Arm Poised to Outpace Nvidia in AI Inference with 20% Royalty Growth

Context Inference – the phase where trained AI models generate outputs – now consumes two‑thirds of AI computing power, up from half a year earlier. Deloitte projects the inference‑focused chip market at $50 bn this year, and McKinsey sees data‑center inference demand rising from 21 GW to 93 GW by 2030, a 35% CAGR.

Key Facts - Arm’s royalty stream from its AI‑optimized Armv9 architecture is set to expand at a 20% compound annual growth rate from fiscal 2026 to 2031, outpacing its historic 14% growth. - The royalty rate for Armv9 is nearly double that of the previous Armv8 generation, boosting per‑chip earnings for licensees. - Arm’s own silicon ambitions could generate $15 bn in annual revenue by fiscal 2031, contributing to a projected total revenue of $25 bn – more than five times its recent $4.7 bn. - Nvidia, AMD, Broadcom and Intel are all developing inference chips, but they rely on Arm’s architecture for products such as Nvidia’s Grace CPU and Vera processor, which Nvidia expects to become a $20 bn business. - Cloud giants Google and Amazon, as well as smartphone makers Qualcomm, MediaTek and Apple, already use Arm IP for in‑house AI inference CPUs.

What It Means Arm’s licensing model gives it a diversified revenue base across data centers, edge devices and consumer electronics. As inference workloads dominate AI compute, demand for energy‑efficient designs will rise, driving more companies to adopt Arm’s IP and pay higher royalties. The 20% royalty CAGR suggests Arm could capture a larger slice of the $50 bn inference chip market than pure‑play silicon vendors that sell finished products.

Looking ahead, the pace of inference adoption and the rollout of Arm‑based custom chips will determine whether the royalty surge translates into the projected $25 bn revenue target. Watch for quarterly updates on Arm’s silicon sales and licensing deals as the inference market matures.