The wafer math: every AI server steals the RAM out of your next laptop
The mechanism behind the price shock is brutally simple and physical. There is one global pool of DRAM and NAND wafers, and AI data centers are now draining it. Industry coverage frames it bluntly: every silicon wafer going into an AI server is one fewer wafer being made into the RAM chip inside your next laptop [1]. Data centers are estimated to consume roughly 70% of the memory chips produced worldwide, and the three makers who control over 95% of DRAM output - Samsung, SK Hynix and Micron - have strategically reallocated capacity toward high-bandwidth memory for AI accelerators, lifting HBM to about 23% of DRAM wafer output in 2026 from roughly 19% [2][3]. That reallocation is the consumer's problem, because HBM wafers are not making consumer DIMMs.
The result is a supply pool that is already fully committed. Micron's 2026 HBM4 capacity is sold out and meeting only 50-65% of key-customer requests, and SK Hynix's HBM, DRAM and NAND are essentially sold out for the year [2]. When supply is gone and demand keeps climbing, price is the only release valve - and it moved fast: TrendForce pegged 1Q26 DRAM up 55-60% quarter over quarter and NAND up 33-38%, with some reporting putting full-year DRAM increases near 200% [4]. That is why Apple's official statement points the finger squarely at the rapid expansion of AI data centers as the cause of the surge in memory and storage demand [5].



