The market heard “less memory needed” and sold memory stocks.

That may be too simple.

If AI gets cheaper to run, the real question is not whether memory per model falls. It is whether lower cost expands total usage fast enough to overwhelm the efficiency gain.

Principle: In technology, efficiency often removes one bottleneck by making the whole system bigger.