Historical memory prices 1960-2026

Original Article ↗ Hacker News Discussion ↗

Interpreting the graph

  • Chart is log-scale, so early high prices and later changes compress visually; several commenters note many people misread it.
  • Prices are nominal, not inflation-adjusted. Some argue inflation wouldn’t change qualitative trends much on a log scale; others say comparisons like “back to 2010” are misleading without adjustment.
  • Many debate whether $/GB is meaningful across decades:
    • Critics say it ignores how much RAM was “typical” or required for useful tasks at each time.
    • Defenders say $/GB is an objective, reusable unit; “$ per usefulness” is subjective and bakes in assumptions.
  • Some argue it’d be more informative to look at $ per typical workstation memory, or per minimum OS requirement.

Historical context

  • Early systems didn’t think in gigabytes; MBs were huge, core memory was kilobytes.
  • There were rare outliers with multi‑GB RAM in the 1980s, but GB-scale memory only became consumer-relevant in the 21st century.
  • Several anecdotes underline how extraordinary capacities once seemed (e.g., 64 MB, 8 GB laptops).

Software bloat and “usefulness per GB”

  • Many complain modern OSes, browsers, Electron/web apps, containers, telemetry, and antivirus consume vast RAM.
  • Some note that we could do “more” (or at least more efficiently) with tens of MB in the 1990s than with GBs today.
  • Others report perfectly usable experiences on 8–16 GB machines even now, especially for non‑developer workloads.
  • Divergence noted: developers often see 32–64 GB as minimum; typical users are happy on 8 GB.

Price trends, cycles, and causes

  • Broad consensus: we’re paying more per GB than a few years ago, but nowhere near historical highs.
  • Disagreement over how far “back” we’ve regressed: some say roughly 2010; others, looking carefully at the curve and inflation, say more like mid‑2010s.
  • Memory is seen as a classic cyclical industry: overinvestment → glut → crash → underinvestment → spike.
  • Crypto and AI are blamed for recent spikes; others see a slowdown in cost decline coinciding with broader scaling challenges (Moore/Dennard).

Technical and product nuances

  • DRAM scaling is hitting physical limits around ~10–20 nm due to capacitor charge constraints.
  • Questions about multi‑level DRAM (multiple voltage levels per cell) draw skepticism: leakage, tiny capacitors, and performance penalties make it unattractive, unlike NAND flash.
  • Some note the chart ignores speed: each DDR generation brings higher bandwidth, so $/GB alone misses a key dimension.
  • High‑bandwidth memory (HBM) and SSD + cache architectures may shift emphasis from capacity to speed.

Market structure, supply, and policy

  • There is concern that high prices plus fear of overinvestment will prolong tight supply.
  • Some advocate government-backed fab capacity to stabilize prices of this critical input.
  • Discussion of China’s DRAM efforts: they’re close to state‑of‑the‑art node sizes; competition might lower global prices, but suppliers may still price near market levels.
  • Speculation that hyperscalers burned by high prices will try to vertically integrate or diversify suppliers.

Data and methodology issues

  • The dataset descends from an older community resource; people are glad it was preserved but question longevity.
  • Critique that recent DRAM points are for DDR3 and small capacities, potentially painting too-rosy a picture versus DDR4/DDR5 reality.
  • Several emphasize this is mostly a data series, not a full analysis, and that key context (cartel years, production volumes) is missing.