Arm Holdings
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Arm's 325 billion cumulative chip shipments have created the world's most extensive software ecosystem — every operating system, compiler, and AI framework is optimised for ARM architecture first, creating a self-reinforcing standard that 20 years of competing alternatives have failed to displace.
ARM's competitive position rests on three pillars: Ecosystem Network Effects, Talent-Based IP Advantage, and the v9 Royalty Ratchet:
- Ecosystem Network Effects — The Software Standard: With 325 billion cumulative chips shipped, ARM architecture has accumulated the largest software ecosystem in computing: iOS/Android/macOS/Windows all run on ARM-optimised binaries; AI frameworks (PyTorch, TensorFlow) prioritise ARM inference optimisation; cloud providers (AWS, Google, Microsoft) offer ARM instances with AWS Graviton, Axion, and Cobalt chips. RISC-V, despite being free and open-source, cannot replicate this ecosystem depth — the opportunity cost of migrating 325 billion device-years of software to a new ISA is insurmountably high.
- Armv9 — The Structural Royalty Ratchet: Arm's business model charges a percentage royalty on each chip that uses its architecture. Armv9 commands a materially higher royalty rate than v8 — and now represents over 50% of royalty revenue. Every v9 chip shipped for smartphones, data centers, or edge AI increases ARM's per-device economics without ARM having to acquire new customers. This is a structural royalty ratchet: as the industry upgrades to v9 and eventually v10, ARM's royalty per device grows automatically, making revenue growth partially independent of unit volume growth.
- Data Center AI Inflection: Data center royalty revenue more than doubled year-on-year in Q3 FY2026, driven by AWS Graviton4 (8th generation), Apple's M-series chips in Macs and iPad Pro, and Microsoft Azure Cobalt 100. As AI inference workloads require efficient compute, ARM's power-performance leadership over x86 makes ARM-based server chips the economically rational choice. The Total License Agreement (TLA) structure — where hyperscalers pay a large upfront fee for full IP access — is transforming ARM's revenue from lumpy license payments to more predictable, annuity-like streams.
Moat Verdict
ARM is a strong net beneficiary of AI — AI inference is the highest-growth compute workload, and ARM's power-efficiency advantage makes ARM-based chips the economically rational choice for inference at every scale from smartphone to data center. The Armv9 architecture's performance improvements for AI workloads (larger matrix math units, improved memory bandwidth) directly increase ARM's royalty rates as customers upgrade. The primary AI risk is that AI training (not inference) favors NVIDIA's CUDA ecosystem, which could push the AI compute center of gravity toward x86/NVIDIA architectures at the training layer — though inference at the edge and hyperscaler inference (AWS Graviton, Microsoft Cobalt) strongly favors ARM.
75.9 resilient · 69.7 vulnerable · 80/20 = 74.7 · = 75
Open a moat to read its note.
SoC architects and chip designers build on ARM's development environments (DS-5, Keil MDK), debug tools, and IP configuration interfaces for years. The familiarity with ARM Cortex-A, -M, and -R series architectures is embedded in engineering teams globally — switching to RISC-V requires retraining and re-tooling significant portions of chip design teams.
Chip designs built on ARM IP — including custom implementations by Apple (Firestorm, Avalanche cores) and Qualcomm (Oryon) — embed years of silicon engineering on ARM's architecture. The Total License Agreement structure means that licensees configure ARM IP into their custom designs, creating deep technical and contractual embeddedness.
ARM is an IP licensing company, not a data business. Public data access is not a meaningful moat dimension.
ARM employs 6,000+ engineers who design CPU microarchitectures at the cutting edge of performance-per-watt. This is among the most specialised engineering talent in the world — CPU architects who can design competitive Cortex-A cores are extremely scarce. RISC-V has struggled to match ARM's pace of innovation precisely because ARM's talent concentration is irreplicable.
