Hard drives on backorder for two years as AI data centers trigger HDD shortage

Shortage drivers and memory supply dynamics

  • Commenters link 2‑year enterprise HDD backorders to hyperscalers pivoting to QLC SSDs, which then drive up NAND and DRAM prices.
  • One camp argues this is largely a demand/mix shock: vendors cut NAND output after a consumer slump and are now ramping back up; fabs exist, they’re just rebalancing.
  • Others insist it is a genuine chip shortage: fab capacity (e.g., DRAM/HBM/NAND) is fully booked for years, lead times are long, and high‑margin AI products crowd out everything else.
  • Several point to deliberate supply constraints and past DRAM price‑fixing as evidence the big memory vendors operate like a cartel.

Apple and consumer hardware impact

  • Debate over whether Apple is shielded by long‑term contracts and its scale, or still constrained because it shares the same DRAM/NAND fabs and wafers.
  • Users report large price jumps for RAM, SSDs, and GPUs versus late 2024; some see Apple machines becoming relatively less expensive, others expect Apple to simply raise prices too.

AI bubble, ROI, and macro risks

  • Strong skepticism that AI can earn back even a fraction of current capex; some expect >90% capital loss.
  • Others argue that in scenarios where AI disrupts most work, AI infra may “lose less” value than other assets.
  • Several worry about macro instability if AI displaces many jobs without UBI: collapsing consumer demand, mortgage defaults, contagion to non‑AI sectors.
  • Counterpoints cite historical tech revolutions where overall economies grew, though commenters stress they produced many losers and unclear new job pathways.

Used hardware, data security, and quality

  • Some anticipate a flood of cheap GPUs and maybe drives if the bubble bursts; others say large providers shred or instant‑erase drives and keep datacenter GPUs in service.
  • Long thread warns that SMART stats can be reset and even capacity faked; “new” drives bought via marketplaces have been found full of old data.
  • Recommendation: buy from trusted channels, physically inspect drives, and do your own wiping; don’t rely on SMART alone.

QLC SSDs and “cold” storage

  • QLC is acknowledged to have lower write endurance but much higher density; with infrequent writes and read‑heavy workloads, it can be “good enough” for cold/read‑oriented storage.
  • Endurance is extended via overprovisioning; commenters disagree on how aggressive this is in practice but agree enterprise SSDs trade raw capacity for longevity and performance.
  • Some note that, at scale, huge QLC SSDs can already beat HDDs on perf/TB and total system cost for certain workloads.

Broader sentiment and analogies

  • Many feel AI is “eating” HDDs, SSDs, DRAM, GPUs, power, and even water, with unclear ROI, likening it to crypto or Chia‑driven shortages.
  • Others see this as a familiar semiconductor boom‑bust cycle that will eventually overbuild capacity and push price‑per‑TB down—after several painful years.