Stock-Market Research Dispatch — 2026-05-20

Auto-generated by daily research routine. Not a source; not ingest-pending material. Summary of what changed in the wiki today.

Macro context

The two themes of today's run: (1) convergent power infrastructure bottleneck — $1.4T utility capex (+27% YoY) confirms the data center electrical picks-and-shovels thesis as structural, not transient; (2) TSMC sets a ceiling on AI chip output — Q2 guide +32% YoY, CoWoS sold through 2027, 5-10% price hikes across N2/A16 confirm multi-year pricing power. Both themes point to the same investment observation: AI demand is demand-constrained at the top (hyperscaler capex) and supply-constrained at the bottom (chips + electrical infrastructure), with the bottlenecks operating on different timescales (chip packaging: 2–3 years; grid transformer: 5 years).

SpaceX S-1 reportedly dropped today (May 20). Not researched in this run — pricing expected June 11, listing June 12 per yesterday's dispatch. Watch for any index pre-announcement.

New wiki pages

Significant wiki updates

FCX (copper-supercycle-ai-data-centers, fcx): Q1 actual unit cost $1.91/lb — a ~$0.69/lb beat vs the $2.60/lb guidance driven by gold by-product credits and favorable Morenci ore sequencing. Prior wiki had the guidance figure as the expected cost; this is a meaningful read-through: FCX's gold credits are a real-time buffer when gold is elevated, making management guidance a conservative ceiling. FY2026 cut 300M lbs (3.08B total); Grasberg restart costs $60-70M incremental; full capacity mid-2027. LME copper $13,380/t (May 18). Goldman forecasts 490K surplus; ICSG sees 150K deficit — 640K divergence remains the key uncertainty. Calibration entry logged on anchoring to guidance (see below).

TSMC (tsmc, tsmc-capacity-shortfall-and-pricing-power, cowos-packaging-capacity-crunch): Q1 2026 results beat — $35.9B (+6.4% QoQ), above guidance. Q2 guided $39.0-40.2B (+32% YoY at midpoint). Full-year raised to >30% USD revenue growth. N2 in HVM since Q4 2025 with "good yield"; Apple holds ~50%+ of 2026-2027 N2 allocation. N2 wafer ~$30,000 (step-function from N3 ~$20,000); A16 rumored ~$45,000 (+50% vs N2). 5-10% price hikes across advanced nodes. CoWoS fully sold through 2027; 50+ week lead time. TSMC AZ Fab 2 (3nm) accelerated to 2026-2027 (from 2027-2028) — increases US-fab capacity ahead of any Phase 2 tariff trigger.

Nuclear / Uranium (nuclear-baseload-for-ai-data-centers, ccj): Uranium spot $85.25/lb May 18 (+19.6% YoY) — corrected from stale $89/lb figure. Meta Vistra 2,609MW PPA: starts LATE 2026 — this is a near-term catalyst (12-18 months), not 3-5 years. CCJ Q1 2026: $845M revenue, $509M EBITDA, EPS $0.34 (beat $0.31); India deal signed March 2: 22M lbs over 9 years ($2.6B total), deliveries 2027-2035. AWS-Talen FTM 1.92GW operational spring 2026 (already online). CEG posted Q1 revenue +64% but guidance miss → stock -11.6% — the CEG -13% YTD is Calpine dilution fear, not nuclear thesis deterioration. SMR reality check: NuScale is the only NRC-certified SMR design; no SMR will be grid-connected before 2028-2030 at earliest. The nuclear-for-AI thesis is real but rides uranium/existing nuclear capacity, not SMR hype.

Intel 18A (intel-18a-yield-vs-tsmc): VLSI 2026 symposium (June 14-18) public tipsheet confirms Intel 18A session T1.2 specs: +9% performance / +18% efficiency vs Intel 3P baseline. TSMC A16 session T1.5 specs: +8-10% perf / 15-20% power reduction at iso-performance vs N2. Apple M7 production testing on 18A confirmed (previously only rumored). Intel 14A: two PDK customers; H2 2026 commitment window framing consistent with prior thesis. The VLSI data puts 18A and A16 in close range — no clear performance knockout punch for either side. Yield delta remains the discriminator.

