Today, Arm officially launched the Arm AGI CPU - this is a brand-new mass-produced chip built on the Arm Neoverse platform, specifically designed to 

empower the next generation of artificial intelligence infrastructure. 

During the over 35-year development history of Arm, this is the first time we have launched our own self-developed chip product. This move will further 

expand the capabilities of the Arm Neoverse platform from IP and Arm computing subsystems (CSS), enabling customers to deploy it. 

The Arm computing architecture offers more options: customers can either independently develop customized chips, integrate platform-level solutions, 

or directly deploy processors designed by Arm. This layout aligns with the requirements of the artificial intelligence infrastructure. 

This rapid evolution trend also meets the growing demand of ecosystem partners for an Arm production-grade platform that can be rapidly scaled up 

for deployment. 

The rise of intelligent agent-based artificial intelligence infrastructure 

Nowadays, artificial intelligence systems are gradually achieving 24/7 continuous operation worldwide. In the past, humans were the bottleneck in the 

computing process - the efficiency of human-computer interaction determined the processing speed of the business within the system. And

In the era of intelligent agent-based artificial intelligence, this limitation has been completely broken: software intelligent agents can autonomously 

coordinate tasks, connect with multiple artificial intelligence models, and make real-time decisions. 

As the continuous operation of artificial intelligence systems and the increasing complexity of workloads continue, the CPU has become the core 

scheduling unit of modern infrastructure, bearing the heavy responsibility of ensuring the efficient and large-scale operation of distributed artificial 

intelligence systems. In

In modern artificial intelligence data centers, the CPU needs to manage thousands of distributed tasks, including coordinating accelerators, controlling 

memory and storage, scheduling workloads, and facilitating data flow across systems; in the context of agent-based artificial intelligence scenarios, 

The CPU also needs to coordinate the task distribution work for a large number of intelligent agents. 

This transformation imposed new requirements on the CPU and also spurred the iterative upgrading of processor technology. 

The Arm Neoverse architecture has become the technical foundation for many of the current top-level ultra-large-scale cloud computing and artificial 

intelligence platforms. Amazon Web Services' Graviton, Google's Axion, Microsoft's Azure Cobalt, and NVIDIA's Vera are all examples. 

Based on this architecture. As the artificial intelligence infrastructure expands globally on a large scale, ecosystem partners are all looking forward to 

Arm further leveraging its technological advantages. And the Arm AGI CPU is precisely designed to address this industry transformation. 

Arm AGI CPU: Designed for intelligent energy efficiency at the cabinet level 

The workload of intelligent agent-style artificial intelligence requires chips to maintain continuous high-performance output at an extremely large scale. 

The design intention of Arm AGI CPU is to enable thousands of computing cores to operate under the power and heat dissipation constraints of modern data centers. 

When running in parallel mode, it can still maintain excellent single-task performance under continuous load. 

From the operating frequency to the memory and input/output architecture, every design detail of the Arm AGI CPU has been meticulously crafted to 

meet the requirements of large-scale parallel and high-performance intelligent agent-style artificial intelligence workloads in high-density cabinet 

deployment scenarios. 

Core technological advantages 

✅ Intelligent Response Performance: Up to 136-core Arm Neoverse V3 processor, each core equipped with 2MB dedicated secondary cache, with a 

maximum clock speed of up to 3.7GHz 

✅ Input/output capabilities for modular artificial intelligence systems: 96 PCIe 6.0 channels, native support for CXL 3.0 (enabling functions such as 

memory expansion), equipped with AMBA CHI expansion links 

✅ Industry-leading Arm energy efficiency: Achieving unparalleled energy efficiency through advanced 3-nanometer manufacturing process, maximizing 

computing density, with a thermal design power of only 300 watts. 

✅ Dual-core design: The memory and input/output modules are integrated onto the same chip. The memory latency is less than 100 nanoseconds, 

and the memory access has been optimized for latency. 

✅ Exceptionally high memory performance: Single-core memory bandwidth is 6GB/s, the maximum memory capacity of a single chip is 6TB, supports 

DDR5-8800 specification, and the memory architecture is specifically optimized for computing scenarios. 

The reference server launched by Arm adopts a 1OU dual-node design. Each blade server integrates two Arm AGI CPUs, and is equipped with dedicated 

memory and input/output modules. The total number of cores reaches 272. This blade server can fully utilize its capabilities. 

