• Type: spark (subtype: perspective)
• event_id: 2026-05-01-ai-power-delivery-spark
• Source: NuttyCLD AI Power Crisis series — six parts, March 30 → May 1, 2026:
  • Part 1 — Two languages of electricity (2026-03-30)
  • Part 2 — Rack becomes new grid (2026-04-17)
  • Part 3 — Last few centimeters (2026-04-19)
  • Part 4 — Last micrometer (2026-04-25)
  • Economics of the last 1cm (2026-04-28)
  • The next Bloom Energy (2026-05-01)
• Shift: Perspective drafted and admitted to ACTIVE.json. Four-layer voltage-stack frame (Grid / Rack / Board / Die) established. EV-AI supply-chain crossover identified as the core asymmetry — auto-side bear was directionally wrong, AI is a second demand curve absorbing the slack. Forked from parent ai-power-bottleneck to separate "where does power come from?" (parent) from "how does it reach the transistor?" (this perspective).
• Effect on thesis: Perspective born. Initial key tickers: MPWR, AOSL, VICR, IFNNY, NVTS, POWI, BESI, ON, STM, WOLF, VRT, ETN, AIXG, VEEC, AMAT, LRCX, ASML, MRAAY, INTC, TSM. Parent: ai-power-bottleneck. Initial catalysts list: MPWR Q1 earnings, Vicor Gen 5 ramp, BESI/SK hynix orders, Intel 18A Panther Lake, TSMC A16 SPR timing, NVIDIA Feynman 2028, Delta vs Foxconn, Wolfspeed substrate share, MLCC allocation language.
• Per-ticker (initial reads): MPWR ↑ (~70% Vera Rubin socket / +60% ASP), VICR ↑ (Gen 5 VPD), AOSL ↑ (OpenVReg co-lead, $900M cap, asymmetric), POWI ↑ (topology IP), BESI ↑ (hybrid bonding majors), MRAAY ↑ (MLCC capacity), INTC ↑ (PowerVia 18A in production), TSM neutral (A16 ramp slipped), WOLF watching (substrate-share threshold).

Pre-spark history (informational, not log entries)

The perspective consolidated thinking that had been accumulating across earlier inputs and adjacent threads. Kept here as breadcrumbs in case future archeology asks "where did this come from?" — they are sources of the spark, not their own thesis shifts.

• Earlier scattered references to MPWR, VRT, ETN as monsters in monster-scan outputs
• BESI surfaced via supply-chain-traces agent trace (different paradigm, converged on same name)
• Optical sibling thesis (optical-supercycle) was developed in parallel — same "shorten the path" logic at I/O layer
• Parent perspective ai-power-bottleneck covers the generation side and predates this one

🔄 2026-06-25 full-scan tape check: CONFIRMING (semicap leg)/mixed — AMAT RSI 71 (+43.5% 30D, -0.5% off ATH), LRCX 64, ASML strong-up are the cleanest names on the board, but the wide-bandgap challenger leg the perspective prefers cracked (NVTS -43% 30D, WOLF -35%) and ETN is firmest (not confirming the incumbent-bear); thesis pays via the shovels, not the asymmetry names. (Data tables spliced fresh to the 2026-06-25 close; full cross-scan read in 2026-06-25-full-scan-market-brief §6.)

🔄 2026-07-05 operating book: Operates as a LAYER of book PRIMARY 2 (power complex, anchored at ai-power-bottleneck §Breaks). Spec: operating-book.