The Arm AGI CPU marks the first time Arm Holdings has designed and produced its own silicon product, extending the company’s compute platform beyond intellectual property licensing into physical data center chips. Arm announced the move on March 24, 2026, targeting the growing demand for infrastructure built around agentic AI workloads.
For more than three decades, Arm has supplied processor architecture to partners who built their own chips. This expansion into production silicon gives those partners a third option alongside Arm IP licensing and Arm Compute Subsystems (CSS), offering what the company describes as the broadest set of choices for building on Arm technology.
What the Arm AGI CPU Delivers
The Arm AGI CPU packs up to 136 Arm Neoverse V3 cores per chip, with 6 GB/s memory bandwidth per core at sub-100 nanosecond latency. Its 300-watt TDP design dedicates one core per program thread, which the company says eliminates throttling under sustained load. In air-cooled configurations, the chip supports up to 8,160 cores per rack; liquid-cooled systems can reach more than 45,000 cores per rack.
Arm claims the chip delivers more than 2x performance per rack compared with x86 platforms, based on internal estimates. The company also projects that this efficiency could enable up to $10 billion in capital expenditure savings per gigawatt of AI data center capacity.
Meta as Lead Development Partner
Meta served as the lead partner and co-developer of the Arm AGI CPU. The social media company plans to use the chip alongside its own Meta Training and Inference Accelerator (MTIA) silicon to improve orchestration across large-scale AI systems. Arm and Meta have committed to collaborating across multiple generations of the chip’s roadmap.
“Delivering AI experiences at global scale demands a robust and adaptable portfolio of custom silicon solutions, purpose-built to accelerate AI workloads and optimize performance across Meta’s platforms. We worked alongside Arm to develop the Arm AGI CPU to deploy an efficient compute platform that significantly improves our data center performance density and supports a multi-generation roadmap for our evolving AI systems.”
Santosh Janardhan, Head of Infrastructure, Meta
Ecosystem and Commercial Momentum
Beyond Meta, Arm confirmed commercial commitments from Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. These companies plan to use the Arm AGI CPU for accelerator management, control plane processing, and API and application hosting. More than 50 companies across hyperscale, cloud, silicon, memory, networking, and software sectors have expressed support for the platform expansion.
Notable supporters include AWS, Broadcom, Google, Marvell, Micron, Microsoft, NVIDIA, Samsung, SK hynix, and TSMC. Jensen Huang, CEO of NVIDIA, said the two companies are building “one seamless platform, from cloud to edge to AI factories.” TSMC confirmed it is manufacturing the Arm AGI CPU using its 3nm process technology.
“AI has fundamentally redefined how computing is built and deployed. Agentic computing is accelerating that change. Today marks the next phase of the Arm compute platform and a defining moment for our company.”
Rene Haas, CEO, Arm
Why Agentic AI Is Driving CPU Demand
The shift from training AI models to running continuously active AI agents is increasing the volume of tokens generated across systems. Arm estimates that as organizations scale agent-driven applications, data centers will need more than 4x their current CPU capacity per gigawatt. This creates demand for processors that handle reasoning, coordination, and data movement at scale, without the architectural overhead associated with x86 designs.
The Arm AGI CPU is designed specifically for this environment, offering deterministic performance under sustained load and high core density within standard power envelopes. These characteristics make it suited for the infrastructure requirements of large-scale artificial intelligence deployments.
Availability and Manufacturing
Early systems from OEM and ODM partners including ASRock Rack, Lenovo, Quanta Computer, and Supermicro are available now. Broader availability is expected in the second half of 2026. TSMC’s 3nm process technology underpins the chip’s performance and energy efficiency targets.
Arm Holdings (NASDAQ: ARM) is headquartered in Cambridge, England. The company’s compute platform currently reaches what it describes as 100 percent of the connected global population through partner devices and systems.
