Question: The @mollysoshea capture (2026-05-15) surfaced Coatue Public Investments CIO Jaimin Rangwalla's Spring 2026 framing on the Sourcery podcast. The CPU 2nd-source confirmation and 107%/4% YTD reproducibility test were already filed into research/investigations/2026-05-26-cpu-shortage-thesis-first-principles-derivation.md. Five named follow-ups remained: (1) pull the actual Coatue deck; (2) verify the divergence empirically; (3) decide hyperscaler-silicon disposition; (4) close the Samsung 005930.KS coverage gap; (5) wire the Coatue letter into the recurring-document watch.

Verdict: ratified — the Sourcery extract suffices as primary source; the deck PDF is auth-walled but identified; cohort empirics reproduce with Samsung now fetched; hyperscaler-silicon stays a sub-thesis annotation on ai-power-bottleneck rather than a new perspective; Samsung 005930.KS is now in memory.json; cadence wired for two future Coatue letters.

What we're asking

The Coatue Spring 2026 letter ratifies three of our active perspectives in one institutional document. Coatue is a $70B AUM crossover fund with deployed capital behind the thesis. Five concrete follow-ups, scoped to producer-only work — pull the source, run the math, make the coverage decisions, wire the next-firing dates. Not a publish pass; Codex reviews.

Source pointers (carried from inputs/tweets/2026-05-15-mollysoshea-coatue-rangwalla-sellers-buyers-of-shortage.md):

• Sourcery newsletter extract: https://www.sourcery.vc/p/exclusive-coatues-public-investments
• Podcast (Coatue mirror): https://www.coatue.com/blog/podcast/sourcery-podcast-20260515
• Coatue landing page: https://www.coatue.com/blog/perspective/public-markets-update-2026-05-06

What we found

(1) Coatue Spring 2026 slide deck — identified, auth-walled

The Coatue landing page (/blog/perspective/public-markets-update-2026-05-06) is the official housing for what Rangwalla called the "Spring 2026 Public Markets Update." Page body is a thin wrapper; the deck PDF link is:

• Deck (Google Drive): https://drive.google.com/file/d/1Zp_KoGg0lIB2MpcB_labOI12dFc1Kfu4/view

The PDF requires owner permission ("Sorry, the owner hasn't given you permission to download this file"). Coatue does not surface a public PDF on coatue.com. The Sourcery podcast extract therefore remains the de-facto primary source for our system; the slide deck pull will resolve only via (a) a Coatue contact, (b) an investor with the deck, or (c) Coatue posting it publicly. Status: identified-but-locked. Treat the Sourcery extract as primary until the deck unlocks; cite both URLs.

(2) Empirical reproducibility — Samsung now closes the cohort

The 107%/4% YTD divergence was already verified in 2026-05-26-cpu-shortage-thesis-first-principles-derivation.md for 12 of 13 named "sellers of shortage." The 13th name — Samsung Electronics 005930.KS — was a coverage gap (no 12m OHLC in our system). 12m fetched 2026-05-27: 242 entries, 52wk high 302,000 KRW, last close 299,000 KRW (2026-05-26).

Slots above the seller-cohort trimmed mean. The 12-name trimmed mean (drop SNDK outlier) was +109.6%; adding Samsung at +149.9% pulls the trimmed mean to +113.0% — still within rounding of Coatue's "~107%" claim, now with the full 13-name cohort priced. The reproducibility test does not break; it sharpens. Samsung is a clean memory-supercycle name (HBM3/HBM3E share with SK Hynix + Micron) and tracks the memory-cohort directionally (MU +163%, WDC +181%, STX +196%, SK Hynix +145%).

(3) Hyperscaler-silicon disposition — sub-thesis annotation, not new perspective

Rangwalla's "Amazon and Google are in an unusual camp" framing (because they own Trainium / TPU and capture some shortage premium internally) reproduces empirically:

Spread within the buyer basket: ~26 pp.

Decision: annotate ai-power-bottleneck as a sub-thesis layer, do not fork a new perspective. Reasons:

1. Single-quarter divergence is not yet a multi-source thesis. One quarter of YTD data plus Coatue's framing is two data points. Forking a perspective on 2 points over-fits; annotating preserves the cohort for multi-quarter sharpening.
2. The mechanism nests under the existing perspective frame. ai-power-bottleneck already tracks "where does the supplier-bottleneck premium accrue." The hyperscaler-silicon-capture wrinkle is a refinement of that question on the buyer side, not a distinct frame.
3. Cohort consolidation. AMZN / GOOG / MSFT already sit in mag7 + cloud watchlists; adding a third "hyperscaler-silicon" watchlist creates routing ambiguity for the next agent who touches the names.
4. Promotion gate is identifiable. If Q2/Q3 2026 hyperscaler earnings show MSFT Cobalt/Maia revenue ramp (currently early-stage) closing the −7.1% gap, or if Coatue Summer 2026 update names a 3rd silicon-owner, the sub-thesis converts to a standalone perspective. Until then, annotation is the correct shape.

