Dell COO Jeff Clarke said that growth in AI tokens demands more memory to compute. To provide such computational power, it is necessary to expand data center capacity through supply of advanced memory chips, which are high-performance GPUs made by tech giants like NVIDIA. These GPUs need High-Bandwidth Memory (HBM) to run LLMs, enchancing AI infrastructure.
To procure HBMs, AI chip companies like NVIDIA approached memory makers like SK Hynix when the memory market was depressed around 2022-23, and secured long-term contracts. Memory makers started serving as dedicated partners, serving one customer, pivoting to customized HBMs, a protfolio transformation. Thus, in October 2025, SK Hynix sold out its memory supply for the year of 2026 to NVIDIA.
Making HBM causes 3-to-1 wafer penalty where fabricating 1-bit of HBM consumes wafers roughly worth 3-bits of conventional memory (DRAM). And there’s a shortage of clean rooms for conventional memories. These factors reduce production of DRAMs and NAND, causing shortage. Some memory makers like Micron even announced exit from their end-consumer business to manufacture HBMs.
This shortage leads to rise in DRAM prices. DRAM is used in laptops, smartphones, gaming consoles, graphics cards, DTH set-top boxes, etc., therefore, resulting in pricey devices. To keep the prices reasonable, some makers are even reversing memory capacity expansion in devices that leads to mediocre performance. Memory procurements by IT firms are either delivered late or completely stalled. Hence, slowing down IT services. Games are not upgrading, and delaying bugfixes due to pricey memory requirements.
So, now you’re paying more for a mediocre device. And also paying more for slowed down tech products and services. As I’ve read the linked articles and you should read them as well, you can see this pivot by memory makers to AI has slowed down your pricey device!