My thinking is: there are many ways to make matter do compute. Semiconductor-based digital logic chips are only
one of many possibilities.
You can make an analog computer from just about anything, it's a matter of identifying things you can interpret as flows and stores (or, as the professors who taught control systems at my uni, faucets, drains and bathtubs), and arranging them just right. You can make a digital computer from just about anything too - it's a matter of identifying things you can interpret as a NAND gate, and stacking them just right. You can encode a neural network model as grooves in plastic sheets, stack them into layers, and have it naturally process incoming light. Etc. The possibilities are limitless, because computing is more about your mental model, and less about the substrate itself.
So I imagine, the breakthrough for nanotech will be when someone figures out a way to make some kind of programmable computer that works at nanoscale, can be easily produced, and is somewhat robust against the environment. It doesn't need to be a fast computer - we're used to multi-GHz CPUs and multi-MHz microcontrollers, but at nanoscale, even an equivalent of an old 8008 will do, or even something much weaker: single nanobot doesn't need much compute. You'll be deploying them by thousands or millions at a time anyway, and you'll want to program them as a swarm anyway.