RAM now represents 35 percent of bill of materials for HP PCs

Original Article ↗ Hacker News Discussion ↗

Impact on consumers and platforms

  • Many commenters feel lucky to have recently “overbought” RAM; others report being unable to source DDR4/DDR5 at any price or only with uncertain lead times.
  • PC buyers note that OEM RAM upgrades are suddenly competitive or cheaper than third‑party modules.
  • Several argue that high RAM prices make MacBooks more attractive: good perf/W, Apple’s supply-chain contracts, and unified memory that’s not directly competing with commodity DIMMs.
  • Counterpoint: Apple’s Linux support and lack of upgradability remain dealbreakers for some.

Why RAM is so expensive

  • Dominant explanation: an extreme demand shock from AI/LLM workloads, especially HBM and datacenter DRAM, with limited short‑term fab capacity.
  • Some claim OpenAI pre‑bought a huge fraction of DRAM wafers; others doubt the exact “40%” figure and say numbers are speculative.
  • RAM makers are described as an oligopoly with a history of boom/bust and alleged collusion; they’re seen as reluctant to overbuild capacity in case AI demand proves transient.
  • Debate whether modern DRAM truly needs EUV; several say much is still made on DUV, but fabs and tooling remain hugely capital‑intensive and slow to ramp.

Will prices normalize?

  • One camp: this is cyclical. As new HBM/DRAM fabs from existing players and Chinese entrants come online over 2–5 years, prices should drop toward “normal,” though maybe not to 2024 lows.
  • Other camp: AI demand (datacenter now, on‑device later) is durable, so high prices may persist unless there’s an AI crash.
  • A minority predict an AI investment bust, after which RAM prices “tank” and consumers benefit from cheap large‑model local inference.

Europe, China, and industrial policy

  • Strong thread arguing Europe should build its own DRAM capacity for strategic resilience, analogous to food production.
  • Objections: loss of manufacturing know‑how, lack of DRAM IP, high energy and labor costs, slow regulation, and intense Asian subsidies and ecosystems.
  • Some see new EU fabs as risky: by the time they’re online, prices might crash and plants become uneconomical without long‑term subsidies.
  • Others counter that this is precisely why to start now; strategic tech supply shouldn’t be left entirely to market cycles.

Software bloat, optimization, and “RAMmageddon”

  • Many hope high prices will finally penalize bloated software: Electron apps, heavy web stacks, memory‑hungry tools like Teams/Outlook, etc.
  • Skeptics think we’ll instead get under‑RAMed “thin clients” and more cloud dependency, not leaner software.
  • There’s detailed discussion of how modern performance often intentionally trades RAM for speed (caching, GC behavior, lookup tables no longer beating compute), and how cache behavior now matters more than sheer RAM.
  • Embedded and low‑end Android work already treats RAM as a hard cost driver; some report active efforts to cut footprint because low‑end BOMs are becoming untenable.

Gaming, GPUs, and VRAM

  • RAM and VRAM constraints are expected to push more optimization in games, but several argue modern games already run well on older hardware if you dial down textures.
  • Complaints that GPU VRAM capacity has stagnated at mid‑range price points for a decade; some foresee APUs (e.g., Strix Halo‑class) eventually replacing discrete GPUs for many users.
  • VRAM and DRAM scarcity is seen as another lever for big AI players to keep local, consumer‑grade models less capable.

Other themes and anecdotes

  • Historical comparisons to past DRAM shocks (e.g., 1990s, 1999 Taiwan quake) but consensus that today’s AI‑driven squeeze is larger and longer‑lasting.
  • Some entertain a “strategic hoarding” theory: AI firms buying wafers not just for need but to starve both competitors and on‑device open‑source AI.
  • zram/compressed memory and swap on fast SSDs are mentioned as partial mitigations, but not substitutes for workflows that truly need large contiguous RAM.