Today, Arm is unveiling the Arm AGI CPU, a groundbreaking silicon solution built on the Arm Neoverse platform. This innovation is set to drive the future of AI infrastructure.
In a historic milestone for Arm, with our over 35 years of experience, we are now launching our own silicon products. This evolution extends the Arm Neoverse platform beyond just IP and Arm Compute Subsystems (CSS), offering customers versatile options for deploying Arm compute. They can choose to develop custom silicon, integrate platform-level solutions, or utilize Arm-designed processors. This shift highlights the rapid changes in AI infrastructure and the increasing demand for production-ready Arm platforms that can be deployed quickly and at scale.
The Emergence of Agentic AI Infrastructure
AI systems are increasingly designed to operate continuously on a global scale. Traditionally, human interaction acted as a bottleneck in computing, dictating the speed at which tasks could be completed. However, the rise of agentic AI removes this limitation, as software agents autonomously coordinate tasks, engage with various models, and make real-time decisions.
As AI systems function continuously and workloads become more complex, the CPU is crucial for modern infrastructure. It ensures that distributed AI systems operate efficiently at scale. In contemporary AI data centers, the CPU manages thousands of tasks, orchestrating accelerators, managing memory and storage, scheduling workloads, and transferring data across systems. With the advent of agentic AI, it also coordinates numerous agents on a large scale.
This transition places new demands on the CPU, necessitating an evolution in processor design.
The Arm Neoverse platform already supports many leading hyperscale and AI applications, including AWS Graviton, Google Axion, Microsoft Azure Cobalt, and NVIDIA Vera. As AI infrastructure expands globally, our partners are asking Arm for enhanced solutions. The Arm AGI CPU is our response to this demand.
Arm AGI CPU: Optimized for Rack-Scale Agentic Efficiency
Agentic AI workloads require sustained performance at an unprecedented scale. The Arm AGI CPU is engineered to provide exceptional performance for individual tasks while handling massive loads across thousands of cores simultaneously—all within the power and cooling capabilities of modern data centers.
Each component of the Arm AGI CPU, from its operating frequency to memory and I/O architecture, has been meticulously crafted to support high-performance agentic workloads in densely populated rack setups.
Arm’s reference server configuration includes a 1OU, 2-node design, featuring two chips with dedicated memory and I/O, totaling 272 cores per blade. These blades are designed to fully utilize a standard air-cooled 36kW rack, accommodating 30 blades that together deliver 8,160 cores. Additionally, Arm has teamed up with Supermicro to create a liquid-cooled 200kW system capable of housing 336 Arm AGI CPUs, which translates to over 45,000 cores.
In this optimal setup, the Arm AGI CPU can achieve more than double the performance per rack compared to the latest x86 systems*, thanks to the inherent advantages of the Arm architecture and a strategic alignment of system resources:
- The Arm AGI CPU boasts top-tier memory bandwidth, enabling more effective execution threads per rack. In contrast, x86 CPUs experience degradation as cores compete under sustained loads.
- Efficient, high-performance single-threaded Arm Neoverse V3 CPU cores surpass legacy architectures, as every Arm thread completes significantly more tasks.
- The combination of more usable threads and higher work-per-thread results in tremendous performance gains for each rack.
Early Momentum Across the AI Ecosystem
The Arm AGI CPU has already gained significant traction with partners eager to scale their agentic AI infrastructure. Upcoming deployments will encompass accelerator management, agentic orchestration, and the densification of services, applications, and tools necessary for expansive task execution. Additionally, enhanced networking and data plane compute will be integrated to support the AI data center.
Meta is our lead partner and customer, co-developing the Arm AGI CPU to optimize gigawatt-scale infrastructure for its suite of Meta applications, working alongside Meta’s custom MTIA accelerators. Other initial partners include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, SAP, and SK Telecom. Together, they are collaborating with Arm to accelerate AI-driven services across cloud, networking, and enterprise environments. Commercial systems are currently available from ASRockRack, Lenovo, and Supermicro.
To further streamline adoption, Arm is launching the Arm AGI CPU 1OU Dual Node Reference Server, engineered to meet the Open Compute Project (OCP) DC-MHS standard form factor. Arm intends to contribute this reference server design along with its firmware and additional resources, including system architecture specifications and diagnostic tools for all Arm-based systems. More details will be revealed at the upcoming OCP EMEA Summit.
A New Chapter for Arm Infrastructure
The launch of the Arm AGI CPU marks a significant new chapter in Arm’s journey in data center innovation and solidifies our position as a leader in computing advancements. As AI continues to reshape the industry landscape, Arm is dedicated to fostering progress across the ecosystem—supporting clients ranging from hyperscale cloud providers to emerging AI startups.
