Today, Arm is announcing the Arm AGI CPU. It is a new class of production-ready silicon built on the Arm Neoverse platform and designed to power next-generation AI infrastructure.
For the first time in its more than 35-year history, Arm is offering its own silicon products. We extend the Arm Neoverse platform beyond IP and Arm Compute Subsystems (CSS), giving customers greater choice in how they deploy Arm compute, from building custom silicon to integrating platform-level solutions or deploying Arm-designed processors. This reflects both the rapid evolution of AI infrastructure and the growing demand from the ecosystem for production-ready Arm platforms that can be deployed at pace and scale.
The rise of agent AI infrastructure
AI systems are increasingly operating continuously on a global scale. Historically, humans have been the bottleneck of computing. The speed at which humans can interact with a system determines how quickly work is processed within the system. In the era of agent AI, that limitation disappears as software agents coordinate tasks, interact with multiple models, and make decisions in real time.
As AI systems continue to run and workloads become more complex, CPUs become the pacing element of modern infrastructure and are responsible for keeping distributed AI systems operating efficiently at scale. In modern AI datacenters, CPUs manage thousands of distributed tasks (tuning accelerators, managing memory and storage, scheduling workloads, moving data between systems), and agent AI is now used to orchestrate fanout across large numbers of agents.
This change places new demands on CPUs, which require processor evolution.
Arm Neoverse already powers many of today’s leading hyperscale and AI platforms, including AWS Graviton, Google Axion, Microsoft Azure Cobalt, and NVIDIA Vera. As AI infrastructure expands globally, partners across the ecosystem are looking to Arm to do more. Arm AGI CPUs were created to address this change.
Arm AGI CPU: Built for rack-scale agent efficiency
Agent-based AI workloads require sustained performance at scale. Arm AGI CPUs are designed to deliver high per-task performance at sustained loads across thousands of parallel cores, within the power and cooling constraints of modern data centers.
Every element of the Arm AGI CPU, from operating frequency to memory and I/O architecture, is designed to support massively parallel, high-performance agent workloads in dense rack deployments.
Arm’s reference server configuration is a 1OU, 2-node design that includes two chips with dedicated memory and I/O for a total of 272 cores per blade. These blades are designed to be fully mounted in a standard air-cooled 36kW rack, delivering a total of 8160 cores with 30 blades. Arm also partnered with Supermicro on a water-cooled 200kW design that can house 336 Arm AGI CPUs with over 45,000 cores.
In this configuration, Arm AGI CPUs can deliver more than twice the performance per rack compared to modern x86 systems*. This is achieved through fundamental benefits of the Arm architecture and careful matching of system resources to perform the following calculations:
- Arm AGI CPUs’ best-in-class memory bandwidth means more efficient threads of execution per rack. x86 CPUs perform poorly when cores compete under sustained load.
- High-performance, efficient, single-threaded Arm Neoverse V3 CPU cores deliver better performance than traditional architectures. Each Arm thread performs more work.
- With more available threads and more work per thread, you can significantly improve performance per rack.
Early momentum across the AI ecosystem
Arm AGI CPUs are already seeing strong commercial momentum thanks to collaborations with partners at the forefront of expanding agent AI infrastructure. Planned deployments span the densification of services, applications, and tools needed to scale out accelerator management, agent orchestration, and agent tasks, as well as networking and data plane computing enhancements to support AI data centers.
Meta is a key partner and customer of ours, co-developing Arm AGI CPUs to optimize gigawatt-scale infrastructure for the Meta family of apps and work with Meta’s own custom MTIA accelerators. Other launch partners include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom, each of which will collaborate with Arm on deployments of Arm AGI CPUs to accelerate AI-driven services across cloud, networking, and enterprise environments. Commercial systems are currently available to order from ASRockRack, Lenovo, and Supermicro.
To further accelerate deployment, Arm is introducing the Arm AGI CPU 1OU Dual Node Reference Server, an Open Compute Project (OCP) DC-MHS standard form factor server. In addition to this reference server design and supporting firmware, Arm will also provide further contributions such as system architecture specifications, debugging frameworks, and diagnostic and verification tools applicable to all Arm-based systems. Further details will be announced at the upcoming OCP EMEA Summit.
A new chapter in Arm infrastructure
The launch of Arm AGI CPUs represents a new chapter in Arm’s data center efforts and continued leadership in computing innovation. As AI reshapes industries, Arm remains committed to enabling progress across our ecosystem, from hyperscale cloud providers to AI startups, wherever our customers are.
