Arm Holdings has announced the Arm AGI CPU, its first production silicon designed for AI data centers. This marks the first time the company is extending its compute platform beyond IP and Compute Subsystems (CSS) into full silicon products.
At the same time, Arm confirmed a partnership with Meta, which will act as the lead partner and co-developer for the new CPU platform.
Expansion into production silicon
Arm stated that its compute platform has powered billions of devices over the past three decades through its licensing model. With increasing demand for deploying Arm-based infrastructure at scale, the company is now expanding its strategy to include Arm-designed silicon products.
This approach allows partners to choose between licensing Arm IP, adopting Arm CSS, or deploying complete Arm-designed CPUs.
Designed for agentic AI workloads
The Arm AGI CPU is built to support agentic AI workloads, where AI systems continuously run tasks involving reasoning, planning, and execution.
Arm noted that this shift is increasing the volume of tokens processed across systems and driving higher demand for CPUs to handle coordination, reasoning, and data movement.
As a result, data centers are expected to require more than four times the current CPU capacity per gigawatt, increasing the need for higher compute density within existing power limits.
Arm AGI CPU specifications and capabilities
- Up to 136 Arm Neoverse V3 cores per CPU
- 6GB/s memory bandwidth per core with sub-100ns latency
- 300W TDP
- Dedicated core per program thread for consistent performance under sustained load
Deployment support:
- Air-cooled 1U servers with up to 8,160 cores per rack
- Liquid-cooled systems with more than 45,000 cores per rack
Arm stated that the AGI CPU can deliver more than 2x performance per rack compared to x86 CPUs and may enable up to $10 billion in CAPEX savings per gigawatt of AI data center capacity.
Meta collaboration and use case
Meta will co-develop multiple generations of the Arm AGI CPU and use it to support its AI infrastructure. The company said its data centers are increasingly exceeding the capabilities of traditional CPUs as it builds systems for AI training and inference.
The Arm AGI CPU will also work alongside Meta’s custom silicon, the Meta Training and Inference Accelerator (MTIA), to improve coordination in large-scale AI deployments.
Meta added that the CPU will be used to support its applications and broader AI initiatives, and that it plans to release board and rack designs for the platform through the Open Compute Project later this year.
Ecosystem and industry support
Arm confirmed that the AGI CPU is supported by multiple partners across the ecosystem, including:
- OpenAI
- Cloudflare
- SAP
- SK Telecom
The company is also working with OEMs and ODMs such as ASRock Rack, Lenovo, Quanta Computer, and Supermicro. Early systems are currently available, with broader availability expected in the second half of the year.
In addition, more than 50 companies are supporting the expansion of the Arm compute platform into silicon, including Google, Microsoft, NVIDIA, Samsung, and TSMC.
Speaking on the announcement, Rene Haas, CEO, Arm, said:
AI has fundamentally redefined how computing is built and deployed. Agentic computing is accelerating that shift. 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.
Commenting on the partnership, Santosh Janardhan, Head of Infrastructure, Meta, said:
Delivering AI experiences at global scale demands a robust and adaptable portfolio of custom silicon solutions, purpose-built to accelerate AI workloads and enhance 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.