The Arm AGI CPU represents the first step in Arm’s new line of data center silicon products, now available for order. Future offerings are committed to achieving best-in-class performance, scalability, and efficiency, working in tandem with the Arm Neoverse CSS product roadmap to ensure all Arm data center clients advance together in terms of platform architecture and software compatibility.
As we embark on this new chapter, our mission remains steadfast: to provide the computational foundation that empowers innovation across various industries. Moreover, the ecosystem is in full support of our endeavor, with over 50 leading companies across hyperscale, cloud, silicon, memory, networking, software, and system design rallying to expand the Arm compute platform into silicon. With the Arm AGI CPU, we are not merely defining the architecture of the AI-native data center but actively building it.
Discover insights from our partners involved in the Arm AGI CPU deployment:
Cerebras
“At Cerebras, we design AI infrastructure specializing in ultra-fast, large-scale inference. As this becomes the predominant workload in AI, the significance of composed, high-performance systems increases. These systems require dedicated AI acceleration alongside efficient, scalable CPUs for optimal data movement, networking, and coordination. Introduce the Arm compute platform into AGI-class infrastructure, and we enhance the ecosystem for customers deploying AI at a global scale.” – Andrew Feldman, CEO, Cerebras
Cloudflare
“In our mission to help build a better Internet, Cloudflare requires infrastructure that scales effectively across our global network. The Arm AGI CPU delivers high-performance, energy-efficient computing tailored for the next generation of workloads.” – Stephanie Cohen, Chief Strategy Officer, Cloudflare
Meta
“To provide AI experiences on a global scale requires a robust and adaptable portfolio of custom silicon solutions, specifically engineered to enhance AI workloads and optimize performance across Meta’s platforms. We collaborated with Arm to develop the Arm AGI CPU, which offers an efficient computing platform that significantly boosts our data center performance density and lays the groundwork for our evolving AI systems.” – Santosh Janardhan, Head of Infrastructure, Meta
OpenAI
“OpenAI operates AI systems at a considerable scale, serving hundreds of millions daily through ChatGPT, businesses relying on our API, and developers utilizing tools like Codex. The Arm AGI CPU will be pivotal as we scale, enhancing the orchestration layer that manages extensive AI workloads while improving efficiency and bandwidth.” – Sachin Katti, Head of Industrial Compute at OpenAI
Positron
“Positron is dedicated to developing purpose-built inference accelerators that achieve remarkable token generation efficiency using standard memory. Arm consistently offers the industry’s most power-efficient computing platforms, making the Arm AGI CPU an ideal foundation for next-gen AI infrastructure. By merging Positron’s inference acceleration technology with the energy-efficient Arm AGI CPU, we see a significant opportunity for data center operators to deploy leading-edge AI models with greater performance per watt and cost.” – Mitesh Agrawal, CEO, Positron AI
Rebellions
“For high-performance AI systems, there’s a necessity for precise coordination between general-purpose computing and accelerator architectures. By integrating the Arm AGI CPU with Rebellions’ NPUs in new high-density server configurations, we’re delivering an energy-efficient, scalable platform finely tuned for AI inference applications at scale.” – Marshall Choy, Chief Business Officer, Rebellions
SAP
“SAP’s successful deployment of SAP HANA on Arm-based AWS Graviton highlights the robustness and effectiveness of the Arm ecosystem for enterprise workloads. The Arm AGI CPU broadens this opportunity, delivering scalable, efficient computing designed to support next-gen AI-driven business solutions.” – Stefan Bäuerle, Senior Vice President, Head of HANA & Persistency, SAP
SK Telecom
“SK Telecom is advancing into large-scale, full-stack AI inference data center infrastructure, utilizing the Arm AGI CPU alongside Rebellions’ AI accelerator chip. By combining our sovereign A.X foundation model with inference-optimized AI servers, we are equipped to deliver top-tier solutions globally while enhancing our AIDC competitiveness.” – Suk-geun (SG) Chung, CTO and Head of AI CIC, SK Telecom
Forward-Looking Statements
This blog post includes forward-looking statements related to Arm’s product roadmaps, projected performance, proposed contributions, and partner deployments. These statements are based on current expectations and are subject to risks and uncertainties that could lead to actual results diverging significantly. For a discussion of factors that could impact Arm’s outcomes, please refer to Arm’s filings with the U.S. Securities and Exchange Commission.
Performance claims are based on internal estimates comparing a fully populated rack of Arm AGI CPU-based servers against similarly configured x86-based servers 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 holders.
*Based on estimates
