Your Next PC Upgrade Just Got Five Times More Expensive, and AI Is the Reason
DDR5 memory prices are up nearly 500 percent in twelve months. A 128GB kit that once bottomed out at 329 dollars now lists at 3,399. The cause is not scalping, it is AI data centres eating the wafers.
If you have priced a computer upgrade recently, you already know something is wrong. Tom’s Hardware published price tracking on Monday showing DDR5 memory up by roughly 380 to 485 percent over twelve months, depending on the kit. A 128GB DDR5-6400 kit now lists at 3,399 dollars against a lowest-ever tracked price of 329. A perfectly ordinary 64GB kit that cost 191 dollars last August now averages 1,118.
The numbers are not a US quirk. German site ComputerBase reports average European RAM prices up 345 percent against September 2025, with hard drives and SSDs both climbing more than 125 percent in the same window. Older DDR4 has not escaped either: buyers fleeing DDR5 have pushed DDR4 kits up between 120 and 180 percent, so the classic advice of “just stay on your old motherboard” no longer saves you much.
Here is what is actually driving it, and it has almost nothing to do with retailers. AI accelerators do not use the ordinary memory sticks that go in your PC. They use HBM, high bandwidth memory, which is built by stacking many thin layers of DRAM on top of each other. Stacking is expensive in raw silicon: a single wafer redirected from ordinary DDR5 to HBM removes roughly three gigabytes of consumer memory from the market for every one gigabyte of HBM it produces. Every memory maker on earth has been rebalancing toward HBM because AI customers pay far more for it. SK hynix, Samsung, Micron and China’s CXMT have all doubled or tripled revenue inside a year. Mainstream DRAM chips are now reportedly worth more than half the price of gold per kilogram.
The uncomfortable part is the timeline. Hyperscale buyers have reportedly already reserved almost all global DRAM output for 2027 with advance deposits. SK hynix chief executive Kwak Noh-jung has warned that 2027 will be the worst memory supply year in the industry’s history, with demand outstripping supply into 2030. ADATA’s chairman went further and suggested a decade. Those are vendor forecasts, not neutral analysis, and vendors benefit from scarcity talk, so treat the far end of the range with some caution. But the direction is not seriously disputed.
What this means for you: if you were planning to build or upgrade a PC this year, budget differently or postpone. Memory is no longer the boring, cheap line item it was for two decades. If you are curious about running AI models on your own machine, this is the practical wall you will hit first, because local models are hungry for RAM above almost anything else, and a 64GB machine is now a serious purchase rather than a mild indulgence. And if you rent cloud computing instead, expect this to arrive on your bill eventually too. For most people the honest answer is: use the machine you have a while longer. That is not defeatism, it is just arithmetic.
Sources
A Legal AI Company Built Its Own Model, and It Started From an Open Chinese One
Harvey launched Harvey II with a Memory feature that learns each lawyer's drafting style, plus Tenet, its first in-house legal model, post-trained from a version of Moonshot's open-weight Kimi K3.