Total License Agreements bundle CPU (Cortex), GPU (Mali), NPU, interconnect (AMBA), physical and security IP into one portfolio licence, so a TLA licensee rarely swaps out one block. The lock that matters is the Arm instruction set and its software ecosystem, already rated strong under network effects; no TLA attach or renewal figure is on file to prove the bundle adds to it, so it rates intact (one fact, one pillar).
ARM's decades of silicon performance, power, and area (PPA) data from hundreds of chip implementations represents a design knowledge base that enables ARM to stay at the frontier of performance-per-watt. This proprietary benchmarking and architectural data is not public and cannot be replicated without building hundreds of competing implementations.
Arm holds no government mandate, certification or procurement lock-in. Export controls and UK national-security oversight are regulatory risk (China licensing), charged in the growth keyRisk rather than scored here as an eroded moat. Previously weakened as 'a risk rather than a moat'.
The ARM software ecosystem is the largest in computing: 325B+ cumulative chips create a base of ARM-compiled software (iOS, Android, macOS, cloud VMs) that makes any new architecture uncompetitive by default. RISC-V has had 20 years and billions in backing but cannot match the software ecosystem — the network effect makes ARM the rational choice for any new chip design that must run existing software.
ARM is an IP licensing company. Transaction embedding does not apply.
ARM does not operate a system of record for any business function. Not applicable.
Fabless or equipment franchise whose edge is IP and process know-how, rated elsewhere; manufacturing volume does not give it a unit-cost lead rivals cannot buy.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Arm's 325 billion cumulative chip shipments have created the world's most extensive software ecosystem — every operating system, compiler, and AI framework is optimised for ARM architecture first, creating a self-reinforcing standard that 20 years of competing alternatives have failed to displace.
Growth Score
Q1 FY2027 (reported July 29, 2026) delivered another record quarter for a fiscal first quarter: revenue $1.289B (+22% YoY), royalty revenue $715M (+22%), licensing revenue $574M (+23%), and non-GAAP EPS $0.45 (+29%). Data-center royalties more than doubled again, Armv9/CSS adoption lifted royalty rates, and AGI CPU customer demand now exceeds $2B across FYE27-FYE28 versus the $1B opportunity outlined last quarter. Q2 guidance of $1.38B revenue and $0.47 non-GAAP EPS keeps the FY27 path around low-20s growth, but the stock has already repriced the AGI CPU optionality.
Valuation Score
At $271.87, ARM is almost exactly at the rebuilt base case ($275), after trading above the stale $220 bull case. The ladder is reset around the new reality: Q1 FY2027 proves the data-center and AGI CPU story is real, but a ~$286B market cap already prices a large share of that optionality. The stock is fair-to-slightly-rich rather than outright unscored upside; further gains require the >$2B AGI CPU demand signal to turn into shipped revenue without compressing the IP model's margin quality.
The Architectural Standard Moat
ARM's competitive position rests on three pillars: Ecosystem Network Effects, Talent-Based IP Advantage, and the v9 Royalty Ratchet:
- Ecosystem Network Effects — The Software Standard: With 325 billion cumulative chips shipped, ARM architecture has accumulated the largest software ecosystem in computing: iOS/Android/macOS/Windows all run on ARM-optimised binaries; AI frameworks (PyTorch, TensorFlow) prioritise ARM inference optimisation; cloud providers (AWS, Google, Microsoft) offer ARM instances with AWS Graviton, Axion, and Cobalt chips. RISC-V, despite being free and open-source, cannot replicate this ecosystem depth — the opportunity cost of migrating 325 billion device-years of software to a new ISA is insurmountably high.
- Armv9 — The Structural Royalty Ratchet: Arm's business model charges a percentage royalty on each chip that uses its architecture. Armv9 commands a materially higher royalty rate than v8 — and now represents over 50% of royalty revenue. Every v9 chip shipped for smartphones, data centers, or edge AI increases ARM's per-device economics without ARM having to acquire new customers. This is a structural royalty ratchet: as the industry upgrades to v9 and eventually v10, ARM's royalty per device grows automatically, making revenue growth partially independent of unit volume growth.
