R31-B — AI-capex pricing-in confirmed at chip-IP + AI-networking layers; 3m magnitudes harden the thesis; ChurchillWw grid-stress reinforces generation-side gating

Update

R31-B — AI-capex pricing-in confirmed at chip-IP + AI-networking layers; 3m magnitudes harden the thesis; ChurchillWw grid-stress reinforces generation-side gating

  • Type: observation (subtype: pricing-in-confirmation across multiple cohort layers)
  • event_id: 2026-05-23-ai-power-bottleneck-r31b-ai-capex-pricing-in-confirmed
  • Source: R31-B authorized perspective-touch session — workflow/RUNS.md 2026-05-23 r31b-r26c-ai-capex-perspective-touch. Source ticket: R26-C TASKS-PERSPECTIVES.md AI-capex pricing-in observation, accumulated extensions through R26-C → R27-D → R27-E → R28-E → R29-C → R30-B (6 extensions over ~10 days). Underlying scan substrate: R30-A precomputed summaries{ai-power,ai-infrastructure,optical-supply-chain,cybersec,semis}.json(fresh 2026-05-22 bars; strict-freshness gate PASSED on critical-market baseline). Capture substrate: R30-Bpublished/captures/2026-05-23-r30b-discord-authenticated-tab-recovery-sweep.md` (ChurchillWw 2× grid-stress data points).
  • What's new (one observation, four named evidence layers): AI capex is now visibly being priced in across the thesis's full vertical — chip-IP royalty layer (ARM), AI-networking equipment (NOK / EXTR / FFIV / ATEN / ERIC / CRDO), optical-component layer (ALAB / LPTH / TSEM / SIVEF / AXTI), and generation-side delivery (independent grid-stress data from ChurchillWw). The R23-G CONTESTED designation on optical-supercycle is consistent — pricing-in is what late-stage cohort behavior looks like — but the breadth of layers participating is the new signal.
  • Layer 1 — chip-IP royalty (ARM the named example, R26-C origination):
    • 2026-05-20 sell-side wave around ARM — Bernstein $300 PT, RBC $260 (from $175), Jefferies $290, TD Cowen "secular demand inflection." ARM +39.49% 7d / +147.62% 3m (cited R26-C; verified at extension time).
    • Reads as the AGI-CPU royalty leg of the AI-capex thesis being priced. NOT a new thesis — the bottleneck framing has always implicitly priced chip-IP exposure; this is the equity tape catching up.
    • ARM coverage state: not in any AI-power-anchored watchlist; reachable only via the general semis cohort. Coverage-gap candidate flagged in R26-C; this entry confirms the gap without adding ARM to key_tickers (royalty layer is adjacent, not core; the bottleneck thesis is generation-side, chip-IP is a beneficiary not a load-bearing leg).
  • Layer 2 — AI-networking equipment (R27-E coverage extension, fresh 5/22 numbers):
    • Equipment cohort fresh-5/22 RSI/3m: EXTR RSI 72 / +85.78% 3m; FFIV RSI 82 / +46.05% 3m; ATEN RSI 69 / +54.29% 3m; ERIC RSI 72 / +23.63% 3m; CRDO RSI 65 / +76.07% 3m; NOK RSI 72 / +104.36% 3m; NBIS RSI 63 / +113.47% 3m.
    • Incumbent-lag pattern (R28-E framing, R29-C nuanced): ANET prints RSI 54 / +20.87% 3m / -10.73% 30d — materially lagging the equipment cohort it sits in. R29-C corrected: the lag is 30d/3m, not 7d (ANET participated in the most recent weekly leg at +10.25%). Reads as either (A) "safe carry with less to give back when cohort cools" or (B) "incumbent disruption — challengers take share." Both readings are live; the next earnings cycle will discriminate.
    • These tickers are already in ai-infrastructure.json per R27-E publish package — coverage is current; no key_tickers edit needed on this perspective. The equipment-layer cohort sits one tier above optical-component layer (which optical-supercycle covers); the bottleneck thesis benefits from both layers but doesn't need to absorb them as key_tickers.
  • Layer 3 — optical-component cohort (R26-A non-cybersec RSI-extreme cluster + R29-C corrected 7d):
