Latency is the remaining hurdle for local AI. We'll get there, but my prediction is that it will take around 10 years.