That makes some sense. NAND Flash is massively parallel by its nature. That is rarely exposed outside the die though. You'll have that 8 bit double data rate bu

That makes some sense. NAND Flash is massively parallel by its nature. That is rarely exposed outside the die though. You'll have that 8 bit double data rate bus and you'll learn to like it. Now that model inference at scale is a thing though? Model weights, cached prefixes? There's a considerable demand for "slow writes, fast high bandwidth reads" memory. And every bit of storage you didn't have to use RAM for you can use for fatter KV cache instead.