ARM
ARM Holdings
Summary
What they do:
License CPU architecture IP — instruction sets, core designs, and subsystems — that every custom AI chip, smartphone processor, and increasingly every data center CPU is built on, sitting at Layer 06 as the architectural foundation beneath the compute silicon.
Why they matter:
ARM holds a near-monopoly on CPU architecture for custom AI ASICs — every Google TPU, Amazon Trainium, Meta MTIA, and NVIDIA Grace uses ARM cores, and switching to RISC-V would cost hyperscalers billions in software recompilation, making ARM the deepest architectural lock-in in the AI supply chain.
Recent performance:
Q3 FY2026 revenue $1.24B (+26% YoY), fourth consecutive billion-dollar quarter. Royalty revenue $737M (+27% YoY). Stock at ~$166, market cap ~$159B. Q4 FY2026 earnings expected May 2026.
Our Verdict
The instruction set monopoly powering every custom AI chip — 95%+ gross margins, 26% growth, and near-zero churn — but at ~150x+ forward P/E, the market has priced ARM as a perpetual growth compounder, leaving no margin of safety.
Structural trends
Structural
76
/ 100
Moat
8/10
Instruction set monopoly — 95% GM, near-zero churn, RISC-V is only long-term threat
AI Exp.AI Exposure
High~35% AI
Play Type
EstablishedAI Growth
~27%
Rel. Value
19
PREMIUMPriceLIVE
$161.22
+2.31%
Live via Yahoo Finance · refreshes every 5 min
Market Cap
$171.2B
P/E Ratio
212.1
P/S Ratio
36.7x
52W High
$183.16
52W Low
$95.32
52W Chg
69.1%
Beta
3.34
ARM is a pure intellectual property company. It employs roughly 7,000 engineers across Cambridge (UK), Austin, San Jose, and Bangalore who design CPU cores, instruction set extensions, and compute subsystems. ARM does not manufacture anything. It licenses blueprints — RTL code, verification vectors, and implementation guidance — to chip designers who send those blueprints to TSMC for fabrication.
The business model is elegant and capital-light. When Broadcom designs Google's TPU or Marvell designs Amazon's Trainium, both license ARM's Cortex CPU cores as the control processor. ARM collects two revenue streams: an upfront license fee ($1–10M per engagement) and a per-chip royalty ($0.50–2.00 per unit shipped). With billions of ARM-based chips shipping annually across mobile, data center, automotive, and IoT, this royalty stream compounds at 20%+ growth with 95%+ gross margins and near-zero marginal cost.
The company's relevance to AI infrastructure is accelerating. Every custom AI ASIC designed by Broadcom and Marvell — the two dominant custom chip houses — uses ARM CPU cores for control, I/O management, and host processing. Google's TPU, Amazon's Graviton and Trainium, Microsoft's Maia, Meta's MTIA — all ARM-based. NVIDIA's Grace CPU, designed to sit alongside Blackwell GPUs in AI servers, is ARM-based. The ARM v9 architecture introduced Scalable Matrix Extensions (SME) optimized for AI workloads, commanding higher per-chip royalties. And in 2026, ARM launched its first in-house chip — the AGI CPU — in partnership with Meta, targeting agentic AI workloads from smart glasses to data centers.
Q3 FY2026 (December 2025) revenue was $1.24B, up 26% YoY — the fourth consecutive billion-dollar quarter. Royalty revenue hit a record $737M (+27% YoY), driven by growth across data center, smartphones, and edge AI. License revenue of $505M (+25% YoY) reflects accelerating design-in activity. Q4 guidance of $1.47B implies full-year FY2026 revenue approaching $4.8B. ARM's operating margin runs at 40–50%, producing ~$2B in annual operating profit from pure IP licensing.
Supply Chain Dependencies
The Catch
ARM's valuation assumes the company is an unassailable monopoly that will compound royalties for a decade without competitive disruption. The 150x+ P/E leaves zero room for error — any deceleration in custom ASIC proliferation, any credible RISC-V progress at a major hyperscaler, or any regulatory action on ARM's licensing terms would crater the multiple. The RISC-V threat is real on a 10-year horizon: hyperscalers have the capital and engineering talent to develop RISC-V ecosystems if they decide ARM's royalty toll is too expensive. ARM's instruction set lock-in protects existing designs but cannot prevent hyperscalers from choosing RISC-V for future chips. And ARM's new AGI chip gambit — designing its own silicon — risks alienating the customers who license ARM IP, creating a compete-with-your-licensees dynamic that ARM has historically avoided.
If They Win
If ARM maintains its instruction set monopoly through the custom ASIC era — if every AI chip shipped globally for the next decade uses ARM cores, from NVIDIA's GPUs to hyperscaler ASICs to edge inference chips — ARM becomes the toll booth of artificial intelligence. Revenue compounds to $10B+ by 2030 with 50%+ operating margins, generating $5B+ in annual profit from pure IP licensing with near-zero capital requirements. The AGI chip succeeds and ARM adds direct silicon revenue on top of royalties. RISC-V remains a niche academic project, unable to match ARM's ecosystem depth. ARM's market cap reaches $500B+ as the market recognizes that in a world where compute is the scarce resource, the instruction set that every compute chip must speak is the most durable monopoly in technology.
Others in The Finished Chip
Not financial advice. All scores generated via AI algorithms using public data.