ARM Holdings: The Royalty Machine Quietly Collecting From Every AI Chip

Every smartphone you have ever owned ran on an ARM-based chip. Every Apple M-series Mac runs on ARM. Nvidia's Grace CPU is ARM. AWS Graviton is ARM. Google Axion is ARM. Microsoft Cobalt is ARM. The custom chips Meta is building. ARM. Tesla's HW4 and HW5 in-house compute. ARM. Even the smallest microcontroller in your washing machine is ARM more often than not.

ARM does not manufacture a single chip. ARM owns the architecture. Every customer pays a license fee plus a royalty per chip shipped. The business model is closer to a software company than a chip company. The margins look like software. The growth rate looks like AI infrastructure. The risk profile is nothing like the cyclical commodity semis people are used to.

This is the cleanest royalty machine in the global tech stack and most retail investors still treat it like a chip stock. That is the opportunity.

ARM is a high-margin, recurring royalty business riding two compounding waves at once, smartphones moving to v9 architecture at higher royalty rates, and AI inference moving to the edge on ARM-based silicon across every device class.

ARM designs the instruction set architecture and reference cores that chip makers license. There are two revenue streams. License fees, paid upfront when a customer signs a multi-year agreement. Royalties, paid on every chip shipped using the architecture. Royalties scale with the price of the chip and the architecture version.

When a smartphone maker moves from Armv8 to Armv9, the royalty rate goes up. When a chip vendor uses higher value compute IP, the royalty rate goes up. When AI workloads push more compute into every device, the royalty per chip goes up.

This is the closest thing to a tax on global compute that exists in public markets.

The first leg of the bull case is the v9 transition. ARM has been migrating its installed base from v8 to v9 architecture. v9 pays a higher royalty rate. That mix shift compounds every quarter through the rest of the decade.

The second leg is the edge AI story. Every flagship smartphone now ships with a dedicated neural processing unit. Every laptop is racing to add on-device AI. Every wearable. Every car. Every smart speaker. The dominant architecture across every one of those device classes is ARM. As AI inference moves out of the cloud and onto the device, compute per unit increases, chip prices rise, and royalty revenue per device goes up.

The third leg is the data center story. AWS Graviton, Nvidia Grace, Google Axion, Microsoft Cobalt, Ampere, and a growing number of hyperscaler custom CPUs are all ARM. Data center penetration is still small but accelerating fast. Every percentage point of share gain inside hyperscaler CPU spend is meaningful royalty growth.

The fourth leg is automotive. The ARM share of automotive silicon is climbing as every major automaker moves to centralized compute architectures with in-cabin AI. Per-vehicle silicon content is exploding. ARM is the default architecture.

Royalty revenue growth quarter over quarter. v9 mix as a percentage of total royalty revenue. Hyperscaler design wins. Compute Subsystems (CSS) adoption. Annualized contract value on multi-year license deals. Operating margin trend.

The metric I care about most is royalty per chip. As long as that is climbing, the long-term thesis is intact even if unit volumes wobble.

ARM trades at a premium multiple because the market correctly identifies it as a software-like business. The multiple looks high until you build the model. Royalty growth at sustained mid-teens to twenty-percent compounding, paired with operating leverage on a fixed cost base, produces earnings growth that justifies the premium.

The bear argument is always that the multiple is too high. The same bear argument has been made on Microsoft, Visa, Adobe, and every other recurring revenue platform business at one point or another. It is a real risk but it is also a reason these names compound for decades.

The first risk is RISC-V. RISC-V is an open architecture that could eat into ARM share in lower-end markets over time. The threat is real but the moat is also real. ARM has a multi-decade head start on tooling, ecosystem, customer relationships, and a software stack that just works.

The second risk is customer concentration. A handful of mega customers drive a large share of revenue. Any single major customer renegotiating terms is a meaningful event.

The third risk is geopolitics. ARM operates in a US-China trade conflict zone. Chinese smartphone exports, Chinese chip licensing, Chinese fab access. Any tightening of export controls could pinch growth in one of ARM's largest markets.

The fourth risk is valuation compression. If multiples compress across software-like names, ARM compresses with them.

The fifth risk is execution on the data center push. ARM's data center growth depends on hyperscalers continuing to build custom ARM silicon and on traditional ARM server vendors actually scaling. The trend is in place but it is not a foregone conclusion.

ARM is a core long-term position. Not a trading vehicle. Not a sentiment play. A multi-year compounder where the business model does the work.

I size it like any other quality compounder. A meaningful position. Adds on weakness when the market panics about a single quarter. Patience.

I want to be clear about why this is structurally different from buying a fabless chip designer or a foundry. Cyclical chip names get killed when inventory corrections hit. ARM keeps collecting royalties on every chip that ships regardless of who designed it or where it was manufactured. The royalty model is durable across the cycle in a way that pure cyclical semis simply are not.

ARM also benefits when AI compute moves to the edge in a way that pure data center AI plays do not. If Nvidia keeps winning the training market, Nvidia keeps building ARM-based Grace CPUs and ARM keeps collecting royalties. If Apple keeps shipping iPhones, ARM keeps collecting royalties. If AWS keeps deploying Graviton, ARM keeps collecting royalties. The business does not require you to pick which AI hardware company wins.

That is the elegance of the model.

Pull ARM's most recent earnings release and read the royalty revenue breakdown. Look at the v9 mix line. Look at the CSS adoption commentary. Look at the data center share number. Now do the same exercise for Nvidia, Intel, AMD, and one foundry name. Compare margin profile, revenue volatility, and customer concentration. ARM's profile will look strikingly different. That is the case.

Read next: MRVL: The Quietest AI Monster of 2026 | NVDA: Why The King Is Not Done

*⚠️ Important Disclaimer: MentorSurge is not a financial advisor. This post is for educational and entertainment purposes only. Nothing on this site constitutes financial, investment, or trading advice. Semiconductor stocks are volatile and subject to industry cycles and geopolitical risk. Always do your own research and consult a licensed professional.*