- Data Center AI Inflection: Data center royalty revenue more than doubled year-on-year in Q3 FY2026, driven by AWS Graviton4 (8th generation), Apple's M-series chips in Macs and iPad Pro, and Microsoft Azure Cobalt 100. As AI inference workloads require efficient compute, ARM's power-performance leadership over x86 makes ARM-based server chips the economically rational choice. The Total License Agreement (TLA) structure — where hyperscalers pay a large upfront fee for full IP access — is transforming ARM's revenue from lumpy license payments to more predictable, annuity-like streams.
Moat Verdict
ARM is a strong net beneficiary of AI — AI inference is the highest-growth compute workload, and ARM's power-efficiency advantage makes ARM-based chips the economically rational choice for inference at every scale from smartphone to data center. The Armv9 architecture's performance improvements for AI workloads (larger matrix math units, improved memory bandwidth) directly increase ARM's royalty rates as customers upgrade. The primary AI risk is that AI training (not inference) favors NVIDIA's CUDA ecosystem, which could push the AI compute center of gravity toward x86/NVIDIA architectures at the training layer — though inference at the edge and hyperscaler inference (AWS Graviton, Microsoft Cobalt) strongly favors ARM.
75.9 resilient · 69.7 vulnerable · 80/20 = 74.7 · = 75
Open a moat to read its note.
SoC architects and chip designers build on ARM's development environments (DS-5, Keil MDK), debug tools, and IP configuration interfaces for years. The familiarity with ARM Cortex-A, -M, and -R series architectures is embedded in engineering teams globally — switching to RISC-V requires retraining and re-tooling significant portions of chip design teams.
Chip designs built on ARM IP — including custom implementations by Apple (Firestorm, Avalanche cores) and Qualcomm (Oryon) — embed years of silicon engineering on ARM's architecture. The Total License Agreement structure means that licensees configure ARM IP into their custom designs, creating deep technical and contractual embeddedness.
ARM is an IP licensing company, not a data business. Public data access is not a meaningful moat dimension.
ARM employs 6,000+ engineers who design CPU microarchitectures at the cutting edge of performance-per-watt. This is among the most specialised engineering talent in the world — CPU architects who can design competitive Cortex-A cores are extremely scarce. RISC-V has struggled to match ARM's pace of innovation precisely because ARM's talent concentration is irreplicable.
Total License Agreements bundle CPU (Cortex), GPU (Mali), NPU, interconnect (AMBA), physical and security IP into one portfolio licence, so a TLA licensee rarely swaps out one block. The lock that matters is the Arm instruction set and its software ecosystem, already rated strong under network effects; no TLA attach or renewal figure is on file to prove the bundle adds to it, so it rates intact (one fact, one pillar).
ARM's decades of silicon performance, power, and area (PPA) data from hundreds of chip implementations represents a design knowledge base that enables ARM to stay at the frontier of performance-per-watt. This proprietary benchmarking and architectural data is not public and cannot be replicated without building hundreds of competing implementations.
Arm holds no government mandate, certification or procurement lock-in. Export controls and UK national-security oversight are regulatory risk (China licensing), charged in the growth keyRisk rather than scored here as an eroded moat. Previously weakened as 'a risk rather than a moat'.
The ARM software ecosystem is the largest in computing: 325B+ cumulative chips create a base of ARM-compiled software (iOS, Android, macOS, cloud VMs) that makes any new architecture uncompetitive by default. RISC-V has had 20 years and billions in backing but cannot match the software ecosystem — the network effect makes ARM the rational choice for any new chip design that must run existing software.
ARM is an IP licensing company. Transaction embedding does not apply.
ARM does not operate a system of record for any business function. Not applicable.
