What's this about?
Memory prices have risen roughly 500% in a year, reaching as much as ten times the lowest prices ever tracked — 128GB of DDR5 now costs $3,399. Storage has followed: one commenter's 4TB M.2 drive went from £229 in September 2025 to £500, and a £30 1TB SATA SSD now costs over £100. The cause is the AI build-out consuming manufacturing capacity, and the effect is that ordinary consumers and small businesses are paying for it.
What exactly is driving this?
Memory fabrication capacity is fixed in the short term — a new fab takes years and billions. AI accelerators need enormous quantities of high-bandwidth memory, and the manufacturers who make it are the same ones making consumer DRAM. When one buyer will pay far more per wafer than the consumer market, production shifts. This is the same dynamic as the GPU shortages during crypto mining and COVID, with one difference commenters kept noting: those were bubbles that popped in months, while data-center memory demand is backed by multi-year capital commitments.
What's the community saying?
- The best question in the thread wasn't about RAM. "If the frontier model business wasn't economically feasible without VC and Nvidia money three years ago, how is it feasible now?" — component costs rising while model prices fall by half is a hard circle to square.
- The repair trap. Several noted the danger isn't buying new, it's a stick of RAM failing — people with adequate machines may be forced into a market they'd planned to avoid for years.
- "The fake economics of the AI sector are damaging the real economy." The most-upvoted expression of a widespread sentiment.
- A programming-languages tangent that got traction: "high time to stop using memory-inefficient languages for serious work" — efficiency becoming an economic argument rather than an aesthetic one.
- The GPU-shortage veterans. People who overpaid during the crypto era warning others not to repeat it, while acknowledging this shortage has different and possibly more durable causes.