I agree too. Mostly because the RAM that goes into AI is not the RAM PCs need. So, eventually that will stabilise, given this is just rushed produce for now. Maybe Q1 or Q2 of 2027 we will start seeing some decrease in prices, given China is almost competing with the other monopolizers...

Also, because Memory on PCs/Servers does not grow like on AI (which mostly uses HBM chips, which are more difficult/costly to produce, taking more fab space allocation), so basically, eventually AI will make memory cheaper for consumers and the enterprise market too as a result of the need to ramp up more manufacture of new chips (new fabs, upgrades, etc). This is just a "rush" moment, and once that rush ends, there will be plenty of volumes of RAM to be consumed. This happened years ago with the HPC market too (but in less volume/impact).

Now, NVMe and QLC in specific is a different problem. Because there is NO alternative to this (PLC is still very much not a win yet, and it's even more expensive to produce), as there is on HBM RAM (which is DDR). And QLC is already expensive to manufacture, and requires much larger sustained volumes for the datacenter industry than memory does. Simply because they wear out! So, it needs a constant supply to keep the constantly growing data being safe.

TLC can't cope with the density of QLC (aka does not work to replace as a technology), and QLC is gaining traction on the speed too (mostly because of a more mature manufacturing process of QLC itself, and smaller, more efficient chips for the controllers).

Anyhow... if I had money, I would buy a ton of them :D "They don't age if you don't use them" - and are very easy to re-sell!