Honestly, I have a background in semiconductor fabrication process, and my undergraduate was in electrical engineering. Experience in semiconductor and analog circuit design leads me to believe that this is the case.
Such a low amount of noise is extremely rare, and it's even LESS likely from a circuit design that is densely packed devices. It's an annoying design consideration but it's the reality of the game. It's obviously a lot to explain here and requires about 30 minutes to explain to somebody why transistors work the way they do, but that sample of no noise innately from the Nikon is a little unbelievable to me. There is always a noise floor.
I don't know what, but something is going on under the hood that is a part of the process of the output data of all the millions of devices on the sensor substrate.
The noise floor from the sensor is no less than that of Canon sensors. If anything, Canon's sensor has a slightly better SNR than the D800 sensor (although not the D810) which gives it a slightly better image at extremely high ISOs. But Canon uses an off-sensor A/D conversion chip, which seems to introduce a lot of read noise, both random and pattern.
Even if Nikon's black point were clipped, so what? You need to clip the blacks at the noise floor anyway, because anything below it is random noise anyway and not part of the camera's dynamic range - it's just like applying a low-pass filter to the intensity reading of individual photosites. May as well use the 16384 levels of a 14-bit file to record actual detail, not just noise.