Arm Holdings Is Quietly Becoming the CPU Backbone of the Agentic AI Era

For years, the AI investment narrative centered on one thing: GPUs. Nvidia became the poster child of the boom, and for good reason — training large language models is enormously GPU-intensive. But a shift is quietly underway, and Arm Holdings is at the center of it.

Bernstein this week published a market note arguing that Arm stands at the beginning of a CPU renaissance driven by agentic AI. The distinction matters. Chatbots run inference in bursts. AI agents — the kind that take multi-step actions, browse the web, write code, and orchestrate other tools — run continuously. That changes the compute profile entirely. Continuous, low-latency workloads favor power-efficient architectures, and Arm’s CPU designs are unrivaled there.

Bernstein projects that Arm could capture a fourfold increase in server CPU market share over the next four years, growing its addressable market to $137 billion. The firm also sees a path to five times current profitability by 2030. That’s not a moonshot number — it’s a compound growth thesis built on the structural shift from chatbot to agent.

The latest earnings give that thesis real footing. In Q4 FY2026, Arm reported a 49% jump in net income to $313 million, with revenue rising 20% to $1.49 billion. For Q1 FY2027, guidance calls for $1.26 billion in revenue — another ~20% year-over-year gain. This isn’t a company waiting for the future to arrive.

What makes Arm particularly interesting for long-term investors is its business model. Arm doesn’t manufacture chips — it licenses its chip architecture to semiconductor companies and device makers worldwide. Apple, Qualcomm, Amazon, and virtually every major chip designer uses Arm’s IP. Every time one of those companies ships a chip, Arm collects royalties. As AI agents proliferate across smartphones, data centers, and edge devices, the royalty stream compounds quietly in the background.

The stock has surged 46% in a week on the back of the Bernstein note and earnings — which means the near-term trade is probably not the point. The more interesting question for patient investors is whether the 5x profit thesis plays out over the next four years. If AI agents become as ubiquitous as smartphones — and there are real reasons to think they will — Arm’s royalty model could be one of the most durable compounding machines in tech.

Not every AI winner has “AI” in its name. Sometimes it’s the company collecting tolls on the road everyone else is building.