You're aware that this would mean the next Nintendo device won't have BC with current Switch games nor would transfer Switch R&D and knowledge, right? Basically the next device would be a reset, blank-slate, from current Switch. Nintendo needs Nvidia if they want to release a Switch sucessor.
Basic due diligence requires they at least consider a non-Nvidia solution. The things you are saying are absolutely true, and it's the job of the due-diligence process to put together a report that explains that clearly in executive speak, showing the loss of BC On related sales outstripping the savings of going with, say, Qualcomm.
Every now and then, the due diligence strawman wins. An example would be the Switch itself, where the TX1 was the due diligence option. We're talking about a 2024 release, T239's specs leaked in 2022. Imagine if the Mont Blanc design had leaked in 2014, right before Nintendo made the move to Nvidia.
All that said, I think that ship sailed by September last year. I can think of one, single example of a product that made it to market, whose chip made it to tape out, that saw any kind of "back to the drawing board" retrenchment.
Honest question from a tech-minded (I'm actually a developer, just in a completely unrelated field) but with no game-specific skills at all: why is that the case?
Just adding to Ol' Footsie's answer here.
Shaders are programs that run on the GPU, and just like programs on the CPU, they've gotta be compiled. Unlike the CPU, shaders on PC are generally compiled just-in-time, the moment the shader is needed, rather than ahead-of-time. There are both technical and historical reasons for this, but the big one is that, unlike x86 or ARM, there is no standardized instruction set for GPUs. AMD, Nvidia, Intel are completely different, and in fact, aren't even public.
AMD has chosen to keep their ISA backwards compatible for the last 11 years. This makes the silicon more complicated, as they have to support old instructions that might be superseded by more modern solutions, or might run poorly on the more modern GPU design, but it also means that they can more easily reuse their investment in their compiler software.
Nvidia has chosen to rethink their ISA every generation. This is a boon to the hardware design, because they can kick old silicon to the curb, and then build the optimal design for the new ISA. But their shader compiler has to support lots of neither backward nor forward compatible strains of the ISA.
For PC gaming, there is no difference for the player. Shaders are compiled just-in-time regardless. This has become a problem in recent years as mid-game stutters while the shader's get compiled are becoming more and more common, but it's equally bad on AMD and Nvidia.
On the server side, AMD's solution is nicer, but in practice, you're recompiling your software often enough that Nvidia's solution isn't exactly painful.
But on the consoles, AMD's solution is a clear winner. In the console space, you generally compile your shaders ahead-of-time. You've only got one hardware target, and you're heavily optimizing for that hardware anyway. By compiling ahead of time you not only eliminate the stutter that PCs have, you can take longer to do the compile, generating a higher quality, more optimized shader binary.
Switch games have all these embedded, compiled shaders, that won't run on any non-Maxwell GPU. Emulation solves this problem all the time - they capture the shader before it hits the GPU, de-compile the binary back to the source code, and then recompile it for the GPU underneath. But that takes a lot of CPU power, and introduces that dreaded stutter that PC games suffer from.
Nvidia and Nintendo will need to solve that for the Successor. Their advantage is that, unlike PC emulators, they only have one target machine to run on. Their disadvantage is they have to do this with what is basically a tablet CPU, and unlike PC emulation, their audience expects little-to-no stutter.