# Special Situations Report — Weekly Call (readable notes)

- **Title:** Earnings season (banks/ASML/TSM beat but tape won't rally), soft CPI/PPI (crypto + gold bounce), the Iran war into day 8 (Strait of Hormuz closed again, WTI ~$84 / Brent >$90), the RWTS 10% Redwood baby bond income add, the featured Nebius (NBIS) NeoCloud deep-dive (net-debt-neutral + ~$8B of private stakes), a VISN expert-call update (customer concentration found, upside trimmed), the merger-arb sweep (ATAI/Eli Lilly, PYPL/Stripe-Advent topping-bid, Lionsgate rumor, WBD antitrust suit), the historic momentum crash (Morgan Stanley Momo −28%), the "why is the VIX low" dispersion piece, the open-source-AI / hyperscaler-chip-demand thesis (Jevons's paradox), the Morgan-Stanley-memory-peak-vs-bull debate, and single-name earnings (IBM miss, UNH/ABT beats, TSM/ASML sell the beat, SNDK/MU short-term buys, NFLX −12.5%, LCID bankruptcy-halt refuted, DGXX NY moratorium)
- **Show:** Weekly SSR research call (premium subscriber recording — no public video)
- **Guest/host:** Jay Singh (founder, Special Situations Report; ex-Goldman Sachs)
- **Date:** 2026-JUL-19 (Sunday call; the deck/transcript title it "2026-07-19 - SSN")
- **Source URL:** the premium recording is distributed via the SSR Discord — [post ↗](https://discord.com/channels/1005949505671278603/1018706480700076073/1528617399463448779) (no public YouTube video)
- **Source PDFs (this folder):** `transcript.pdf` (premium transcript), `report.pdf` (the ~54-page weekly strategy deck), plus supplemental research Singh references: `Nebius NBIS - Tracking Model.xlsx`, `GS Summary 7-16.pdf`, `GS TMT TODAY 7.16.26.pdf`, `HSBC Monthly House Views.pdf`, `JPM Income Portfolio insights.pdf`
- **Note:** readable notes auto-extracted from the premium PDFs; fillers (um/uh/you know/false starts) removed, wording otherwise verbatim from the recording. Section headers carry the recording's `MM:SS` cue. No public video, so the per-name table has no deep-links (the Ref column is the section time as plain text). Auto-transcript garbles corrected to the right entity and listed here: **"SSN" = SSR** (Special Situations Report); **"atai Beckley" = atai Life Sciences (ATAI)** (post the Beckley Psytech combination), taken out by **Eli Lilly (LLY)** at $6.75 cash + up to $2.50 CVR; **Vison / Vistanc / VSEN = Vistanc Networks (VISN)** (ex-CommScope; VSEN "customers" Comcast/Charter are VISN's cable customers); **Nebius = NBIS** (its private stakes: **ClickHouse**, **AV-Ride/Avride**, **Toloka**, **TripleTen**; comps **DataBricks**, **CoreWeave (CRWV)**); **"AVT" beat = Abbott Labs (ABT)** (the transcript body discusses Abbott, not Avnet); **"DGFX/DGXX" = DGXX** (Digi Power X); **"Kimmy/Kimi K3" = Kimi (Moonshot AI)** Chinese open model; **"Momo index" = Morgan Stanley Momentum index**; **"Pumafish" = Google's Broadcom-made 2nm ASIC**; **"Vera Rubin / R200" = Nvidia's next GPU platform**; **"SATS ... now ECHO" = EchoStar (SATS)**; **"Fable 5" = an Anthropic model** cited in the cost-per-million-tokens chart; **"Stena/Stemma Research"** = a sell-side shop cited on the DeepSeek-2.0 "not so fast" piece (spelling uncertain); the transcript footer's "Date of Transcript: 2024-06-25" is a stale template artifact — the real call date is 2026-07-19.

