ARM

Universal CPU IP underneath the AI compute stack:

- Every hyperscaler custom server CPU is Arm v9 - AWS Graviton4, Google Axion, Microsoft Cobalt 100, Meta MTIA host

- Every NVIDIA AI platform pairs a Grace CPU (Arm v9, Neoverse V2) with the accelerator - Grace+Blackwell shipping at scale, Rubin generation continues the pairing

- Apple silicon (M-series, A-series, S-series, R1) - 100% Arm v9 across the entire device stack

- v9 royalty rate is ~2× v8 → structural ASP expansion on every chip shipped, independent of unit growth

- Compute Subsystems (CSS) chiplet IP pushes ARM up the value chain from ISA licensor to full system reference design → royalty per AI accelerator rises

Bear:

- Qualcomm v9 licensing dispute pending (Oryon / Snapdragon X) - outcome materially affects PC + handset royalty per-chip economics

- SoftBank still owns ~88% of float post-IPO → structural overhang on multiple expansion

- Forward PE 98 (semis-forward-pe-2026-05-03) is the richest multiple on the board - every percentage point of v9 mix has to print or the multiple compresses