mstack

ARM

5 min read

ARM — Arm Holdings

TL;DR

ARM’s core IP licensing business hit an inflection in the datacenter: over 1 billion Neoverse CSS cores deployed, royalty revenue more than doubled YoY, and CSS is projected to exceed 50% of royalty revenue within a few years as every major hyperscaler adopts Neoverse for custom CPUs. The business model is now shifting further — ARM is transitioning from pure IP licensor to full chip vendor with its “Phoenix” platform, taking Meta and OpenAI as anchor customers. That move puts ARM in direct competition with its own CSS licensees, which is the key structural tension. The investment risk flagged in the corpus: SoftBank’s Masayoshi Son likely has large margin loans against his ARM stake — forced selling is the tail risk.12

Business

IP licensing company. Designs CPU and interconnect architectures licensed to chip designers globally. Revenue model: upfront license fees + royalties on shipped chips. Two primary franchises: (1) Neoverse CSS — compute subsystems for datacenter CPU/DPU; licensed to Apple, Qualcomm, Amazon (Graviton), Google, Ampere, and others; (2) Cortex-A/Cortex-X — mobile application processor cores inside every major smartphone SoC. Majority owned by SoftBank. Listed on Nasdaq.1

Thesis

  • CSS is winning the datacenter. More than 1 billion Neoverse cores have been deployed across datacenter CPUs and DPUs. 21 CSS licenses signed across 12 companies. Datacenter royalty revenue more than doubled YoY; CSS is projected to represent >50% of royalty revenue “in the next couple of years.” Every major hyperscaler (AWS/Graviton, Google/Axion, Microsoft/Cobalt, Meta/Phoenix) has adopted Neoverse — ARM is the de facto standard for hyperscaler custom CPUs.1
  • Phoenix: business model pivot from licensor to chip vendor. In 2026, ARM began offering full datacenter CPU designs — not just IP — under the “Phoenix” platform. Meta is the first Phoenix customer (128 Neoverse V3 cores, TSMC 3nm, 12 DDR5 channels, 96 PCIe Gen 6 lanes, TDP 250–350W). OpenAI is a second Phoenix customer (part of the Stargate/SoftBank venture). Cloudflare is a potential third. Next-generation “Venom” CPU is in development.1
  • Mobile royalty base is large and recurring. Every Android flagship SoC (Snapdragon, Dimensity, Exynos) licenses ARM Cortex-A/X cores. The royalty base is sticky even if per-chip ASPs are under pressure.

Risks

  • Phoenix competes with CSS licensees. By offering full chip designs, ARM now competes directly with Qualcomm, Ampere, and others who license Neoverse CSS to build their own CPUs. This is a business model risk: ARM is simultaneously licensor and product competitor to its largest clients.1
  • SoftBank / Masayoshi Son forced-selling tail risk. Irrational Analysis (Feb 2026 earnings): “ARM terrifies me because Masa definitely has massive loans against his shares. Nothing else really matters when such an existential risk is in play.” SoftBank’s ownership stake at ARM’s elevated valuation creates forced-selling risk if SoftBank’s broader portfolio deteriorates.2
  • Smartphone volume exposure. ARM management guided a 15% smartphone unit volume reduction at the Feb 2026 earnings call; the Irrational Analysis author flagged 30–50% as potentially more realistic. Mobile royalties are the volume base — a larger handset correction would compress ARM’s royalty income meaningfully.2
  • RISC-V displacement at the margin. Open-source RISC-V is not a near-term threat to ARM’s core franchise, but hyperscalers building their own RISC-V microcontrollers and embedded cores could erode the TAM at the bottom of the stack over time.

Recent catalysts

  • 2026-02-06 — Irrational Analysis earnings roundup: ARM reported strong results but stock tanked on earnings night alongside QCOM; ARM recovered the next day while QCOM did not. Smartphone volume guidance flagged as potentially too optimistic; Masa loan risk highlighted.2
  • 2026-02-09 — SemiAnalysis “CPUs Are Back”: full ARM section — CSS business metrics, Phoenix platform details (Meta anchor, OpenAI customer, Venom roadmap), business model shift to chip vendor.1

Second-order reads

  • 2026-02-09 — SemiAnalysis, CPUs Are Back — Meta’s Phoenix CPU deployment uses TSMC 3nm → positive for TSM N3 demand from a new hyperscaler socket type; neutral/negative for AMD and INTC to the extent Phoenix displaces general-purpose CPU sockets.
  • 2026-03-28 — Chipstrat, Agentic AI Needs CPUs — AWS Trainium3 head nodes moved from Intel Sapphire Rapids to Graviton4 (Neoverse V2) → directly validates ARM CSS royalty TAM expansion in hyperscaler clusters; negative for INTC x86 share.1

Valuation & positioning

No valuation multiples in the corpus. The investment tension: ARM’s CSS business has clear secular momentum (datacenter CPU royalty doubling YoY), but the Phoenix pivot risks alienating existing licensees and the SoftBank overhang is an unquantified forced-selling tail risk. The corpus frames ARM as a strong underlying business with a dangerous structural shareholder.2

Sources

QCOM — major Cortex-A licensee; Oryon (Nuvia) is ARM v9-derived; Apple cliff affects QCOM, not ARM royalties AMD INTC — competing datacenter CPU architectures; ARM CSS is the disruptor TSM — Phoenix CPU on TSMC 3nm; ARM datacenter royalty growth is a read-through to N3 demand NVDA — Vera CPU uses ARMv9.2 cores; ARM captures royalty on every NVDA Vera deployed

Footnotes

  1. SemiAnalysis — CPUs Are Back: The Datacenter CPU Landscape — 2026-02-09 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  2. Irrational Analysis — Earnings Roundup: COHR QCOM $ARM — 2026-02-06 ↩ ↩2 ↩3 ↩4 ↩5

Graph View

Backlinks