Memory everywhere not no one can buy it
AI kit is eating wafer capacity, and the rest of the industry is left fighting over scraps as prices jump.
Silicon has picked a favourite, and your laptop is not it. According to EETimes, Analysts are calling it the “Great Memory Pivot,” with 2026 opening on a split market where AI infrastructure is booming and consumer electronics are facing shortages.
The crunch is coming from wafer capacity being diverted into high-margin AI parts, particularly high-bandwidth memory, which is muscling out the chips that typically feed phones, PCs and consoles. For SK Hynix, Samsung Electronics, and Micron, the AI boom has triggered a profit supercycle that rewards value over volume.
After choking on excess inventory in 2023, the big three have swung production towards HBM and enterprise DDR5 for generative AI workloads. PC memory modules with their slim margins are being ignored, while advanced server modules can hit margins as high as 75 per cent, which is better than many AI accelerators.
Manufacturers have effectively sold out HBM capacity for the year, and Micron is even talking up a gross margin of 68 per cent in Q2 2026. The shortage is not just demand hype, because HBM3E burns roughly three times the wafer space of standard DDR5 thanks to bigger dies and uglier yields from vertical stacking.
That makes every AI wafer a consumer wafer that never gets made, and it is already hurting PC brands, according to IDC bean counters.
“This is a zero-sum game: every wafer allocated to an HBM stack for an Nvidia GPU is a wafer denied to the LPDDR5X module of a mid-range smartphone or the SSD of a consumer laptop,” IDC said.
The knock-on is nasty and straightforward, and it lands on anyone buying kit in 2026. Analysts reckon average DRAM prices could rise by 50 to 55 per cent in early 2026, with consumer electronics prices increasing by up to 20 per cent once vendors pass the bill down the chain.
Grey Box Shifter, Dell’s chief operating officer, Jeff Clarke, warned investors that the company has “never seen costs move at the rate” they are currently rising. Lenovo CFO Winston Cheng admitted the company is now “stockpiling memory chips and other critical components” to buffer against shortages.
That hoarding tightens the vice, especially when everyone tries it at once. The market split is already showing up in the forecasts, with smartphones and PCs expected to decline by a few points while average selling prices creep up as makers quietly cut specs in cheaper models.
PCs look particularly awkward thanks to the so-called “AI PC paradox”, where vendors want to sell local AI machines that need 16GB to 32GB of memory, right as memory becomes harder to source. As memory costs climb to as much as 18 per cent of a PC build cost, some modular PC brands have even stopped selling individual components to discourage hoarding.
IDC predicts PC shipments could fall by nine per cent in 2026 as “AI PCs” become priced out of reach for many buyers. In the data centre lane, AI servers are still expected to post double-digit growth, with premium pricing baked in by hyperscaler spending and HBM scarcity.
The scramble for capacity is getting weird, with Micron discussing a deal with Taiwan’s Powerchip Semiconductor Manufacturing Corp to tap its 12-inch fab space because Micron’s own lines are flat-out. China is eyeing the gap, with ChangXin Memory Technologies planning a $4.2bn (€3.6bn) IPO to expand capacity and target the boring mainstream memory that everyone else is neglecting.
ByteDance and Alibaba are stockpiling. ByteDance is reportedly planning to spend almost $23bn (€19.7bn) on AI infrastructure in 2026, thereby removing more chips from circulation before anyone else can get near them.
The industry used to treat cheap memory as a free upgrade path, but now it acts as if it had never encountered a constraint before. The “AI tax” is showing up in the bill of materials, and it will make your next mid-range laptop feel like a premium indulgence.
