That's probably better explained by TSMC 7nm capacity rather 5nm capacity. I'm sure Nvidia would have preferred even TSMC 7nm over Samsung 8nm.
Nvidia used TSMC 7nm for the first Ampere chip, the GA100 used in the HPC-focused A100. The density vs. the Samsung 8nm manufactured GA102 in the RTX 3090 & 3080 is drastic. 65.6M / mm² vs. 45.1M / mm².
Either Samsung gave Nvidia a hell of a deal or TSMC's 7nm was booked up.
Could also be a bit of both. But I wouldn't be surprised if it's mainly Samsung giving Nvidia a great deal. For years it's all over town that Samsung seems to have the best wafer prices in the industry.
Isn't the rumour that they went to TSMC and "balked" at the price - dangling how much cheaper Samsung would offer their fab capacity, demanded a discount and were told to go whistle since TSMC already has enough customers to run their foundries at 100%?