I know; that's why I went for the largest connector I could find (400V, 32A)
Though I doubt that the BM1387 can pull 15W anyway, right?

The S9 has 189 of these chips and pulls 1372W at 13.5TH/s, so we should get roughly 70GH/s per chip at 7W of power. That's only 17.5 amps.
This was anyways just meant as a starting ground to figure out the connections and protocol, but obviously Skot has already done all that. I will try to replicate his setup based on my pure breakout and building the rest on breadboard. I actually have 2 of these so I can also try the concatenation aspect and then maybe make an x2 or x4 board.
It may even be easier to build a 0.8V or 1.6V buck converter instead of a 0.4V one due to less drastic voltage drop / higher duty cycle, right?

But mostly, I am looking to build a buck converter next (which I think is also missing in Skot's project) as running off a lab PSU is at best good enough for experimentation.

Yup, he uses a synchronous buck controller IC together with 2 mosfets.
Looks something like this. They are switched on and off by the controller, which creates accurate timings to achieve the required duty cycle and thus output voltage.


I mean 2 of these can be powered through USB; sidehack proved it. But while I'd love to have an open-source USB based open-source miner, I agree that it will be easier, more practical and even cost-effective (economy of scale) to have a power supply circuit feeding a handful of ASIC chips instead of needing a buck per chip. That would then obviously exceed USB specs and be powered directly off of 12V, for sure. Even a 6-pin PCIe connector could be used. They are rated for up to 75W each.