Thanks for sharing this. While PS5 and XBSX/S were launched at pretty much exactly the wrong time from a supply chain point of view, it's possible that the new Switch model could be the opposite; launching at a time when new capacity has been coming online, while demand is dropping. Of course there's more to it than just the SoC, as there's always the possibility of shortages of smaller components like display driver ICs, PMICs, or whatever, but it looks like manufacturing will be much less supply-constrained in 2023 than it has been in the past couple of years.
The other thing I just noticed on SemiAnalysis is
this article. Mainly regarding Nvidia's costs for Ada GPUs, it includes this comment, which I think is quite relevant for this thread:
I've been saying for a while that, despite the increased wafer cost of TSMC's 5nm processes, it could actually be cheaper per chip than Samsung 8nm, and this seems to confirm it.
If we assume Drake is 8 billion transistors (just a rough guess), on 8nm we would expect a 175.4mm2 die size, and 337 dies per wafer. Assuming 0.1 defects per cm2 on 8nm, that gives us an average of 280.1 yielded dies per wafer. On TSMC 5nm, we would get a 66.1mm2 die size, and 916 dies per wafer. With a 0.1 defect rate per cm2, that would leave 854.4 yielded dies on average per wafer. Combine these with the 2.2x price difference between 8nm and 5nm, and you get a Samsung 8nm Drake being 39%
more expensive than a TSMC 5nm Drake. I would actually expect the real numbers to show a larger gap, as Samsung 8nm yields are unlikely to be as good as TSMC 5nm yields. If we bump the 8nm defect rate up to 0.2, it becomes 70% more expensive on a per-die basis.