Fabless or equipment franchise whose edge is IP and process know-how, rated elsewhere; manufacturing volume does not give it a unit-cost lead rivals cannot buy.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Growth Analysis
Growth Drivers
Key Risk
If RISC-V or fully in-house hyperscaler CPU designs capture >15% of incremental data-center processor deployments by end-2027, while Arm's AGI CPU supply ramp stays below the >$2B demand signal, the data-center royalty growth trajectory slows materially and the market has to reprice ARM as an expensive IP licensor rather than a compute-platform company.
Score Derivation
85.3 base + 2.7 trajectory − 5 risk = 83
Base 85.3 (23% midpoint from 20–26% CAGR, centred on the measured FY2026 +24% / Q1 +22% / Q2 guide ~22% rather than above all three) + 2.7 trajectory (data center and AGI CPU accelerating; Armv9 royalty mix stable) + 0 stable margin − 5 moderate RISC-V / hyperscaler-in-house risk = 83
Price Scenarios (12–24 Months)
Valuation Multiples
| Trailing P/E (GAAP) | >250× |
| Forward P/E (NTM, non-GAAP) | ~115-125× |
| PEG Ratio | ~4.5× |
| Price / Sales (NTM) | ~48× |
| Price / FCF | ~200× |
ARM now trades at a software-like sales multiple and a triple-digit forward earnings multiple, reflecting scarcity value in the IP royalty model plus a new AGI CPU narrative. That is defensible only if data-center royalties keep doubling and the AGI CPU business scales without turning ARM into a lower-margin silicon vendor. The scenario reset raises fair value, but it does not make the stock cheap.
Approximate figures as of August 2026.
Where We Are vs Targets
Loading live price…
RISC-V gains material data-center traction, AGI CPU supply/margins disappoint, and the royalty ratchet slows — ARM is revalued as a premium IP licensor rather than an AI compute platform.
- RISC-V achieves 15% data-center processor share by end-2027 as hyperscalers and Chinese cloud providers push custom designs, slowing ARM data-center royalty growth from more than doubling to below 30%
- AGI CPU demand proves supply-constrained and lower-margin, converting less than $1B of the >$2B customer-demand signal by FY2028
- Royalty revenue falls back to low-teens growth after Q2 FY2027, and the multiple compresses toward ~30× FY2027 sales — roughly $160 per share on a ~$5.8-6.0B revenue base
ARM sustains low-20s FY2027 revenue growth, data-center royalties remain the fastest-growing royalty pool, and the AGI CPU converts enough of the >$2B demand signal to justify today's premium multiple.
- FY2027 revenue lands around $5.9-6.1B, with Q2 meeting the $1.38B midpoint guide and licensing growth near 30% YoY
- Data-center royalties more than double again in FY2027 while Armv9/CSS mix keeps blended royalty growth above unit growth
- AGI CPU shipments convert the original $1B opportunity across FYE27-FYE28 and preserve a path toward the larger >$2B demand pool
- The stock holds around the current high-40s sales multiple because investors view ARM as the default CPU platform for AI inference rather than a cyclical semiconductor supplier
The AGI CPU scales faster than management's initial capacity plan, AI inference makes ARM the dominant architecture in edge and cloud, and FY2028 revenue visibility moves above $8B without margin collapse.
- Arm secures manufacturing capacity for the full >$2B AGI CPU demand signal and demonstrates a path to 50% gross margin as more work moves in-house
- AI edge inference in smartphones, PCs, cars, and IoT creates a premium royalty category for NPU-integrated ARM chips at several times the rate of general-purpose cores
- Neoverse-based platforms keep gaining hyperscaler share, with AWS, Microsoft, Google, and Chinese cloud providers expanding ARM-based instances for inference workloads
- FY2028 revenue visibility moves above $8B and the market keeps a ~50× forward-sales multiple, supporting a bull case near $380 without assuming the prior $453 cycle high