The deployment is carried out in a standard 36-kilowatt air-cooled cabinet. With 30 blade servers, a total computing power of 8,160 cores can be achieved. 

Additionally, Arm has collaborated with Supermicro to develop a 200-kilowatt liquid-cooled cabinet solution, which can accommodate 336 chips. 

Arm AGI CPU has a total of over 45,000 cores. 

Under the aforementioned configuration, the single-cabinet performance of the Arm AGI CPU can exceed that of the latest x86 system by more than 

twice. This advantage stems from the inherent technical characteristics of the Arm architecture, as well as the precise matching of system resources with 

computing demands. Specifically 

It is reflected in three aspects: 

The Arm AGI CPU boasts industry-leading memory bandwidth, enabling a single cabinet to support a greater number of effective execution threads; 

while x86 CPUs, under continuous load, experience resource competition among cores, which leads to performance degradation. 

The single-core of the Arm Neoverse V3 processor combines high performance with high energy efficiency, performing better than traditional 

architectures. Each Arm thread can handle more tasks. 

More available threads, combined with the enhanced task processing capability of a single thread, have jointly contributed to a significant improvement 

in the performance of a single cabinet. 

The early positive feedback of the artificial intelligence ecosystem 

The Arm AGI CPU has received high recognition from the leading partners in the artificial intelligence ecosystem, and its commercial implementation 

progress is remarkable. The planning and deployment scenarios of the partners cover accelerator management, intelligent agent task orchestration, and 

the expansion of intelligent agent tasks. 

High-density services / applications / tools, as well as an enhanced network and data plane computing capability specifically designed for artificial 

intelligence data centers. 

**The Meta Universe** (Meta) is our primary partner and initial client. The two parties jointly developed the Arm AGI CPU, aiming to optimize the 

performance for the gigawatt-level infrastructure of Meta's entire range of applications, and to integrate it with Meta's own research and development. 

The MTIA accelerator is operating in a collaborative manner. Other initial partners include Cerebras, Cloudflare, F5, OpenAI, Positron, Rebellions, and SIA. 

PwC (SAP) and SK Telecom of South Korea. These companies have all engaged in in-depth cooperation with Arm, by deploying Arm AGI CPUs, to 

accelerate the implementation of AI-driven services in cloud, network and enterprise environments. Currently, Huoqing 

The companies of Erqi, Lianxi and Chaohui have initiated the commercial system ordering based on the Arm AGI CPU. 

To further accelerate the implementation of the product, Arm has simultaneously launched the Arm AGI CPU 1OU dual-node reference server. This 

product adopts the Open Compute Project (OCP) DC-MHS standard form. Arm plans to use this reference server 

The design plan and associated firmware will be made available for open-source contribution. Additionally, the system architecture specifications, 

debugging framework, as well as diagnostic and verification tools applicable to all Arm-based systems will also be made available. More details will 

be announced at the upcoming OCP conference in Europe. It was announced at the summit of Asia, the Middle East and Africa. 

A brand-new chapter of Arm's infrastructure business 

The release of Arm AGI CPU marks a new stage of development for Arm's data center business, and also demonstrates Arm's continuous leading position 

in the field of computing innovation. In the current era where artificial intelligence is reshaping the entire industry, Arm has always... 

Committed to promoting technological progress across the entire ecosystem, we provide tailored solutions to different types of clients, including 

large-scale cloud service providers and AI startups. 

Arm AGI CPU is the first product of Arm's new data center chip product line. It is now available for order. Arm has already planned the subsequent 

product development roadmap and will continue to strive for industry-leading performance, scalability and energy efficiency. 

At the same time, this product line will be advanced in parallel with the Arm Neoverse CSS product roadmap, ensuring that all Arm data center 

customers can achieve synchronous development in terms of platform architecture and software compatibility. 

Entering a new stage of development, Arm's mission remains unchanged: to build the foundation of computing power and empower the innovation 

of various industries. And this vision of Arm has also received strong support from the entire ecosystem: over 50 companies from large-scale computing, 

The leading enterprises in the fields of cloud computing, chips, memory, networks, software, system design and manufacturing are helping Arm's 

computing platform expand from IP to chip products. With the Arm AGI CPU, we not only define the native artificial intelligence... 

