Arm Holdings’ stock has tripled since the start of the year as investors have rushed toward the company’s growing role in AI data centers, where demand is shifting fast from graphics chips to CPUs. The move has given new weight to a business that has spent years dominating mobile device CPUs but is now gaining ground in server racks that power the latest AI systems.
That shift matters because Arm now says royalty revenue growth should accelerate to a 20% compound annual rate over the next five years, up from 14% over the last five. The company has made that case just as hyperscalers, after years of pouring most of their money into GPUs and custom AI accelerator chips, are adding large numbers of CPUs to handle the orchestration work that AI clusters need. Those processors decide where data moves, manage machine-to-machine communication and help coordinate servers full of GPUs.
The market has already noticed. In March, Arm estimated the total addressable market for data center CPUs would reach $100 billion by 2031, up from $50 billion in 2026. AMD has separately lifted its own server CPU market estimate to $120 billion by 2030, while Nvidia said on its most recent earnings call that it has visibility to nearly $20 billion in total CPU revenue this year. The numbers point to the same conclusion: CPUs are moving from a supporting role to a larger slice of AI infrastructure spending.
Lip-Bu Tan, Intel’s chief executive, said the CPU-to-GPU ratio could eventually shift from 1:4 to 1:1, a striking marker of how quickly the market is changing. Arm said in its fourth-quarter letter to shareholders that its share of CPU compute among the top hyperscalers is about 50%, and it has been benefiting from that rise as Amazon’s Graviton system is used extensively by 98% of the company’s top 1,000 customers. Amazon also recently signed a deal with Meta Platforms to deploy thousands of Graviton chips, while Microsoft and Alphabet have already started rolling out their own Arm-based CPUs in their data centers.
Still, the economics are not as simple as the stock chart suggests. Arm says licensing its intellectual property delivers very high margins, but that licensing accounts for only a tiny fraction of the total spending on the chips themselves. That leaves a gap between Arm’s influence over the architecture of AI data centers and the amount of hardware revenue it can actually capture. The company wants to make and sell its own chips, but whether it can turn its growing presence inside hyperscalers into a larger direct hardware business is the unanswered question hanging over the next phase of the story.
