ARM Holdings sits in an unusual spot in the AI chip stack. The company does not fabricate silicon, does not sell GPUs, and does not compete head to head with Nvidia in training accelerators. It licenses CPU design IP and collects a royalty on every chip shipped using that IP. After Q4 FY26 results landed on May 6 with data center royalties more than doubling year over year, the bull case is louder than ever. So is the valuation.
For investors already running an AI chip basket of NVDA, AMD, and AVGO, the real question is not whether ARM is a good business. It is whether ARM adds exposure you do not already own.
ARM's Royalty Model: Why It's Different From NVDA, AMD
ARM does not ship physical silicon. It licenses architectures (Armv9, Compute Subsystem) to companies like Apple, Qualcomm, Amazon, and Nvidia, then collects a royalty per chip those partners ship. Royalty rates step up when partners adopt newer architectures or use ARM's own pre packaged Compute Subsystem blocks.
Smartphones, autos, and the new data center leg
For most of ARM's life the royalty pool was smartphones. That base is mature but still growing on mix shift to v9. The newer engine is data center: hyperscalers building custom CPUs on ARM cores, including Amazon Graviton, Microsoft Cobalt, and Google Axion.
What this means for cyclicality
Because ARM gets paid per chip shipped, its revenue lags partner shipment cycles but is less violent than fabless peers. ARM does not eat inventory write downs the way AMD did in 2022, and it does not face GPU allocation risk the way NVDA does. The flip side: ARM's upside per chip is capped at the royalty rate.
Valuation: ARM's Premium and What Justifies It
ARM trades at a steep premium to other chip names on every standard multiple. Q4 FY26 revenue of $1.49 billion brought full year revenue to $4.92 billion, with royalty revenue at $2.61 billion. Management is guiding toward $15 billion in chip related revenue by fiscal 2031.
The bull math
If data center royalties keep compounding at the current rate, ARM's mix shifts from smartphone heavy to AI infrastructure heavy. AI infrastructure royalties carry higher rates per chip and are tied to hyperscaler capex, which remains in expansion mode. Bulls argue the multiple compresses naturally as revenue catches up.
The bear math
The premium assumes near flawless execution. Any slip in v9 adoption, pricing pushback from large customers, or rotation back to merchant silicon would compress the multiple fast.
If you are building an AI chip basket and you already own NVDA, AMD, and AVGO, ask whether ARM adds new exposure or duplicates capex sensitivity you already have.
Own, Pass, or Wait: Three Scenarios
Not every AI portfolio needs ARM. The right answer depends on what else you hold and what risk you are trying to add.
Own it
You believe hyperscaler custom silicon (Graviton, Cobalt, Axion) keeps taking share from merchant CPUs, and you want a name that benefits from that shift without picking which hyperscaler wins. ARM gets paid regardless.
Pass on it
You already own NVDA and AVGO and view ARM as duplicating capex beta without adding GPU or networking exposure. Your basket is full and ARM's premium does not earn its slot.
Wait for a pullback
You like the thesis but want a better entry. ARM has historically given investors 15 to 20 percent drawdowns around guidance resets or hyperscaler capex scares. A waitlist position is reasonable.
ARM's Place Alongside NVDA, AMD, AVGO in a Chip Basket
Most AI chip baskets we have seen in retail portfolios skew GPU heavy. ARM plays a different role.
Complementary exposure
NVDA captures GPU training and inference. AMD adds GPU competition plus server CPU share. AVGO captures custom ASIC and networking. ARM captures the CPU royalty pool across mobile, edge, and increasingly cloud. The four names overlap less than they appear.
Position sizing
A reasonable construction is NVDA as the core, AVGO as the custom silicon and networking sleeve, AMD as the GPU optionality position, and ARM as the royalty layer. Sizing ARM smaller than NVDA reflects the higher multiple and the v9 adoption risk. For a broader survey of names beyond the GPU leader, see our 5 AI Stocks Beyond NVIDIA writeup.
Conclusion
ARM is the cleanest pure play on the CPU side of AI infrastructure. The royalty model is differentiated, the data center growth is real, and the long term chip revenue target is credible. The valuation is the friction. If you are starting an AI basket from scratch, ARM earns a slot.
If you already own NVDA, AMD, and AVGO, the honest question is whether ARM adds exposure or just adds beta.
For more on how hyperscaler earnings reprice this group, our Mag 7 read through piece walks through the mechanism.
FAQ
Is ARM Holdings stock a good AI play?
ARM is a credible AI play through CPU royalties on hyperscaler custom silicon and AI accelerator host CPUs, but the premium valuation assumes continued execution.
How is ARM's royalty model different from Nvidia or AMD?
ARM licenses CPU design IP and collects per chip royalties, so it does not carry inventory or fab risk but also caps its upside per unit shipped.
Should I own ARM if I already hold NVDA and AVGO?
Only if you believe ARM adds CPU royalty exposure your basket lacks, otherwise it duplicates AI capex beta you already own.
What did ARM's Q4 FY26 results show about AI growth?
ARM reported $1.49 billion in Q4 revenue with full year royalties of $2.61 billion and data center royalties more than doubling year over year.
