In the case of Taalas, the pipeline was said to be 2 months:

> From the moment a previously unseen model is received, it can be realized in hardware in only two months ( https://taalas.com/the-path-to-ubiquitous-ai/ )