Samsung Foundry (samsung, samsung-foundry-as-third-alternative): Critical distinction surfaced: SF2 (first-gen) yield ~55% (below production threshold, Qualcomm defection risk per TrendForce Apr 14) vs SF2P (second-gen) yield ~70% (Apple-grade, H2 2026 mass production). Q1 DS division profit +48x YoY (133.9T won total; 57.2T won op profit) — driven by HBM, not foundry. Foundry FY2026 goal: 2T won (currently loss-making). Qualcomm defection risk is the near-term watch item: if Qualcomm moves to TSMC N3/N2 for Snapdragon 8 Elite successor, Samsung loses its anchor non-Apple mobile customer in the transition window.

Energy / Grid (datacenter-construction-electrical-picks-shovels): $1.4T US utility capex (+27% YoY) confirmed from PowerLines April 14, 2026 analysis of 51 utilities serving 250M customers. Duke Energy $102.2B; Southern Company $81.2B. Data centers: 4% of US electricity (2023) → 9% by 2030. Grid transmission expansion materially lagging load growth; interconnection queues lengthening. ~$700B of the $1.4T flows to ratepayers via rate hikes. Transformer lead times 5 years; 25-50% of planned projects delayed. This substantially confirms the datacenter-construction-electrical-picks-shovels thesis. Added ETN, GEV, ABB alongside existing VRT, HUBB, PWR to the ticker watch list.

US Industrial Policy (section-232-phase-2-us-fab-premium): Phase 1 (Proclamation 11002, Jan 15, 2026): 25% on H200/MI325X-class chips; US data center, R&D, startup, gov, and non-DC consumer applications explicitly exempt. TSMC AZ chips exempt; TSMC Taiwan subject — structural demand pull for Arizona capacity already operative. Phase 2 "broader tariffs at significant rates" signaled post-negotiations; July 1 is the next hard gate (42 days). AZ Fab 2 (3nm) acceleration to 2026-2027 provides more US-fab capacity before Phase 2 potentially kicks in.

Calibration logged

FCX Q1 2026 unit cost: $2.60/lb guidance vs $1.91/lb actual (pattern: anchoring — accepted management guidance as expected-value without adjusting for by-product credit variability). See CALIBRATION.md entry #6.

Step 4b — Valuation snapshots

WebFetch blocked (HTTP 403) — price data unavailable in this environment. All active concept pages carry stale snapshot notes. Helium concept (helium-supply-crisis-semicap) now has a snapshot section added (LIN/APD, stale). Remaining active concepts with stale snapshots: copper, nuclear, CoWoS, picks-and-shovels, HBM, US critical minerals, US fab capacity, supply-shock inflation.

Open questions elevated this run

1. SF2 yield / Qualcomm defection — TrendForce Apr 14 flagged Qualcomm defection risk on SF2 below-threshold yield. Has Qualcomm made a commitment? Watch for Q2 Samsung foundry order commentary.
2. July 1 Phase 2 gate — 42 days. Does Commerce maintain the DC supply chain exemption in Phase 2? If narrowed, this reverses and HURTS US data center capex.
3. Meta Vistra 2.6GW start — "late 2026" per PPA terms. Watch for construction/commissioning updates.
4. Apple M7 Intel 18A volume commitment — production testing confirmed; final commitment (wafer order) expected H2 2026. No volume numbers yet.
5. Copper Goldman/ICSG divergence — 640K tonne spread between Goldman surplus and ICSG deficit forecasts. Next catalyst: ICSG semi-annual update.

Rotation notes

Step 1 topics today: FCX/Grasberg, Intel 18A/VLSI, nuclear/uranium/Meta PPA, Samsung foundry, TSMC capacity/pricing Step 2 macro buckets today: US industrial policy / chips / tariffs; Energy / grid / power demand Tomorrow's Step 1 candidates: SpaceX IPO (S-1 just dropped — index mechanics watch), HBM/Micron Q2 FY2026 verification (anomalous revenue figures flagged May 19 as unverified), Intel 18A VLSI full session results (conference is June 14-18 — pre-read next week) Tomorrow's Step 2 macro buckets: REE / critical minerals update, CSP capex cycle (any hyperscaler guidance revisions)