Memory chips, long considered the commodity basement of the semiconductor world, are about to become the majority of it. Gartner’s latest forecast projects memory semiconductor revenue will reach $837.3 billion by the end of 2026, accounting for 54% of the projected $1.555 trillion global semiconductor market.
That’s up from just 27% in 2025. In other words, memory’s share of the total chip industry is expected to double in a single year.
The numbers behind the surge
DRAM revenue is projected to climb 246.6% year-over-year in 2026, while NAND flash is expected to jump by 371.9%. The driver is AI. Specifically, the insatiable appetite of AI data centers for high-bandwidth memory (HBM) and advanced DRAM to feed the compute demands of training and running large models. Gartner expects AI data centers to account for 36.5% of total semiconductor revenue by 2026, eventually surpassing 53% by 2030.
Memory semiconductor revenue is expected to surpass $1 trillion in 2027, as the broader global semiconductor market reaches $1.94 trillion.
Who wins, and who’s spending
Three companies sit at the center of this transformation: Samsung Electronics, SK hynix, and Micron Technology. SK hynix currently holds 56.4% of the HBM market. Samsung leads in overall DRAM production. Micron rounds out the trio as the only major US-based player in the space.
On the demand side, Nvidia’s procurement commitments for memory surged from $119 billion to $279 billion in a single quarter, with payments scheduled through 2029.
The catch: cyclicality hasn’t been repealed
Memory chips have always been the most volatile corner of the semiconductor market. Boom-bust cycles are practically a defining feature. Prices surge when demand outstrips supply, manufacturers rush to build new capacity, and then a glut forms that sends prices crashing. Analysts have warned that this cycle is unlikely to be different in kind, even if the AI demand wave is different in scale.
For Samsung, SK hynix, and Micron, the challenge is balancing massive capital expenditure programs against the possibility that AI spending growth could decelerate before all that new capacity comes online. Memory fabs take years to build and cost billions of dollars.
Nvidia’s $279 billion procurement commitment also creates a new kind of supply chain risk. If Nvidia’s AI accelerator demand were to slow, even modestly, the downstream impact on memory producers would be amplified by the scale of those commitments.