32GB of DDR5 now costs $375 – AI shortage continues to squeeze PC building

Original Article ↗ Hacker News Discussion ↗

Scale of Price Increases

  • Many reports of 2–5× jumps in 1–2 years:
    • DDR5 2×32 GB kits going from ~$150–200 to $700–900+.
    • DDR4 64 GB kits from ~$85 to ~$375.
    • Even DDR3 has roughly tripled or quadrupled from its lows.
  • SSDs and HDDs also surged: 4 TB SSDs cited rising from ~$250 to $500–700; large HDDs for NAS now $600–700 each.
  • Several people note their existing PCs’ RAM alone is now worth more than the entire system cost when built.

Causes Discussed

  • Strong consensus that AI datacenters are the primary new demand:
    • HBM and server DIMMs pull wafer capacity away from consumer DDR4/DDR5 and even DDR3.
    • Claims that large AI buyers pre-booked huge portions of future DRAM output, triggering panic buying and “dynamic pricing.”
  • Some argue US tariffs / China sanctions and regulatory delays for new fabs are major contributors, especially in the US.
  • A minority claims this is exaggerated and that outside US-tariff channels, prices rose less, though availability is still tight.
  • New capacity: references to multibillion-dollar, multi‑year fab builds; SK Hynix, Chinese DRAM vendors (CXMT, YMTC) expanding but on slow timelines.

Impact on Consumers, Hobbyists, and SMEs

  • PC building and upgrading:
    • Many regret not buying RAM/SSD earlier; some abandon planned builds or stick to old laptops/desktops.
    • Prebuilt PCs and bundle deals (e.g., CPU+board+RAM) are often cheaper than buying parts separately.
    • Mid/high-end gaming PCs are seen as returning to a “prosumer luxury” niche; 32 GB is still OK for gaming, but 64+ GB is now painful.
  • Small businesses and EDA/compute users report extreme quotes for server RAM and SSDs, with short quote validity and long lead times.
  • Concern that consumer PC ecosystem (boards, cases, coolers, etc.) will shrink as high-end component demand collapses.

Local Models vs Cloud AI

  • Some see local/open models as “good enough” for many coding and general tasks and a potential alternative to subscriptions.
  • Others argue local models still lag top proprietary systems, especially for complex, non-coding work.
  • Debate whether widespread local inference would reduce or further increase overall DRAM demand.

Outlook and Broader Concerns

  • Split views:
    • One side expects classic boom‑then‑glut pricing crash once capacity catches up or an AI/datacenter bubble pops.
    • Others note DRAM makers remember past busts and may deliberately avoid overbuilding, so no glut.
  • Broader worries about:
    • “Negative externalities” of AI (hardware, energy, water) on non‑AI users.
    • Long‑term damage to hobbyist computing and new entrants to PC gaming.
    • Whether this is a temporary shock or a “new normal” where high‑RAM systems are luxury goods.