Arm AGI CPU Opens New Revenue Path for Arm
Arm AGI CPU launch with Meta as lead partner signals Arm's move into selling AI chips and projects multibillion revenue, shifting investor positioning.
KEY TAKEAWAYS
- Arm unveiled the Arm AGI CPU, its first in-house data-center CPU with Meta as lead partner.
- Arm projected the AGI CPU would add billions in annual revenue, signaling a shift toward selling silicon.
- It supports up to 136 Neoverse V3 cores and Arm said rack designs deliver more-than-2x x86 performance.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
Arm Holdings on March 24, 2026 unveiled the Arm AGI CPU, its first in-house designed data center CPU for agentic AI workloads, with early systems beginning to ship and broader availability planned for the second half of 2026.
AGI CPU Technical Specs and Performance
The Arm AGI CPU targets agentic AI workloads—reasoning, planning, and acting—and is built on Neoverse V3 cores. It supports up to 136 cores per socket, delivers about 6 gigabytes per second of memory bandwidth per core at sub-100-nanosecond latency, and has a 300-watt thermal-design power rating.
Arm designed the chip for dense deployments, stating that 1U air-cooled servers can reach 8,160 cores per rack, while liquid-cooled configurations exceed 45,000 cores per rack. The architecture offers more than twice the rack performance of comparable x86 systems. The company also cited up to $10 billion in capital expenditure savings per gigawatt as part of the platform’s economics. Taiwan Semiconductor Manufacturing Co. (TSMC) will produce the CPU using a 3-nanometer process.
Customers and Commercial Outlook
Meta is the lead partner and initial customer, collaborating with Arm on a multi-generation roadmap. Other customers include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. Original equipment manufacturers and original design manufacturers such as ASRock Rack, Lenovo, Quanta, and Supermicro have committed to building systems around the design.
The platform has backing from more than 50 firms across silicon and cloud ecosystems, including Amazon Web Services, Broadcom, Google, Marvell, Micron, Microsoft, NVIDIA, Samsung, and SK hynix. Early systems are available now, with broader availability expected in the second half of 2026.
Arm projected the AGI CPU would add billions of dollars in annual revenue and described a multi-billion-dollar opportunity in AI data centers. The company said agentic AI workloads will require more than four times the current CPU capacity per gigawatt. This projection, combined with partner commitments and early hardware availability, signals a strategic shift toward selling silicon alongside its licensing business, potentially altering Arm’s revenue profile as data centers adopt higher core counts and denser rack economics.
Rene Haas, Arm’s chief executive, said, "Today marks the next phase of the Arm compute platform and a defining moment for our company."
