A JP Morgan "long software, short semiconductors" basket jumped 12.1% on
Tuesday, its largest single-day gain, reflecting a sharp intra-AI trade
rotation. Investors are pricing a potential slowdown in frontier AI model
training, prompting flows out of some AI infrastructure and long-cycle
semiconductor equipment into software, cybersecurity, large cloud platforms and
lagging AI names. JP Morgan warns that a slower R&D cadence does not equal
halted training—tighter safety, alignment and third-party evaluations could
increase compute demand—so this appears to be a scenario reassessment rather
than confirmation of reduced AI compute needs (JP Morgan report Sept. 15).