Out of curiosity, if for some commercial reasons Nintendo and/or Nvidia wanted to favor Samsung, do you think the other nodes you are talking about could be viable alternatives to TSMC in terms of technological performance?
I understand that Samsung 8nm is the cheapest, but could other Samsung nodes for example also be cheaper than their direct competitors at TSMC?
There are a couple of reasons that no one really expects anything other than SEC 8nm and TSMC 4N. The big one is that picking the "cheapest" viable node for each product is, on the whole, more expensive than just picking a small number of nodes for all products, and sticking to them across the board.
The primary reason is just flexibility. Capacity gets purchased in advance. If you don't use all your capacity, you sell it back to the foundry at pennies on the dollar. You need extra at the last minute you pay a massive premium. Using a small set of nodes, and being able to move your own products around to make room, or use up extra is worth the cost of not being able to perfectly cost optimize every product
Second is expertise. In the design phase, knowing really exact details about node behavior can be useful from mapping out performance targets to setting up the simulator. It's obviously useful during the design phase, and then later when bug-fixing, or building the tools to verify the chips as they come off the line. Specializing in a small number of nodes just makes all of this more efficient,
Third is micro-controller reuse. Chips are full of tiny controllers that are reused across lots of different products. Memory controllers, PCIe controllers, security modules. Just like the rest of the chip, these parts need to be designed for each node they're on. Reusing nodes makes it easier to reuse these tiny subsets of the chip across your various products.
To answer your question, it may or may not be possible for Samsung to offer a more advanced node at a better price than TSMC. Whether or not they could offer such a good price that these additional costs and risk factors are worth it, that's pretty doubtful.
I find the idea that Nvidia and Nintendo could reach a relevant energy budget even with Samsung in 8nm interesting and stimulating, but anyway wouldn’t this node be a problem for future updates (Lite model and others)?
No one really knows! The math suggests that SEC 8nm isn't viable. If it is viable, then our data is incomplete. If our data is incomplete, it's impossible to guess how far that node can go.
My personal assumption is that Nintendo talked to Nvidia to ensure that they wound up with a custom chip that matches their long term needs. That either the chip is usable across a range of formfactors, or Nintendo is confident that a redesign for a new node in the future will be a worthwhile use of money, relative to the savings of using this node now.
I truly doubt that Nintendo has locked in plans for the entire generation worth of revisions. I'm sure they have tentative plans, even
strong tentative plans. But I wouldn't be surprised if they were choosing the best chip for their flagship product first, and willing to figure out the details of revisions later. While I expect a smaller/tougher/cheaper variant at some point, it's hard for me to speculate what Nintendo would want out of a revision of a product that hasn't even been announced yet.