Theses›Datacentre & AI compute›AI-hardware circular economy

AI-hardware circular economy

AI-hardware circular economy: GPU reuse + critical-metal recovery

  • Not dated

The call, a draftRetired AI accelerators feed a growing resale and metal-recovery market, and the value sits at the two ends, with the trusted logistics firms and the recovery chemistry. The window is early, and hyperscalers reusing most of their own hardware caps it.

D1D2D3R1R2C1C2foragainst D1D2D3R1R2C1C2foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the drivers.
  1. D1Accelerators are forced out of frontier roles every 18-36 months, which feeds a steady secondary supply.
  2. D2Server boards hold about 100-300g of gold a ton, roughly 50x virgin ore, plus gallium, germanium and indium that are import-constrained.
  3. D3Old accelerators keep their value, with CoreWeave rebooking 2022 H100s at about 95% of the original price.
  4. R1Microsoft reports 90.9% reuse of its own hardware, so hyperscaler insourcing caps the feedstock left for third parties.
  5. R2The GPU-broking middle of the chain has no lasting advantage, so margin there competes away.
  6. C1Datacentre IT asset disposal and AI-hardware decommissioning pass $19B a year.31 Dec 2030
  7. C2A metal-recovery or robotic-disassembly company working on AI hardware reaches a $500M-plus outcome by valuation, sale or revenue.31 Dec 2030
What would prove itdated tests, open and resolved
20272028202920302031today123456predictionswatch 20272028202920302031today123456predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1Microsoft, Amazon, Google or Meta announces a named partnership for critical-metal recovery or decommissioning of AI hardware between 6 Nov 2026 and 31 Jan 2027.Jan 2027 · open
  2. 2By 2030 datacentre ITAD + AI-hardware decommissioning exceeds $19B annually.Dec 2030 · open
  3. 3A critical-metal-recovery or robotic-disassembly company focused on AI hardware reaches a >$500M outcome by 2030.Dec 2030 · open
  4. 4Datacentre IT asset disposal revenue against $19B a year.Dec 2030 · open
  5. 5Any metal-recovery or automated-disassembly company passing a $500M outcome on AI-hardware feedstock.Dec 2030 · open
  6. 6Whether gallium and germanium export controls make recovery structurally valuable.Dec 2030 · open
Expressionslisted, private, technologies
Technologies