For 36 years, Arm Holdings made CPU designs and licensed them to anyone who wanted them, who then designed and made their own custom modifications.
But that is changing. Arm has announced its expansion into production of silicon products for the first time in the company’s history. The company’s first product is the Arm AGI CPU, an Arm-designed CPU purpose built for AI data centers, with Meta as its first customer.
“AI has fundamentally redefined how computing is built and deployed. Agentic computing is accelerating that change,” said Rene Haas, CEO of Arm in a statement. “Today marks the next phase of the Arm compute platform and a defining moment for our company. 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.”
The chip was designed and manufactured by Ampere, the Arm start up founded by ex-Intel executive Renee James. SoftBank, Which owns Arm, acquired Ampere last year for $6.5 billion. Manufacturing of the chip will be done by TSMC on a 3-nanometer process node.
Arm argues that as AI shifts from training models to deploying continuously running agents that reason, plan and act, the volume of tokens generated across AI systems is rapidly increasing and requires significantly more CPUs to handle reasoning, coordination and data movement.
The Arm AGI CPU is expected to be the foundation for agentic data centers. It comes with up to 136 Arm Neoverse V3 cores per CPU, designed for use in SoC, blade, and rack designs, with 6GB/s memory bandwidth per core at sub-100ns latency. Arm claims the AGI CPU delivers more than twice the performance per rack versus x86 CPUs.
With a Thermal Design Power (TDP) of 300 watts, the AGI CPU draw significantly less power than X86 based CPUs from Intel and AMD. It supports high-density 1U server chassis that allow air-cooled deployments with up to 8,160 cores per rack, and liquid-cooled systems delivering 45,000+ cores per rack.
Meta serves as the lead partner and co-developer of the AGI CPU, optimizing it for its family of apps and working alongside Meta’s own custom silicon, called Meta Training and Inference Accelerator (MTIA), enabling more efficient orchestration in large-scale AI systems.
Meta is hardly alone. Arm has confirmed additional commercial momentum with partners including Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. These customers will deploy the Arm AGI CPU for key agentic CPU use-cases.
Arm is partnering with lead OEMs and ODMs to deliver new systems, including ASRock Rack, Lenovo, Quanta Computer, and Supermicro, with early systems available now and broader availability expected in the second half of the year.
Jim McGregor, principal analyst with Tirias Research, said Arm’s hand was forced in this decision. “I think it’s necessary for Arm to branch out because they’re not part of a larger entity like they would have been with Nvidia,” he said.
McGregor noted a lot of questions remain unanswered. Arm is targeting the x86 market with these chips and offered up competitive benchmarks but didn’t give any details on what systems were being used in comparison. “So you couldn’t verify anything. I don’t know what parts they were comparing it to,” he said.
Also, it’s not clear that this is going to be an ideal solution for the enterprise. “I’m not really sure it’s intended to be a general-purpose processor anyway. It’s supposed to be an AI processor like Vera [the Nvidia Arm CPU]. It’s supposed to maximize the utilization of the AI accelerators,” said McGregor.