While I don’t disagree with anything you wrote, the only way that people can actually get their heads around what the DS DAC does is to realise that PDM (aka DSD) and PCM are deeply connected to each other. I like to describe it as DSD being a special case of PCM, because its use of a single bit sample depth gives it unique characteristics which we can exploit to great effect.
The thing that helped me realise this was Ted sharing that he uses the exact same upsampler for DSD and PCM inside the DS DAC. This works because both PCM and DSD directly describe a waveform – plot the points of amplitude versus time, join them with a continuous curve that avoids any frequencies above half the sampling rate, and that’s your analog signal as described by the digital data. They are only intrinsically different in terms of their frequency range and their noise spectrum, but they are practically very different in terms of how amenable they are to numerical processing and what kinds of circuits are needed to perform the D-to-A electrical conversion.
I would also argue that PDM and PWM have important similarities. They’re both based on the principle that you can switch a signal between a positive and a negative voltage at a high switching rate, then low-pass-filter the output to produce a high fidelity signal in the bottom portion of the signal bandwidth. The main difference is that we use PDM as a digital encoding and transmission format, with a requirement for precise amplitude and time domain precision to ensure accuracy of D-to-A conversion, while we use PWM as a transient abstraction to facilitate analog-to-analog conversion (ie amplification) including correction via analog feedback.
Re the original question, the YouTube link posted above is super helpful. I had no idea that Sharp had played with a PDM amp technology or that one existed today in those crazy powered speakers. The presenter does a good job of describing the electronic engineering challenges and downsides of attempting to do PDM at high power levels. Switching precisely between +/- outputs on the order of 70V at several megahertz with vanishingly low jitter is no easy thing to do.