The RAM shortage comes for us all

Original Article ↗ Hacker News Discussion ↗

Apple, integrated RAM, and pricing

  • Debate over whether Apple will raise Mac prices or absorb costs.
  • Some argue Apple rarely increases sticker prices and uses high margins as a buffer; others expect 20%+ hikes or unchanged upgrade pricing with lower margins.
  • Clarification that Apple’s “unified memory” is on-package DRAM dies from the same few suppliers, not on-die and not made by Apple.
  • Consensus that base RAM/SSD tiers are unlikely to become more generous soon.

Scope and mechanics of the shortage

  • Commenters say all major DRAM types (DDR5, DDR4, LPDDR, HBM) are up 2–4x, though not equally.
  • Explanation: the same fabs ultimately produce the DRAM dies; shifting capacity to one type (notably HBM) constrains others.
  • DDR4 production is being wound down; remaining supply is “new old stock” or pulled from used systems and is also spiking.

AI datacenters, OpenAI, and wafer deals

  • Strong focus on OpenAI’s reported agreements locking up a huge fraction of DRAM wafers, especially for HBM, as a key driver.
  • Some see this as a strategic “choke point” to slow competitors and local compute; others frame it as a massive, risky bet that AI demand won’t collapse.
  • Dispute over motives: from “simple supply-and-demand, no nefarious intent” to accusations of market-cornering reminiscent of commodity hoarding episodes.

Memory vendors shifting upmarket

  • Big three (Samsung, SK Hynix, Micron) portrayed as prioritizing high-margin HBM and server DDR5 and exiting or de-emphasizing consumer and legacy lines (e.g., killing Crucial retail brand).
  • Mention that manufacturers are wary of overbuilding after past DRAM busts and are explicitly not scaling up aggressively.

Effects on consumers and ecosystem

  • Many personal anecdotes of planned PC/server builds blown up: RAM triples in weeks, SSDs and even refurb HDDs also jump.
  • Concern that small PC builders and DIY will be priced out, while large OEMs and console makers ride on longer-term supply contracts.
  • Some expect future homelab bonanzas when AI hardware is decommissioned; others note datacenter gear’s power/cooling makes home reuse non-trivial.

Bubble vs long-term trend

  • Split between those seeing an unsustainable AI bubble (expecting a crash, cheap RAM, and broader economic pain) and those thinking sustained demand will justify current build-outs.
  • Worry that if AI demand collapses after fabs retool for HBM, both DRAM makers and downstream markets could be badly hit.

Local computing, software bloat, and efficiency

  • Thread frequently veers into nostalgia for efficient software and old machines with tiny RAM.
  • Some fear a drift toward “dumb terminals + cloud AI,” exacerbated by high RAM prices and local LLM costs.
  • Others argue rising prices will eventually force better RAM efficiency rather than end personal computing.

Fabs, new entrants, and timing

  • Long subthread about how hard and capital-intensive it is to start a fab: ASML tools are just one part of multi‑billion, multi‑year projects.
  • Skepticism that new capacity can arrive fast enough to help before mid/late decade, unless demand crashes first.