A global memory chip shortage is starting to bend the roadmaps of the two companies at the center of the AI boom.
According to a report in Herald Business, the squeeze on high-bandwidth memory — the specialized, stacked memory that sits alongside AI accelerators and feeds them data — is reshaping the strategies of both Nvidia and AMD. With supplies from Samsung Electronics and SK Hynix under strain, the report says the two chip designers are looking at lower-spec HBM for their AI products.
That is a meaningful shift in posture. The AI accelerator race has largely been fought on top-end specifications, with each generation promising more memory bandwidth than the last. Settling for lower-spec memory suggests that, at least for now, getting enough parts to ship matters more than winning the spec sheet.
The pressure is not confined to chip designers. In a separate commentary published by The Globe and Mail, Elon Musk — described there as SpaceX CEO — argued that the memory shortage will drive costs even higher, framing it as a counterpoint to worries about an AI capital-spending bubble. In that telling, the constraint on AI buildouts is less a collapse in demand than a shortage of the components needed to satisfy it.
Both threads point the same direction: memory, not just logic chips, has become a bottleneck in the AI supply chain.
It matters because the cost and availability of a single component now shapes what AI hardware gets built, how fast it ships, and ultimately what companies and consumers pay for AI services.