---

2026-07-19 - SSR - Earnings, Soft CPI, Iran Attacks, RWTS, VISN Update, Merger Arb Update on PYPL/WBD

## Introduction and Strategy Document Overview - 00:00
- Sunday call; ~16-17 attachments incl. the 54-page strategy piece. Roadmap by page: ATAI (Lilly offer), PYPL (Stripe/Advent bid), Lionsgate sale rumor, WBD antitrust update, the RWTS 10% baby bond (Redwood common bought −42% below book / 17% yield, since +10%), the Nebius (NBIS) NeoCloud write-up (4 equity stakes worth >$8B), Dan Niles on AI over-investment (p11), Iran war day 8 (p13), why the VIX is low (p15), the momentum sell-off worsening (Momo −25%, p19), Korea's single-stock leverage-ETF suspension, the open-source AI era + hyperscaler G-spreads >150 (p23), soft CPI/PPI (p26), Morgan Stanley "memory pricing likely to peak" vs the opposing HF bull view, JPM Meta preview (p32), the TSM double beat (stock −high-single-digits), SpaceX unlock (p33), ASML guidance raise (p34), the IBM miss (p35), Shopify catalysts (p37), UNH + ABT beats (p38), DGXX NY data-center regulatory catalyst (p42), the Lucid (LCID) bankruptcy-halt refutation (p43), Argus cutting SanDisk to hold — but SNDK + MU "both short-term buys" (p45), Stripe/Advent topping-bid for PayPal (p46), the VISN expert call + customer concentration (p47), Netflix −12.5% ("could be a buy this year", p52).

## Economic Calendar - 05:07
- Light data week: leading index (~0), initial/continuing claims (23rd), S&P mfg + services PMI + new home sales (24th). Market pricing ~2-2.5% GDP for the year, below original expectations.
- Last week banks + ASML + TSM mostly beat, but the tape didn't rally (AI momentum fears + new Iran strikes).

## Earnings Calendar Review - 05:52
- Big two-week calendar. Held positions reporting: **ServiceNow (NOW)** and **Moody's (MCO)** on Jul 22; **CME** reporting but "we don't have a position" (stock decimated). Also flags Tesla, IBM (pre-announced, down most in years), Alphabet + Intel (Jul 23, the most important for the hyperscaler read), Raytheon, Comcast (related to VISN) and Charter (a top VISN customer).

## Special Situations and M&A Updates - 08:21
- **ATAI** +33% Thu on a definitive agreement with **Eli Lilly**: $6.75 cash + up to $2.50 CVR. **Electronic Arts (EA)** moved higher on EU phase-one clearance of its take-private. Utility-deal housekeeping (NextEra/Dominion — "DNE" filings in VA/NC/SC, close H2 2027; Re/Max HSR; NSA/PSA vote → close ~Jul 22).
- **PayPal (PYPL)** +20% Wed: Stripe + PE firm Advent bid ~$60/share cash (28% premium; $53B / ~8× fwd EBITDA — cheap). Proposal made early May, no response yet; "there could be a topping bid... mid-60s." PYPL ~$6B FCF next year = a 10% yield to deal terms; 8× EBITDA is ~1-1.5× below its 3-yr average → topping-bid case. All-cash so it delists PayPal (not a Stripe back-door IPO). **Lionsgate (LION)** exploring a potential sale via the content-M&A wave.

## Warner Brothers Antitrust Lawsuit Details - 12:09
- **WBD** +2% Tue after 12 state AGs sued to block the Paramount deal — relief because the complaint held "very few surprises." Focus: a 30% top-grossing-film share (top-4 studios → 93% post-merger), worse theater splits (AMC), less output; more on cable-channel overlap, less on labor. At $27, WBD prices ~75% deal-completion odds (mid-teens standalone). Watch: EC phase-one (Jul 22), UK CMA (Aug 7, "most risky"), UK culture secretary intervention.

## Redwood (RWTS) Baby Bond - 15:02
- **RWTS** — the Redwood baby bond ~9.8% at ~$24.90, a new issue from end-May, still adding to the income portfolio ("10% is not bad for an unsecured bond in this environment"). Two ~60¢ dividends (payable Aug 14 + Nov 13) → ~$1.20/share if bought before the Aug 14 x-date. (The Redwood common — RWT — was the separate 42%-below-book / 17%-yield buy.)

## Nebius Deep Dive - 15:38
- **NBIS** — the featured operational build-up (GPU/revenue/Microsoft + Meta contracts; A100→GB200/Vera Rubin ramp). A NeoCloud "undervalued if you include the stakes." Formed from Yandex's divested international assets; **net-debt-neutral** (far less debt/EBITDA than **CoreWeave**), a pure-play GPU-as-a-service + sovereign-AI play (Amsterdam HQ, European alternative). Q1 rev jumped to $399M; de-risked via a $775M asset-backed facility tied to contracted customer cash flows (less equity dilution). "Venture company inside a large-cap wrapper" — optically expensive P/E but EBITDA +5×; targeting >4 GW contracted power by 2026; Vineland NJ build online Q4. Consensus buy $244-265 vs $177 close (+41%); beta >2. **Private stakes worth ~$8B:** ClickHouse (28%, $15B Jan-26 round → ~$4.2B), AV-Ride/Avride (83%, $4B → ~$3.3B), Toloka (50%, $500M → ~$250M), TripleTen (100%, ~$250-500M). ~$46B of signed Microsoft/Meta deals; Wolfe sees $34B rev / $21B EBITDA by 2030, consensus now ~$44B rev. Backed by Orbis, Allan Gray, Two Sigma. Micron's "supply-demand imbalance persists beyond 2027" and Qualcomm's "1,000× tokens/query" support the compute-demand tailwind.