Implementation: log-entry on ai-power-bottleneck (filed alongside this artifact) records the sub-thesis layer + 18.9% / −7.1% empirical anchor + promotion gate.

(4) Samsung 005930.KS — coverage gap closed

• Security family (stonks/watchlists/_securities.json): existing samsung-electronics entry updated; obsolete "scanner/watchlist coverage currently uses MU, HY9H.F, STX, WDC, and DRAM instead" note replaced with Coatue Spring 2026 ratification + empirical YTD cite. trade_access stays research_only (KRX requires Korean broker); EWY remains paper-trade proxy.
• Watchlist (stonks/watchlists/memory.json): 005930.KS appended to symbols (now 7 names: MU, STX, WDC, DRAM, HY9H.F, 005930.KS, EWY). Description block updated with the Coatue + reproducibility-test context.
• Data (stonks/data/stocks/005930.KS/ohlc/12m.json): 242 entries fetched 2026-05-27 via deno task fetch 005930.KS 12m. The 30d cache that already existed is now backstopped by full 12m.

Convention: this follows the BANB.SW / AIXA.DE / HY9H.F pattern — Korean primary kept as primary_listing, no US OTC proxy added because SSNLF is too illiquid to trade reliably.

(5) Recurring-document watch — Coatue cadence wired

Two calendar firings added to stonks/config/cadence.json:

Both ref this investigation. Dates align with the post-Q2-13F-deadline and post-Q3-13F-deadline rhythm — Coatue tends to publish public-markets updates shortly after their own 13F filings reveal positioning. Dates are approximate; cadence:check will surface them as upcoming and we adjust when Coatue actually publishes.

Note: The historical record for Coatue Public Markets Updates is uneven (they don't publish on a strict schedule). If we miss a firing because the actual letter drops in October or December instead of November, that's expected — the calendar serves as a trigger to check coatue.com/insights, not as a hard deadline.

Bonus carry-throughs

OpenAI + Anthropic ARR trajectory ($3.4B → $55B over 24 months, $2.5B/week velocity at the cadence Rangwalla named): logged in research/threads/2026-05-27-anthropic-ipo-compute-financing.md:62-63 as primary-institutional ratification. No further action — the thread is the right home.

TASKS-RESEARCH.md:109 (Anthropic + OpenAI S-1 filing watch) — Rangwalla's ARR cadence framing is a substantive datapoint when those S-1s drop. Cross-link in this artifact's affects; do not promote the S-1 watch itself (filing date unknown, trigger-driven).

Verdict + reasoning

Ratified — Coatue institutional ratification of three perspectives is durable; five follow-ups closed.

What this changes:

• ai-power-bottleneck, optical-supercycle, memory-supercycle each have a 2nd or 3rd independent institutional primary source.
• memory-supercycle cohort coverage is complete with Samsung now priced.
• Hyperscaler-silicon-capture is now a tracked sub-thesis on ai-power-bottleneck with a clean empirical anchor (+18.9% vs −7.1% YTD) and a promotion gate (Q2/Q3 2026 hyperscaler-silicon revenue progress).
• Future Coatue letters will surface via cadence:check rather than ambient capture.

What this does NOT change:

• We still cannot read the actual Coatue Spring 2026 deck; the Sourcery extract is the operative primary source.
• The CPU shortage 1:16 → 1:4 thesis stays gated on the substrate cohort coverage build-out — Coatue ratifies the head-CPU layer (INTC/AMD/ARM), not the substrate layer.
• Trade access on Samsung stays research_only; paper-trade exposure to the Samsung thesis runs through EWY (already in the watchlist) and the rest of the memory cohort.

Follow-ups

• ai-power-bottleneck log-entry filed alongside this artifact for the hyperscaler-silicon sub-thesis layer + 18.9% / −7.1% empirical anchor + promotion gate.
• Coatue deck pull stays open as a low-priority — surfaces only if a contact or investor shares the file; otherwise the Sourcery extract is sufficient.
• 2026-08-15 cadence firing will prompt a check at coatue.com/insights for the actual Summer 2026 letter; refresh this artifact's source-pointer block when it drops.
• Hyperscaler-silicon promotion gate monitored on Q2/Q3 2026 hyperscaler earnings — if MSFT Cobalt/Maia revenue ramps measurably or a 3rd silicon-owner emerges, convert annotation to standalone perspective.

Sources

• Capture: inputs/tweets/2026-05-15-mollysoshea-coatue-rangwalla-sellers-buyers-of-shortage.md
• Prior integration: research/investigations/2026-05-26-cpu-shortage-thesis-first-principles-derivation.md (Coatue 2nd-source block + 107%/4% reproducibility test for the 12-name cohort)
• Sourcery extract (primary): https://www.sourcery.vc/p/exclusive-coatues-public-investments
• Podcast (Coatue mirror): https://www.coatue.com/blog/podcast/sourcery-podcast-20260515
• Coatue deck landing: https://www.coatue.com/blog/perspective/public-markets-update-2026-05-06 (deck PDF auth-walled at https://drive.google.com/file/d/1Zp_KoGg0lIB2MpcB_labOI12dFc1Kfu4/view)