of machine intelligence.