## A Brief Intermission - 34:24
- (water break.)

## Dan Niles on AI Overinvestment - 35:46
- Every industrial revolution over-invests → eventual bust, but Niles thinks the current "speed bump" is not the start of the bust this year. Near-term AI-speed-bump drivers: token minimization (Coinbase cut token spend ~50% routing to cheaper models), cheaper Chinese models, open-source LLMs, rising semi costs, Q3 impact. **Google** is Niles' favorite consumer AI play (complete AI stack; GCP growth accelerating 34%→63%) but not immune to the top-1% cutting AI bills — reports Wed, market will focus on margins. **Intel (INTC)** (reports Thu): Niles more bullish on the AI-infra layer than the commoditized LLM layer — server-CPU bottleneck for agentic AI, advanced packaging attracting hyperscalers, a US-backed foundry. The Momo index fell 28% Jun 22→Jul 16 (24 days); >10% corrections since 1996 average −20% over 48 days, but the prior +40% Mar 30→Jun 22 rally was also unusually sharp.

## Market Update and Geopolitical Situation in Iran - 40:13
- Futures bounced intraday (Nasdaq +0.53%). Momo not fully oversold (RSI 35 vs 31 avg bottom) but risk/reward better after the IBM −26%, ASML −3%, TSM −8% and semi/momentum −10% washouts. Niles: start adding back AI-infra exposure (agentic AI needs 10-100× more tokens), balanced with caution at the public-cloud layer.
- **Iran war day 8:** the US proposed then canceled its own (higher) Strait-of-Hormuz tolls; 6 days of US strikes → 8 by Sunday, IRGC infrastructure hit (6 bridges, desalination plants, power stations); Iran hit the US airbase in Kuwait. Brent → ~$86→>$90, WTI → ~$81→~$84. Blockade reinstated ("guardian of the Hormuz Strait"), reviving inflation risk just as CPI/PPI abated. UAE/DP World planning a Fujairah port to bypass the strait; Rome peace talks (Jul 15) "not fruitful."

## VIX Analysis: Why It's Low Despite Geopolitical Risk - 45:03 / Final Thoughts on the VIX and Falsifiable Edges - 56:15
- The framework: the low VIX is not complacency but **Bayesian updating** — through three "natural experiments" (war onset Feb 28, war end, and the Jul 6 restart) the market learned this war's risk lives in **oil, rates and single-name vol, not broad index correlation**. Round 1 priced blind (VIX to 31, OVX to 126); the restart prices "with a map" (VIX capped 15-17.5). Evidence: OVX repriced ~39%, WTI +9.4% single-day, 2-yr yield to a 15-month high ~4.23% (hike risk, not cut), Nasdaq-100 vol 27.3 vs VIX 16, one-month implied **correlation at a two-year low** (the dispersion mechanism as a number), a steep contango (1-day <10, 3-month 18.57) and a pinned skew 144-146 (bid tails). Falsifiable edge: the risk is **oil to ~$110** with the 2-yr breaking out forcing the Fed hold→hike — that flips a dispersion story into a correlation story (a sustained VIX >25 via oil-inflation would break the framework). "Watch the two-year yield, not the tanker count."

## The Momentum Sell-Off and South Korea's ETF Suspension - 60:31
- Historic July momentum crash: MS Momo −28%, Goldman high-beta momentum −24% MTD (worst since Apr 2009), MS tech-momentum 17-day RoC −35% (worst in its 27-yr history). Korea suspended new single-stock leverage-ETF listings + tripled the minimum cash deposit (10M→30M won) → KOSPI VIX to 81 (2008 levels). **SK Hynix** −40% from its ATH; Korea's PE ~5.89× (earnings feared over-inflated).

