2026-W19-tuesday-reads
Tuesday reading log — AVIS short squeeze, AI value capture shifts to the labs, Big Tech earnings
2026-05-04 → 2026-05-10
Tuesday May 5 reading: Matt Levine on the Pentwater/SRS Avis short squeeze and Section 16(b) matching, SemiAnalysis on value migrating from infrastructure to model labs, TBPN tech-earnings quad kill, Stratechery on Amazon, Google and Meta, plus a few LATAM-payments and Stripe-crypto tweets worth keeping.
TL;DR · the week in 10 points
2 items
May 5Bloomberg
Money Stuff (Matt Levine): the Avis short squeeze — Pentwater + SRS owned 108% of the float via swaps, Section 16(b) caught a different fund
- Avis: $99 (Mar 20) → $713 (Apr 21) → $182. Pentwater + SRS at 108% economic ownership via cash-settled swaps.
- Total return swaps post-Archegos: still off-balance-sheet for the fund, banks hedge by buying the shares themselves.
- Section 16(b) matches lowest-buy with highest-sell — statutory profit decouples from actual P&L.
- Pentwater dodged 16(b) on a parallel fund; sold 4.3M shares for $1.75B in proceeds.
- The arb idea: short interest >30% + creeping past the 16(b) disclosure threshold = self-reinforcing squeeze, no dilution to existing holders.
May 5SemiAnalysis
SemiAnalysis: AI Value Capture — the shift to model labs
- Anthropic ARR $9B → $44B+; inference GM 38% → 70%+. Labs are now the value capture layer.
- Opus blended price ~$0.99/MTok on agentic workloads (300:1 input:output, 90%+ cache hits).
- GB300 → VR NVL72 capex/W: $37.4 → $38.1 — flat despite 2x perf/W. Nvidia is intentionally under-pricing.
- SOCAMM2 = socketed, repriceable memory at ~60% GM. Hidden pricing lever for Rubin.
- Two moats keep lab margins: real frontier quality gap (Kimi K2.6 at $0.95/$4 barely dents Opus) + structural compute scarcity.
Timeline · commits + curated notes
7 entries
- noteTuesday reading log — Avis short squeeze (Levine), SemiAnalysis AI value-capture flows, TBPN earnings quad-kill, two Stratechery posts, LATAM-payments and Stripe-crypto tweets.