Is Arm Holdings Still a Hidden Winner of the AI Chip Race?

Arm's revenue grew 23% and 24% in its past two fiscal years, exceeding its 17%-21% growth rates during the early smartphone era, as Arm-based CPUs gain data-center adoption in Nvidia, Amazon, Microsoft and Alphabet systems. Higher-value Armv9 architecture and compute-subsystem products are increasing royalty revenue per chip while AI-related unit volumes rise. The bullish growth case is tempered by an elevated valuation of 55x trailing sales and 188x free cash flow, despite Arm outperforming Nvidia over the past month, year and three years as of Sept. 17.

Analysis

ARM’s incremental AI exposure is economically real but easily overstated: hyperscaler CPU deployments are a slower, lower-dollar-content attachment to accelerator capex, and royalty recognition trails chip shipments rather than cloud capex announcements. The near-term earnings sensitivity is therefore more dependent on a favorable mix of advanced architecture adoption and licensing milestones than on headline AI spending. At the current valuation, even sustained 20%+ growth can disappoint if royalty revenue fails to accelerate faster than license revenue.

The second-order beneficiary is TSMC (TSM) rather than ARM alone: custom Arm-based server silicon requires advanced-node wafer capacity, packaging, and design ecosystems, while hyperscalers retain much of the cost savings from replacing merchant CPUs. ARM’s competitive risk is not x86 displacement reversing, but customers internalizing more of the design stack and negotiating economics as their volumes scale. Nvidia’s expanding CPU platform also creates a mixed outcome for ARM: it broadens architecture adoption but makes ARM’s economics dependent on a customer with substantial negotiating leverage.

Consensus appears to treat ARM as a clean, capital-light proxy for AI compute volume. The missing distinction is that ARM captures a small fraction of silicon value while its valuation already implies a disproportionate share of AI profit-pool expansion; this leaves asymmetric downside on any deceleration in Armv9 royalty uplift, data-center royalty mix, or smartphone recovery. Over the next 1-3 months, the key catalyst is royalty growth versus licensing growth and management commentary on server-chip shipment timing; over 6-18 months, custom CPU share gains must translate into materially higher royalty dollars per unit to support multiple durability.

Key Decisions for Investors

• Maintain an underweight/short bias in ARM versus long NVDA as a 3-6 month relative-value trade; ARM has greater multiple-compression risk if royalty growth lags AI expectations, while NVDA retains more direct exposure to current accelerator spending. Exit if ARM reports two consecutive quarters of royalty growth materially above total revenue growth and raises full-year royalty guidance.
• Do not chase ARM following AI-related announcements; establish only after earnings provide verifiable data-center royalty mix, Armv9 penetration, and customer shipment timing. A long setup requires evidence that royalty revenue—not upfront license revenue—is accelerating into the high-20% range or better.
• For AI infrastructure exposure, favor long TSM versus ARM over 6-12 months: custom hyperscaler silicon growth increases advanced-node content and manufacturing demand regardless of which CPU architecture wins. Thesis is impaired by a meaningful cut to leading-edge wafer demand, adverse Taiwan geopolitical escalation, or a broad hyperscaler capex reset.
• Monitor ARM’s revenue mix at the next result: licensing-led upside should be treated as lower-quality and less durable than shipment-driven royalty upside. If data-center commentary remains qualitative without disclosed royalty contribution or unit-volume evidence, retain the relative short rather than adding directional exposure.