## Open Source AI Era and Hyperscaler Chip Demand - 62:51 / The Arithmetic of Frontier Margins - 01:08:44 / Impact of Open Models on Hyperscaler Revenue - 01:12:39
- Gavin Baker's mega-bull case: if share shifts from 90%-margin frontier labs to cheaper models, compute demand rises (intelligence-per-dollar up) and margin dollars redistribute to AI-infra providers (NeoClouds like Nebius) and the lowest-per-token-cost winners. OpenRouter data: calls to sub-$1/M-token models 18%→41% (Jan→Jun 2026); Chinese open-model token share <2%→~61%; all open-weight ~69%. Frontier margins: Anthropic gross margin −94% (2024) → mid-60s% (2026), ~36% pre-training; OpenAI adjusted GM 40%→33%. As of late-May Anthropic took 42% of OpenRouter model spend on 11% of tokens — a 25× price gap paid for reliability. The risk to frontier is losing **mid-tier volume**, not margin rate. **Hyperscaler math:** the open-model switch turns a pass-through into a direct sale; a **32% reinvestment (reuse) rate is hyperscaler breakeven**, and real budgets grow 5-7×/yr → hyperscalers earn *more*, so they don't cut compute/memory. (This is a **Jevons's paradox** argument — cheaper compute → more total demand.)

## Soft CPI/PPI Data and Federal Reserve Commentary - 01:21:01 / Geopolitical Impact - 01:23:21
- Soft June CPI + PPI left rates lower — positive for crypto (brief bounce) and gold. Base case: Fed **on hold** (neither hike nor cut) = best for risk near-term; hike only if inflation runs hot. Energy is the wild card; if Brent stays <90 and there's another peace treaty the market can rally (>100 = "shit territory"). **Warsh** testified (less hawkish, low guidance); **Waller** hinted willingness to hike Jul 29 → July-hike odds spiked to 45% intraday before CPI/PPI pulled them back.

## Gold's Performance / Market Sentiment / Wrap-Up - 01:27:10 → 01:29:47
- Gold the worst asset of the last couple months as **real (TIPS) yields** rose; still +19% YoY; fell below $4,000 after a >$5,400 Jan spike. "Gold is suffering from a surplus of Fed credibility." Equities didn't rally on cooler inflation — weakness is the AI-infra-trade unwind.

## HBM Memory Pricing and Its Impact on Chipmakers - 01:30:31
- **Morgan Stanley: memory pricing likely to peak**; the opposing bull view from **BlueFin Research Partners**: HBM4 pricing could double/triple in 2027 (SK Hynix's HBM4-ramp delay a "veiled threat" to Nvidia on margins). HBM4 content: Nvidia R200 288GB (~$5k COGS → $19k→$38k/$58k if 2×/3×; VR200 rack ~$8M, >$9M if HBM4 3×), AMD MI450 432GB, Google Pumafish 576GB (shelving possibly HBM4-cost-driven).

## Meta Pre-Earnings / TSM / SpaceX / ASML / IBM - 01:33:26 → 01:37:52
- **META** — JPM preview; risk = monetizing the huge capex + showing *higher* capex again (stock +20% in two weeks); hyperscaler 12-mo FCF (ORCL/META/GOOGL/MSFT/AMZN) gone from $300B → ~negative.
- **TSM** — double beat (EPS 437 vs 337, GM 67.7%), but ADRs fell on capex raised 60→64B; stock down high-single-digits for the week; Taiwan Taiex into technical correction.
- **SpaceX (SPCX)** — market preparing for the August unlock; the trade is **short SpaceX vs long SATS (now "ECHO")** — SpaceX fell more, so the spread should compress.
- **ASML** — great numbers (net sales 43-45B vs 39B est, GM 55-57%; lifted the annual forecast a 2nd time, +30% NA EUV capacity) but the stock sold off — "the market's very skeptical of these names."
- **IBM** — massive miss; discretionary IT spend worsening (customers shifting spend to servers/storage/memory to hedge price increases; the theme for software this season); "slow to innovate"; stock −20%+ pre-market. Oracle's separate problem = too much OpenAI exposure.

