Bobbers, correct, they will keep memory supply limited to cash-in on AI, as they are not a non-profit, after all.
Although demand should be naturally (very) high, considering how many hundreds AI datacenters are being built right now, and running AI locally is an ever-growing thing, too (topic=315954).
Sourcing from CXMT kinda proves that they can't even produce enough for themselves, or at least that its majority goes to enterprise-paying customers?
Memory fabs are also being built. Even closed-weight LLM providers don't seem to have a secret sauce, as Kimi K3 has shown, so as LLMs become bigger, memory demand also will continue to grow, if there's no major breakthrough in LLM architecture.