Summary: Tom’s Hardware reported on August 17 that PCPartPicker data showed DDR5 average prices up about 500% in 12 months, with one 128GB kit moving from $329 to $3,399. The article attributes pressure to AI infrastructure and HBM demand, while reported capacity additions largely arrive from late 2028 onward. Developers and builders therefore face higher memory budgets now rather than near-term relief.
If it has been a year since you last looked at memory prices, the current listings read like a broken page. On August 17, Tom’s Hardware pulled together PCPartPicker’s tracking data: DDR5 average pricing has climbed roughly 500% across 12 months, and a few kits now sit at ten times their all-time lows. One 128GB DDR5-6400 kit that used to cost $329 is listed at $3,399. DDR4, which by rights should be obsolete, has been swept up by the same wave. Cheap memory is finished for the moment — and the capacity announcements made this month will not change that before 2028.
How Big Are the Increases
The bluntest evidence sits in PCPartPicker’s comparison of average prices, August 2025 against August 2026. DDR5-4800 2x16GB moved from $90 to $425, a 372% jump. DDR5-5600 2x16GB climbed from $116 to $528, up 355%, while the 2x32GB version went from $191 to $1,118, up 485%. DDR5-6000 2x32GB rose from $222 to $1,272, a gain of 473%. Builders priced out of DDR5 retreated to older platforms, and that pulled DDR4 up 120% to 180% along the way — a 3200 MT/s 2x16GB kit went from $105 to $281. Europe traces the same curve: ComputerBase measures RAM prices 345% above their September 2025 level, with HDD and SSD prices more than 125% higher. Tom’s Hardware went as far as the absurdist arithmetic: by weight, mainstream DRAM chips are now worth over half as much as solid gold.
Who Bought Up the Memory
No mysterious supply-demand wobble explains this; the driver is AI infrastructure. HBM, the stacked memory soldered beside training and inference accelerators, consumes several DRAM layers per device, so every wafer yields less standard memory. Measured against data-center customers whose orders are counted in gigawatts and billions of dollars, PC and smartphone makers rank low on the memory vendors’ priority list. The reporting carries a more extreme fact still: hyperscale buyers have reportedly reserved nearly all global DRAM capacity for 2027 through advance deposits. The windfall turns up in vendor revenue — SK hynix, Samsung and Micron, plus China’s CXMT, have all doubled or tripled revenue inside a single year.
More Capacity Is Coming, Just Not Until 2028
StorageReview spent that same week auditing the expansion promises from the big three, and the fine print outweighs the headline totals. Micron committed up to $3 billion to the domestic supply chain, a figure that includes $500 million in strategic financing for GlobalWafers’ Texas wafer fab plus a ten-year wafer supply agreement — upstream supply, not DRAM bits. SK hynix’s August 7 announcement covers 54 trillion won (about $38 billion) for new DRAM and HBM fabs in Yongin and NAND lines in Cheongju, with cleanrooms not scheduled until late 2028 and mid-2029. Its roughly $4 billion HBM packaging facility in Indiana, backed by a $458 million federal award and a $500 million loan, targets mass production in the second half of 2028. SK Group chairman Chey Tae-won told CNBC the market feels “like a war” and that prices “went up too fast”; CEO Kwak Noh-jung warned that 2027 will be the worst supply year in the industry’s history, with demand exceeding production into 2030. NVIDIA, meanwhile, has already locked in HBM supply under a $500 billion arrangement with SK Group.
What This Means for Builders and Developers
The near-term playbook is not complicated. Put off the memory upgrades you do not need: a quarter of waiting now buys a lower spec at the same money. Reprice the RAM lines on cloud instance quotes, since providers will hand these costs onward. And in any cost model, drop the assumption that memory keeps sliding down its historical curve — for memory-hungry local inference workloads, bandwidth was the bottleneck already, and capacity itself has now doubled in price, which rewrites the hardware math for home labs and edge boxes alike. The same arithmetic faces anyone quoting client hardware builds this quarter. ADATA chairman Simon Chen even thinks the crunch could run another decade. There is basically one scenario that brings prices back down: a real contraction in AI capital spending. Wishing for a recession just to get cheap RAM back is not a serious position — but the market is now openly pricing consumer memory as a casualty of the AI build-out.