Memory chips function as the public market's closest proxy for AI infrastructure spending, with revenues tied directly to server unit volumes and data center capital expenditure. CNBC host Jim Cramer said investors are already rotating out of the memory-chip winners from that buildout and into companies with growth that sits outside the data center cycle.
The derivative structure of memory demand is what made these stocks the AI trade. Every system commissioned for AI training or inference carries a memory specification. Memory revenues therefore move with the number of systems being deployed, making them a concentrated bet on buildout rates. Stocks at the center of that demand chain are the ones with the most to reprice when spending assumptions shift.
Cramer's read is that institutional money managers are making that shift now. Where the capital is going, by his account, is toward companies with earnings drivers outside the data center supply chain. His reported comments put the rotation as current rather than anticipated.
Memory carries baseline cyclicality that predates the AI infrastructure story and ranks it among the more volatile corners of the semiconductor sector in ordinary periods. A single dominant demand thesis amplifies that baseline. When that thesis softens, as Cramer described, the exit tends to be faster than the entry was.