## Shopify / UNH / Abbott / Crypto / Precious Metals - 01:39:38 → 01:45:27
- **SHOP** — catalyst update (exponential AI traffic, higher-margin financial scaling, up-market enterprise adoption) — a follow-up to prior work.
- **UNH** — beat + raised outlook (EPS ~$19.5-20).
- **ABT** (transcript's "AVT") — Abbott Labs +11% on strong results (rev +13% YoY to $12.6B) across devices/diagnostics/pharma/nutrition.
- Crypto rallied briefly (Fed-hike odds → ~8%). **GLD** saw a $14B outflow since March, but central banks + retail are restarting gold buys under $4,000 ("buy it under 4,000, not at 5,600"). Silver −20-21% YTD vs gold −7% (real-yield + industrial exposure). US 10-yr 4.57% → new mortgage-rate highs (6.55%, highest since Aug 2025).

## DGXX / Lucid / ASML forecast / PayPal / VISN - 01:47:05 → 01:53:46
- **DGXX** — negative NY catalyst: Gov. Hochul's 1-yr freeze on new hyperscale (50MW+) data-center air permits; DGXX has ~140MW in-state but is **grandfathered** ("carved out"; upgrading existing crypto-mining to tier-3 AI). Local North Tonawanda moratorium (covers expansions) is the more relevant risk, but existing ops continue + a hardship-exception path exists — "not a clean all-clear, but not an outright ban." 140MW of controllable power = a strategic asset.
- **LCID** — Lucid refuted its bankruptcy halt (liquidity to next year); "degens" bought it from ~$270-280 back to ~$736.
- **PYPL** — Stripe/Advent bid $60.50 (28% premium; $50B bank financing, highly levered); PayPal cutting ~20% of staff; owner thinks it's worth ~$70.
- **VISN** — **expert-call update** (Carlos Treves, VP product/growth at Zayo): mapped all the asset sales → ~32% upside on the stub / ~19.5% on the stock (half the cap is cash; ~45% on the business value alone). Trades ~3-4× EBITDA vs a 6-7× fair; **but customer concentration is higher than hoped, so "not a tremendous amount of upside"** — still cheap to fundamentals; awaiting government deal approval.

## China AI Risk / Netflix / Europe - 01:55:30 → 01:57:13
- Stena/Stemma "China AI risk overblown" follow-up (cost-per-million-token chart: Fable 5 ~57.4 vs much-cheaper Chinese models). **NFLX** −12.5% Fri (most in 4 years) on slowing earnings despite heavy content spend + podcasts/gaming — "trying to do too much"; time-spent +2% Q1 — "but I think it'll be a buy again." Europe's falling US reliance / rising China reliance = dangerous for its manufacturing footprint.

## Q&A - 01:57:24 →
- **Nokia (NOK)** — "of all the AI names, probably the least interesting" / not first to rally; but optical + IP networks (~30% of rev) growing 10-12%, earnings +14% on Infinera synergies; JPM €18 target (29× — high); on a PEG basis it looks interesting → "one to do more work on." A momentum-reversion candidate.
- **Nebius vs CoreWeave** — prefers NBIS: **CRWV** "also likely a buy" but has ~$40B debt and burns cash / raising more debt; Nebius grows ~3-4× faster with no net debt.
- **Model reuse (frontier vs open)** — the reuse/Jevons's-paradox explanation: a cheaper model frees budget; reinvesting ~⅓ of the savings gets the hyperscaler to breakeven, and enterprises grow budgets 5-7×/yr → more compute + memory demand. ("Bananas made of gold vs Chiquita bananas.")
- **Kimi's compute sourcing** — Kimi runs on Alibaba (**BABA**) for now (good for Alibaba demand); Microsoft repurposes Chinese open models on **US** compute; even Chinese hyperscalers buy Nvidia GPUs (routed via Singapore). "OpenAI is definitely losing market share, which is why I'm more worried about Oracle."
- **Defense / Joby (JOBY)** — likes the big defense companies (**LMT** etc.) and **AVAV** more than **JOBY** (JOBY burns ~¾B/yr, runs out of excess cash in 3 yrs, no positive EBITDA till past 2030).
- **Data-center delays** — ~70% of 2027 builds delayed (power/equipment) → the cycle extends into 2028 (higher build pace than 2027); over-ordering of memory/GPUs eventually shows up. Morgan Stanley's "memory peaks this year" is likely wrong because the data centers aren't even built yet. AI names "tradable from the sell-off" — a trading opportunity, not a buy-and-hold; needs rates to peak + the war to end.

## Closing Remarks / Final Market Outlook - 02:15:50 / 02:24:20
- Choppiness ahead (higher rates + oil); if the war goes away, optimistic — would add precious metals, a little crypto, AI names, income names ("adding in general"). Holds a lot of cash. "Patience is key." (Plus a World Cup aside — Spain over Argentina.)