Arm AGI CPUs are the first products in Arm’s new data center silicon product line and are available to order now. Successor products strive for best-in-class performance, scale, and efficiency. This will continue in parallel with the Arm Neoverse CSS product roadmap, with all Arm datacenter customers moving forward collaboratively on platform architecture and software compatibility.
As we enter this new chapter, our mission remains the same. It’s about providing the computing foundation that enables innovation across industries. And the ecosystem is fully supporting us. More than 50 leading companies across hyperscale, cloud, silicon, memory, networking, software, system design, and manufacturing support the extension of Arm computing platforms to silicon. With Arm AGI CPUs, we’re not just defining the architecture of the AI-native data center, we’re building it.
Learn more about Arm AGI CPU deployment partners.
cerebrum
“At Cerebras, we are building an AI infrastructure designed for ultra-fast, large-scale inference. As this becomes the dominant AI workload, configurable, high-performance systems are more important than ever. These systems include data movement, networking, and coordination at scale. “Extending the Arm computing platform to AGI-class infrastructure is a positive step for the ecosystem and for customers deploying AI on a global scale.” – Andrew Feldman, CEO of Cerebras
cloudflare
“To continue our mission of helping build a better internet, Cloudflare needs an infrastructure that can scale efficiently across our global network. Arm AGI CPUs deliver high-performance, energy-efficient computing designed for next-generation workloads.” – Stephanie Cohen, Chief Strategy Officer, Cloudflare
meta
“Delivering AI experiences at global scale requires a portfolio of robust, adaptable, custom silicon solutions purpose-built to accelerate AI workloads and optimize performance across Meta’s platforms. We collaborated with Arm to develop Arm AGI CPUs to significantly increase data center performance density and deploy an efficient compute platform that supports a multi-generation roadmap for evolving AI systems.” – Santosh Janardhan, Head of Meta Infrastructure
OpenAI
“OpenAI runs AI systems at scale. Hundreds of millions of people use ChatGPT every day, enterprises build on our APIs, and developers rely on tools like Codex. Arm AGI CPUs will play a critical role in our infrastructure as we scale, powering the orchestration layer that orchestrates large-scale AI workloads and improving overall system efficiency, performance, and bandwidth.” – Sachin Katti, Head of Industrial Computing, OpenAI
positron
“At Positron, we are focused on purpose-built inference accelerators that use commodity memory to deliver breakthrough token generation efficiencies. Arm consistently delivers the industry’s most power-efficient computing platforms, making Arm AGI CPUs a natural foundation for next-generation AI infrastructure. Positron’s inference acceleration technology combined with the energy-efficient Arm AGI CPU platform enables data center operators to deliver frontier AI with greater performance. We believe there is a strong opportunity to deploy the model at scale” per watt and dollar. ” – Mitesh Agrawal, Positron AI CEO
rebellion
“High-performance AI systems require tight coordination between general-purpose compute and accelerator architectures. Combining Arm AGI CPUs and Rebellions NPUs in a new high-density server configuration provides a scalable, energy-efficient platform optimized for large-scale AI inference workloads.” – Marshall Choi, Rebellions Chief Business Officer
SAP
“SAP’s successful deployment of SAP HANA on Arm-based AWS Graviton highlights the maturity and performance of the Arm ecosystem for enterprise workloads. Arm AGI CPUs expand that opportunity, delivering scalable and efficient compute designed to support the next generation of AI-powered business solutions.” – Stefan Bäuerle, Senior Vice President, Head of HANA and Persistence, SAP
SK Telecom
“SK Telecom is expanding into a large-scale, full-stack AI inference data center infrastructure that includes Arm AGI CPUs and Rebellions AI accelerator chips. By integrating our Sovereign AX foundation model and inference-optimized AI servers, we are ready to deliver it to the world while increasing AIDC’s competitiveness.” – Suk-geun (SG) Chung, CTO and Head of AI CIC, SK Telecom
Forward-looking statements
This blog post contains forward-looking statements regarding Arm’s product roadmap, future performance, planned contributions, and partner deployments. These statements are based on current expectations and are subject to risks and uncertainties that could cause actual results to differ materially. For a discussion of factors that may affect Arm’s results of operations, please refer to Arm’s filings with the Securities and Exchange Commission.
Performance claims are based on Arm internal estimates comparing a rack fully populated with Arm AGI CPU-based servers to an equivalent x86-based server configuration using industry standard workloads. Actual results may vary depending on system configuration, workload and other factors.
All product and company names are trademarks or registered trademarks of their respective owners.
*based on estimates