    • Fresh-5/22 numbers (from optical-supply-chain R30-A precomputed): ALAB RSI 79 / +42.58% 7d / +55.35% 30d / +139.3% 3m; LPTH RSI 68 / +55.50% 7d / +51.34% 3m; TSEM RSI 68 / +13.12% 7d / +122.42% 3m; SIVEF RSI 69 / +137.82% 30d / +372.61% 3m; AXTI RSI 68 / +87.10% 30d / +395.36% 3m; CRDO (in ai-infra) +38.25% 7d.
    • These corroborate the R31-A optical-supercycle README cohort-state edit (Wave 1 contested / late-stage at positioning layer; Wave 3 momentum layer extended hard).
  • Layer 4 — generation-side grid-stress (R30-B ChurchillWw 2× data points, primary-source class):
    • ChurchillWw 2026-05-22 tweets cite PJM peak-demand stress + 2025 coal-retirement slowdown (only 2.6 GW of 8.0 GW planned retirements actually executed). Source class: substantive-primary-source per memory (reference_churchillww_candidate_follow), NOT a recycler / pumper / signals-shop.
    • Material for this perspective specifically: the thesis says "you cannot add 500MW to the grid in 6 months" — coal-retirement slowdown is the inverse of the same physics. If coal is staying online longer than planned, that's incremental supply that wasn't in the 2026 deficit math. Does NOT invalidate the thesis (the deficit is still real and growing with AI capex) but adds friction to the rate of de-rating, particularly for IPP-merchant pricing (CEG / VST / TLN) — a partial soft-floor for the IPP leg's recent weakness.
    • PJM peak-demand stress data point confirms the demand side of the same physics (load is growing faster than firm capacity additions).
    • Reads as primary-source confirmation that the perspective's macro-physics frame is still the working frame, just with a slightly slower de-rating cadence than a clean "rolling outages this summer" scenario would imply.
  • Effect on thesis (across all 4 layers): posture unchanged from R20 cohort-split frame — IPP leg weakening (CEG/VST/TLN regulatory + valuation drag, R30-A no new tape change), equipment leg holding (BE 30d +27% strong; GEV carries), AI/HPC-pivot miners sub-leg (R31-A absorb-fold) stabilizing-but-stock-pickers-market, chip-IP royalty pricing-in confirmed, AI-networking equipment pricing-in confirmed late-stage, optical-component cohort pricing-in confirmed momentum-extended, generation-side gating-physics-intact-with-friction. The thesis hasn't changed — the tape is hardening it across the full vertical at once.
  • No key_tickers edit: ARM (royalty adjacency, not load-bearing) + EXTR/FFIV/ATEN/ERIC/NOK (covered via ai-infrastructure.json) + ALAB/TSEM/SIVEF/AXTI (covered via optical-supply-chain.json + optical-supercycle perspective). The cohort coverage is already correct; this entry is the cross-layer pricing-in observation, not a coverage gap.
  • Iran-deal continuity carry (R29-C → R30-B): TheShortBear 2026-05-22 "Iran no deal is close" (R29-C source) was REVERSED by same author's 2026-05-23 "BREAKING: US and Iran are closing in on a deal..." (4× same-day tweets, 8989 ❤; R30-B capture). R27-D airlines combine REINFORCED at the source-pivot level; R29-C's combine-with-contradiction is now contradicted by its own source. Tape-test trigger queued to R31-D for post-Memorial-Day airlines + Brent adjudication (5/27-5/29 window). NO us-energy-dominance / war-ends-playbook log entry from R31-B (cross-perspective writes are out of scope here); the tape-test row in SCANS will adjudicate before the next us-energy-dominance touch.
6 events
DateTypeEvent
May 23 ArticleWeird-ETF rotation read — MSTR-leverage breakdown, AI-stack narrowing, defensives flat May 25 Deep DiveARM May 25 InvestigationOptical supercycle WHY: what specifically drove the CIEN/COHR demand acceleration May 26 Deep DiveBAM May 26 InvestigationCPU Shortage Thesis — First-Principles Derivation May 26 InvestigationSpaceX IPO Deep Dive: Separate Trades + the Oxygen Question

No direct external sources are attached to this read.