$SNDK “Brother Sun” said: “Isn’t storage always in short supply forever?”

In fact, what large models lack most is HBM (bandwidth for running compute), not NAND (capacity for storing data). After training, they only need a few TB, so you can just confidently install it on an ordinary SSD.
And high-quality data has already run out; once AI starts, it competes on cleaning and compression, not on piling up useless cold data.
Storage is still fundamentally a cyclical commodity. Once manufacturers massively expand production and the infrastructure restocking cycle ends, excess supply will arrive—just later than expected. If you blindly believe in “unlimited growth” when valuations are high, it’s easy to become the bag holder.