Arm Holdings (ARM)

CPU architecture/IP licensor — the instruction-set everyone (NVDA/AMD/Apple/hyperscalers) builds on

AI exposure 37% — derived from reported segment revenue, not asserted.

Where it sits

Silicon Layer → Semiconductor Devices · figures rebased to Q1 FY2027, reported 2026-07-29

Per-chip royalties on Arm-based processors (mobile, data-center CPUs, auto, IoT) leads 2 businesses in this sub-tier that together are 100% of revenue, so Semiconductor Devices is effectively the whole company rather than one leg of it.

Reported segments

• Royalty revenue — 55.5% of revenue
• Licensing & Other — 44.5% of revenue

Arm ($1,289M Q1 FY2027, +22% YoY). Royalty 55.5% ($715M, +22%) = primary B1 Semiconductor Devices — per-chip royalties on every Arm-based processor, with data-centre royalties more than doubling YoY on Neoverse adoption; Licensing & Other 44.5% ($574M, +23%, including $193M from SoftBank) — architecture and Compute Subsystem licences, which is where the AI-infrastructure design wins land first. IP licensor, not a manufacturer.

How the 37% is built

• Royalty revenue — 55.5% of revenue × 30% AI-driven (estimate). Royalties accrue across every Arm chip shipped; the data-centre share is rising fast but mobile still dominates the base
• Licensing & Other — 44.5% of revenue × 45% AI-driven (estimate). Licensing skews to Armv9 and Compute Subsystem deals, which is where AI infrastructure customers engage first

Investment read

Moat. Architecture lock-in — Arm ISA is embedded in virtually all mobile + increasingly data-center/custom silicon. Royalty model scales with the whole industry.

Bottleneck / pricing power. Toll-booth economics — a royalty on nearly every custom AI chip without making one, with Armv9 (~2x v8 rates) and CSS (~2x v9) inflating the take per chip contractually. The live constraint is now legal rather than competitive: the FTC opened a formal investigation in May 2026 into whether Arm degrades or denies architecture licences to advantage its own silicon, and Qualcomm's counter-suit is headed to trial.

Role in the AI stack. The architecture layer — the common ISA the custom-silicon wave is built on.

Connected companies

• Broadcom (AVGO) — ARM supplies: CPU IP licensed into custom accelerators (e.g. OpenAI/Broadcom CPU)
• NVIDIA (NVDA) — ARM supplies: CPU IP / architecture license
• Qualcomm (QCOM) — ARM supplies: core architecture license
• Advanced Micro Devices (AMD) — ARM supplies: architecture license
• Amazon (AMZN) — ARM supplies: Arm IP in AWS Graviton data-center CPUs
• Microsoft (MSFT) — ARM supplies: Arm Neoverse IP in Microsoft Cobalt 100/200 data-center CPUs
• Alphabet (GOOGL) — ARM supplies: Arm IP in Google Axion data-center CPUs
• Apple (AAPL) — ARM supplies: Arm ISA/architecture license for Apple Silicon (A/M-series) under agreement extending beyond 2040
• Meta Platforms (META) — ARM supplies: CPU IP for custom silicon for META AI data centers
• Oracle (ORCL) — ARM supplies: Arm CPU architecture underpinning ORCL's Ampere and Grace-based OCI fleet (licensed via Ampere/NVIDIA, not directly to ORCL)
• Synopsys (SNPS) — ARM is supplied by: EDA tools + IP co-optimization for Arm cores (Arm Total Design partner, 3-decade collaboration, Neoverse CSS)
• Cadence Design Systems (CDNS) — ARM is supplied by: EDA tools + design flow for Arm IP implementation/verification
• Intel (INTC) — ARM competes with: Arm architecture vs Intel x86 in data-center/PC CPUs

Others in Semiconductor Devices

• NVIDIA (NVDA)
• Advanced Micro Devices (AMD)
• Broadcom (AVGO)
• Qualcomm (QCOM)
• Marvell Technology (MRVL)
• Intel (INTC)

All 174 companies · What moved this quarter · How the map is built