Arm Launches Its First In-House Chip in 35 Years — An AI Data-Center CPU Targeting $15B/yr
Arm Holdings crossed a threshold this week that would have seemed unthinkable a decade ago: after 35 years as a pure-play chip architecture licensor, the company unveiled its first self-designed silicon — the Arm AGI CPU — built specifically for AI data center workloads. CEO Rene Haas told Reuters the chip is projected to generate $15 billion in annual revenue by 2031 as part of a $25 billion total revenue target. Arm shares jumped 20% on the announcement, with Intel and AMD also rising more than 5% on optimism around broader CPU demand driven by AI inference.
The AGI CPU is designed to handle the kind of sustained, multi-step reasoning tasks that define agentic AI workloads — systems that act on behalf of users with minimal oversight rather than respond to one-off queries. Arm's power-efficient architecture, which already powers the vast majority of mobile AI processing, gives it a credible starting position for inference tasks where energy costs are the dominant constraint. The chip will be manufactured at TSMC and has drawn support from more than 50 ecosystem partners including Google Cloud, Microsoft Azure, NVIDIA, Meta, and Hugging Face.
This marks a fundamental shift in the AI chip landscape. Nvidia has dominated the training market with GPUs, but the inference buildout — which will ultimately dwarf training in scale — has no clear winner yet. Arm entering the silicon business with an inference-optimized CPU adds a credible new challenger and raises questions about how its existing licensing partners, including Qualcomm and Apple, will respond to competing directly with their architecture provider.