So, Arm Holdings (ARM) decided it wasn't enough to design the brains for almost every smartphone on the planet. Now, it wants to build the brains for the AI revolution, too. And it's doing it by becoming something it's never been before: an actual chipmaker.
The company's shares jumped more than 12% in Wednesday's premarket session after it announced a major strategic expansion. For the first time in its history, Arm is launching production silicon—a physical chip it designed and will sell. This isn't just another license for its famous architecture; this is Arm rolling up its sleeves and getting into the fabless semiconductor business with a new AI-focused data center CPU called the Arm AGI CPU.
Think of it this way: Arm built the incredibly successful blueprint that everyone uses to make efficient chips. Now, it's decided to build the house itself, and it's aiming for the most valuable neighborhood in tech: AI infrastructure.
Why Build a Chip When You License the Blueprint?
For decades, Arm's business model was brilliantly simple and asset-light: design incredibly power-efficient chip architectures, license the intellectual property (IP) to companies like Apple (AAPL) and Qualcomm (QCOM), and collect royalties. It was the ultimate behind-the-scenes player. So why change a winning formula?
The answer is the trillion-dollar gravitational pull of artificial intelligence. Arm is betting that the next phase of AI—moving from models that answer questions to "agents" that perform tasks continuously—will require a massive, fundamental shift in data center compute. These always-on AI agents will need vastly more CPU power to manage and orchestrate workloads alongside specialized AI accelerators like those from NVIDIA (NVDA).
Arm's argument is that data centers might need more than four times their current CPU capacity per gigawatt of power to support this future. That creates a huge market for a CPU that is both extremely powerful and extremely efficient. Enter the Arm AGI CPU.
The company says this chip, built for "agentic AI" workloads, delivers over twice the performance per rack compared to traditional x86 server platforms (think chips from Intel (INTC) and AMD (AMD)). The specs are built for density: up to 136 of Arm's latest Neoverse V3 cores, a 300-watt thermal design power (TDP), and support for systems that can pack over 45,000 cores into a single, liquid-cooled rack.
The Meta Partnership and the $15 Billion Target
No tech company makes a move this big alone. Arm's lead partner and co-developer on this project is Meta (META), which is pouring billions into its own AI ambitions. But the customer list reads like a who's who of cloud and AI: Cerebras, Cloudflare, F5, OpenAI, SAP, and SK Telecom. Arm is also working with hardware manufacturers like ASRock Rack, Lenovo, Quanta, and Supermicro to get systems built.
Perhaps most telling is the broader ecosystem support. More than 50 companies are backing the platform, including the cloud hyperscalers—Amazon's AWS (AMZN), Google (GOOG), and Microsoft (MSFT)—alongside NVIDIA (NVDA) and the chip manufacturing titan TSMC (TSM).
This isn't a niche experiment. It's a full-scale assault on the data center CPU market. And Arm has put a staggering number on its ambition: it expects this new chip business to generate about $15 billion in annual sales within five years. To put that in perspective, that figure could overtake the sales from its entire existing core licensing business. It's a bet-the-company kind of pivot.
"Today marks the next phase of the Arm compute platform and a defining moment for our company," said CEO Rene Haas. "With the expansion into delivering production silicon with our Arm AGI CPU, we are giving partners more choices all built on Arm's foundation of high-performance, power-efficient computing, to support agentic AI infrastructure at global scale."
