Skip to content

Tighten supply-chain dynamics: cap parallelism, fix expansion floor, …#97

Merged
jakechinitz merged 1 commit into
mainfrom
claude/review-supply-demand-dynamics-IC3Dm
May 21, 2026
Merged

Tighten supply-chain dynamics: cap parallelism, fix expansion floor, …#97
jakechinitz merged 1 commit into
mainfrom
claude/review-supply-demand-dynamics-IC3Dm

Conversation

@jakechinitz
Copy link
Copy Markdown
Owner

…centralize intensities

  • Expansion floor (5% of capacity per trigger) previously bypassed maxAnnualExpansion for nodes with parallelism caps (grid_interconnect, transformers_lpt, dc_construction, etc.), letting them grow ~3x faster than their stated labor/permitting limits. Floor is now bounded by the per-node monthly cap.
  • Added maxAnnualExpansion to packaging/memory/foundry nodes (hbm_stacks, cowos_capacity, advanced_wafers, hybrid_bonding, abf_substrate, osat_test). Under severe shortage the elasticity formula could otherwise drive organic growth to 150-285%/yr, well above any historical real-world rate.
  • DEFAULT_STARTING_BACKLOGS now derived from nodeIntensityMap x GPU backlog instead of hardcoded per-component magic numbers, so component backlogs stay consistent if intensities change.
  • Centralized SSD intensity and DC labor factors (worker-months/MW, FTEs/MW) into TRANSLATION_INTENSITIES; previously hardcoded in the engine and unreachable from assumptions.
  • Removed unused hybridBondingPackageShare assumption.
  • Fixed efficiency calibration comment to reflect that H_memory is multiplied into the inference gain (5.25x Y1, not 4.2x).

https://claude.ai/code/session_01Mx47TwSbtizZ64FWztSDkU

…centralize intensities

- Expansion floor (5% of capacity per trigger) previously bypassed
  maxAnnualExpansion for nodes with parallelism caps (grid_interconnect,
  transformers_lpt, dc_construction, etc.), letting them grow ~3x faster
  than their stated labor/permitting limits. Floor is now bounded by
  the per-node monthly cap.
- Added maxAnnualExpansion to packaging/memory/foundry nodes
  (hbm_stacks, cowos_capacity, advanced_wafers, hybrid_bonding,
  abf_substrate, osat_test). Under severe shortage the elasticity
  formula could otherwise drive organic growth to 150-285%/yr, well
  above any historical real-world rate.
- DEFAULT_STARTING_BACKLOGS now derived from nodeIntensityMap x GPU
  backlog instead of hardcoded per-component magic numbers, so component
  backlogs stay consistent if intensities change.
- Centralized SSD intensity and DC labor factors (worker-months/MW,
  FTEs/MW) into TRANSLATION_INTENSITIES; previously hardcoded in the
  engine and unreachable from assumptions.
- Removed unused hybridBondingPackageShare assumption.
- Fixed efficiency calibration comment to reflect that H_memory is
  multiplied into the inference gain (5.25x Y1, not 4.2x).

https://claude.ai/code/session_01Mx47TwSbtizZ64FWztSDkU
@jakechinitz jakechinitz merged commit 55c58c9 into main May 21, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants