Here is Freitas’ latest estimate, from his “comprehensive nanorobotic control” article:
What will it cost to develop a nanofactory? Let’s assume research funds are spent in a completely focused manner toward the goal of a primitive diamondoid nanofactory that could assemble rigid diamondoid structures involving carbon, hydrogen, and perhaps a few other elements. In this case, we estimate that an ideal research effort paced to make optimum use of available computational, experimental, and human resources would probably run at a $1$ M/yr level for the first 5 years of the program, ramp up to $20-50 M/yr for the next 6 years, then finish off at a ~$100 M/yr rate culminating in a simple working desktop nanofactory appliance in year 16 of a ~$900 M effort. Of course the bulk of this work, after the initial 5 year period, would be performed by people, companies, and university groups recruited from outside the Nanofactory Collaboration. The key early milestone is to demonstrate positionally-controlled carbon placement on a diamond surface by the end of the initial 5 year period. We believe that successful completion of this key experimental milestone would make it easier to recruit significant additional financial and human resources to undertake the more costly later phases of the nanofactory development work.
So, about a billion dollars and 16 years. Say they started in 2020, we’d expect a nanofactory around 2036, by this estimate.