Nvidia usually secures process node capacity to fabricate multiple products. (One example is that Nvidia secured capacity for TSMC's 7 nm* process node for fabricating A100, BlueField-3, Quantum-2, and ConnectX-7.)Are we totally sure Samsung 5nm is out of the question? If that was the choice it would kind of mirror Tegra X1 being 20nm Maxwell while all other Maxwell GPUs were 28nm.
I believe TSMC's 20 nm* process node was the exception where the Tegra X1 was the only product from Nvidia fabricated using that process node.
And I haven't heard any rumours of Nvidia securing process node capacity for Samsung's 5 nm* process node.
* → a marketing nomenclature used by all foundry companies