Dan Niles Sees a Powerful AI Memory Trade — but Only for the Next Year

Dan Niles sees one of the strongest near-term setups in semiconductors coming from memory.
His core argument is demand.
Niles estimates that agentic AI can consume 10–100x more tokens than traditional chat-based AI, dramatically increasing the amount of compute, memory and storage required to serve each user.
That creates a powerful near-term tailwind for memory suppliers such as Micron $MU and SK Hynix as next-generation AI systems become increasingly memory-intensive.
But Niles is explicitly framing this as roughly a one-year opportunity — not a permanent shortage.
The longer-term threat is China.
Japan disrupted the DRAM industry in the 1980s, South Korea repeated the playbook in subsequent decades, and Niles believes CXMT could now become the next major source of capacity and pricing pressure.
CXMT’s blockbuster Shanghai IPO has given investors a public way to track that threat in real time.
Intel is another wildcard. CEO Lip-Bu Tan has expressed interest in new memory architectures, but the company has not announced a full return to DRAM manufacturing.
For memory investors, the setup is clear:
AI demand looks extremely strong today.
But watch CXMT’s expansion and Intel’s actual strategy before assuming today’s supply tightness lasts forever.