Memory just crossed another threshold. According to a DIGITIMES report (spotted by TweakTown and covered by IGN this week), the DRAM and HBM capacity of the three big memory makers — Samsung, SK Hynix, and Micron — has been sold out through 2027 under long-term purchasing agreements, with “no further supply planned” for 2027 and 2028 not looking much better either. The buyers locking up that capacity are primarily not PC or phone makers. They are AI companies.
The caveat first: none of the three manufacturers has confirmed the report. But a year of component price movements is doing the corroborating for them.
What the Report Says
The key detail is the shape of the deals: five-year long-term purchasing agreements that pre-order memory that has not even been manufactured yet. This is less like buying chips and more like reserving fab output in advance — the same logic as long-term GPU compute contracts. The difference is timing: a GPU ordered today ships this year, while 2027 capacity is being allocated before a single wafer is scheduled. Buyers are no longer pricing in uncertainty; they are paying to remove it. NAND demand is rising too, but it has not been fully absorbed, since NAND has more suppliers. IGN’s summary of the outlook: unless new capacity comes online, retail prices are expected to climb further, with no relief in sight.
Who Is Buying the Memory
AI data centers. Every AI accelerator card carries HBM, every AI server is stuffed with high-capacity DDR5, and these orders are placed at data-center scale. When training and inference clusters expand quarter by quarter, memory makers naturally give their best capacity to the customers with the longest contracts and the deepest pockets. HBM is the sharpest end of the squeeze: it is built on advanced packaging lines that cannot ramp quickly, so every new wave of AI hardware translates directly into multi-year HBM commitments. For AI companies, locking in 2027 HBM is the same move as locking in GPUs or power: every input to compute expansion — silicon, memory, land, electricity — has become a futures market where you buy early or you wait in line.
Consumers Are Already Paying
The price increases are not abstract. IGN ran the numbers: a Western Digital SN7100 1TB PCIe 4 SSD cost about $110 in January and now lists at $189 — up 52%. The Xbox Series X received a price increase this month. Valve’s Steam Machine launched at roughly $750, above the price Valve originally intended, because of memory costs. The reviewer’s wish was modest: she would like to buy a decent RAM kit for under $500 again. Memory used to be the boring, invisible line item of every build; it is now the one most likely to blow the budget. PC builders and console gamers are subsidizing AI infrastructure, one price hike at a time.
China’s CXMT and the OEM Playbook
The shortage is also redrawing the procurement map. Nikkei Asia reports that HP, Asus, and Acer have begun using memory chips from China’s CXMT (ChangXin Memory Technologies) in products for the Chinese market. Guru3D goes further, reporting that CXMT’s DRAM capacity is itself reportedly sold out through 2027 as major OEMs fight for supply. When the big three’s standard-memory output is locked up by AI contracts, supply that used to be treated as the second-choice option is moving into mainstream product lines from first-tier brands — a signal that is geopolitical and supply-chain at the same time.
What Developers and Buyers Should Do
Three suggestions. First, put DRAM and NAND inflation into hardware budgets: whether you are spec’ing workstations, local-inference boxes, or waiting for cloud prices to fall, do not assume memory prices normalize within 2026. If a project depends on big-RAM machines — in-memory caches, local vector databases, beefy CI runners — price the RAM line item separately and revisit it quarterly. Second, lock hardware procurement early: enterprise refresh cycles and GPU server tenders will see memory take a growing share of cost. Third, expect the increases to propagate down the stack — cloud instance pricing, NAS devices, consoles, and phones will all reflect it eventually; the only question is when.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.