According to a Nomura semiconductor report released on July 2, the firm believes it is premature to declare a peak in chip stocks, citing AI model iteration, growing inference demand, expanded data center buildouts, and tight supply in advanced packaging and memory.
Nomura argues that AI capital expenditure reflects competitive pressure among large cloud platforms—Microsoft, Google, Amazon, and Meta—rather than short-term individual choices. As long as these companies compete for AI models and inference traffic, they will continue securing chip and packaging resources despite rising costs. The report highlights that supply chain bottlenecks extend beyond GPUs to advanced packaging, substrates, HBM, and server assembly, with smaller substrate suppliers emerging as potential constraints even as TSMC expands CoWoS capacity.