🧩

ARM

πŸ’‘
Thesis. Arm is the toll booth of computing. It designs the CPU architecture inside almost every phone and a fast-rising share of data-centre chips, and collects a royalty on each one shipped. The model is capital-light and the royalty rate per chip is climbing. The catch is that it trades near 180x forward earnings, is now selling its own CPU in competition with its licensees, and is under an FTC probe.
Reviewed 19 September 2026, after fiscal Q1 2027 results. Next review after Q2 earnings on 5 November 2026.

What it does

Arm licenses chip designs. It doesn't manufacture anything.
The money arrives twice. A customer pays an upfront licence fee for the right to use an Arm design, then pays a royalty on every chip it ships using that design β€” forever. Licensing was $2.31B in FY2026 and royalties $2.61B.
The newer Compute Subsystems (CSS) β€” pre-built blocks rather than raw architecture β€” carry royalties above 10% of a chip's selling price, against roughly 5% on the older Armv9 architecture. Same chip shipped, twice the toll.
Since March 2026 Arm also sells its own data-centre CPU, which puts it in competition with the customers it licenses to. CEO is Rene Haas. SoftBank still controls roughly 90% of the shares.
A royalty business compounds two ways at once: more chips shipped, and a bigger cut of each one. That is the whole bull case in a sentence.

Bull case

  • Q1 FYE27, reported 29 July 2026: record revenue $1.29B, up 22%. Royalties $715M (+22%) and licensing $574M (+23%) β€” both halves growing together
  • The data centre is inflecting. Data-centre royalties more than doubled year over year. Arm cores sit in roughly half of hyperscaler CPU compute
  • Neoverse is compounding. Shipments passed 1.5 billion cores, and the most recent 500 million of those came in just nine months
  • The rate is rising, not just the volume. Growth is attributed explicitly to higher royalty per chip via Armv9 and CSS, with CSS above 10% of selling price
  • Margins are what you'd expect from an IP business. Non-GAAP operating margin 41.2%, non-GAAP EPS $0.45. Essentially all gross margin, very little capital employed
  • Q2 guidance keeps the pace. Revenue $1.38B Β± $50M, non-GAAP EPS $0.47 Β± $0.04

Bear case

  • The valuation leaves no room. Roughly 180x forward earnings with a beta of 3.77 β€” the business can do well and the stock can still fall hard
  • Its own CPU is a smaller story than it looked. Arm says demand for the AGI CPU "exceeds $2 billion across FYE27 and FYE28", with capacity secured for the first $1B. Real, but a fraction of the scale this page previously carried
  • It now competes with its own customers. Selling a data-centre CPU puts Arm opposite Apple, Qualcomm and NVIDIA, who are also its licensees
  • Regulators are looking. An FTC antitrust probe opened 15 May 2026, and the Qualcomm/Nuvia case already loosened licensing terms once
  • RISC-V is the long game. A free, open architecture is a structural threat to a business whose product is a toll
  • Disclosure just got thinner. From Q1 FYE27 Arm stopped reporting remaining performance obligations and licence counts, removing the best forward-visibility metric it had
  • SoftBank control and a thin float amplify every move

Major customers

Arm's customers are licensees β€” they pay upfront for the architecture, then a royalty on every chip. "Customer" here means anyone shipping an Arm-based chip.
  • Apple β€” the architecture underneath every iPhone, iPad and Mac processor
  • Qualcomm β€” Snapdragon, and the counterparty in the litigation that loosened licensing terms
  • NVIDIA β€” Grace and Vera CPUs, plus Arm cores across its data-centre platforms
  • Amazon β€” Graviton server CPUs, one of the most important data-centre design wins there is
  • Google and Microsoft β€” Axion and Cobalt custom silicon programmes
  • MediaTek, Samsung, Broadcom β€” mobile and infrastructure chips
  • Automotive and embedded β€” a long tail of smaller licensees
πŸ”—
Read-through. Arm sits underneath almost everything else in this file. Royalties track total chips shipped rather than AI capex specifically, so smartphone volumes still matter as much as the data-centre story β€” a weak handset cycle hurts Arm even in a strong AI year.
Every data-centre CPU from
πŸ›οΈ
NVDA
carries an Arm royalty, so NVIDIA's CPU ramp reads straight through.
πŸ›οΈ
AMD
running the x86 side means the CPU market is growing fast enough for both, but the share battle is live. The custom-silicon programmes at
πŸ•ΈοΈ
AVGO
and
πŸ•ΈοΈ
MRVL
are accelerators rather than CPUs, but the CPUs paired with them in the rack are increasingly Arm.
🧩
CEVA
is the other IP-licensing model in this category, at a fraction of the scale and without the ecosystem lock-in.
The structural tension: Arm's biggest licensees are the ones most capable of designing around it or negotiating the royalty down. Watch the rate per chip, not only unit growth.

What would change the view

  • Royalty rate per chip β€” the CSS and Armv9 mix is the whole double-compound argument
  • Data-centre royalty growth β€” currently more than doubling year over year
  • AGI CPU revenue actually converting against the stated $2B+ demand across FYE27–28
  • FTC probe developments, and whether any licensee follows the Qualcomm precedent
  • Smartphone unit volumes, which still carry the base of the royalty stream
  • SoftBank actions on the float

Update log

  • 19 Sep 2026 β€” Fiscal Q1 2027 results added (revenue $1.29B, +22%; royalties +22%; data-centre royalties more than doubled). Corrected the AGI CPU demand figure from ">$20B in six weeks" to Arm's own stated ">$2B across FYE27 and FYE28" β€” the earlier number was wrong by roughly ten times. Noted that Arm has stopped disclosing remaining performance obligations. Verdict held at Watch.
  • 7 Sep 2026 β€” Reviewed. Q1 FYE27 figures had not been pulled; page flagged as not current.
  • 6 Jul 2026 β€” Full review.

Research and education only β€” not investment advice.