The architecture of the intelligent data center is not only a result of designing this brand-new infrastructure by ourselves. 

Partner's Message: The Practical Value of Arm AGI CPU 

Cerebras 

Syrup focuses on building ultra-high-speed and large-scale artificial intelligence infrastructure for reasoning, which is becoming the mainstream 

workload in the field of artificial intelligence. In this context, modular high-performance systems have become increasingly important - such systems 

It is necessary to have both specialized acceleration chips specifically designed for artificial intelligence, as well as efficient and scalable CPUs to scale 

up the coordination of data flow, network communication, and task scheduling. Arm has expanded its computing platform to the level of general 

artificial intelligence infrastructure. 

For both customers in the field of ecology and those involved in the global large-scale deployment of artificial intelligence, this represents a significant 

positive breakthrough. 

-- Andrew Feldman, CEO of Synopsys 

Cloudflare 

The mission of CloudFlare is to contribute to the creation of a better internet. This requires that our infrastructure be able to achieve efficient expansion 

across the global network. The high-performance and high-energy efficiency computing power provided by Arm AGI CPU is specifically designed for 

the next-generation workloads. 

It is precisely tailored to meet our needs. 

-- Stephanie Cohen, Chief Strategy Officer of Cloudfire 

Metaverse 

To provide an artificial intelligence experience globally, a powerful and flexible customized chip solution portfolio is required. These chips need to be 

specifically tailored to accelerate artificial intelligence workloads and optimize Meta's platform-wide performance. Jointly developing Arm AGI CPU 

with Arm, aiming to create an efficient computing platform, significantly enhancing the performance density of data centers, and providing cross-

generation technical support for our continuously evolving artificial intelligence systems. 

-- Santosh Janardhanan, Head of Meta's Infrastructure Department 

OpenAI

OpenAI operates large-scale artificial intelligence systems, with hundreds of millions of users using ChatGPT every day. Enterprises leverage our API 

for development, and developers rely on tools like Codex for innovation. As the business expands on a larger scale, 

The Arm AGI CPU will become an important component of our infrastructure, enhancing the orchestration layer capability for coordinating large-scale 

artificial intelligence workloads, and significantly improving the system's energy efficiency, performance and bandwidth. 

-- Sachin Katyi, Head of Industrial Computing Department at OpenAI 

Positron 

Positrons are dedicated to developing customized inference accelerators that can achieve breakthrough token generation efficiency through general-

purpose memory. Arm has always been able to launch the most energy-efficient computing platforms in the industry, which makes the Arm AGI CPU 

the next-generation human-like 

The natural foundation of artificial intelligence infrastructure. By combining positron's reasoning acceleration technology with the high-efficiency Arm 

AGI CPU platform, we expect to help data center operators achieve better performance per watt and per dollar. 

Carry out large-scale deployment of cutting-edge artificial intelligence models. 

-- Mitesh Agrawal, Chief Executive Officer of Positron AI 

Rebellions Control Technology 

High-performance artificial intelligence systems require close collaboration between general-purpose computing and accelerator architectures. By 

integrating the Arm AGI CPU with the Renwei Intelligent Control's neural network processor (NPU) into a brand-new high-density server configuration, 

we are Create a scalable and energy-efficient platform that has been specifically optimized for large-scale artificial intelligence inference workloads. 

-- Marshall Cai, Chief Commercial Officer of RuiLink Intelligent Control 

SAP 

SAP has successfully deployed SAP HANA on the Amazon Web Services Graviton platform based on the Arm architecture. This fully demonstrates the 

maturity and performance of the Arm ecosystem in enterprise-level workload scenarios. 

-- Stefan Baller, Senior Vice President and Head of the HANA and Persistence Technology Department at SAP 

SK Telecom of South Korea 

SK Telecom is building a large-scale, full-stack artificial intelligence inference data center infrastructure, which includes Arm AGI CPUs and Ruilian 

Zhikong artificial intelligence acceleration chips. We will also independently develop the A.X large model and integrate it with the optimized version. 

Combining the optimized artificial intelligence server with this solution is expected to not only bring it to the global market, but also significantly 

enhance the competitiveness of our artificial intelligence data center in the industry. 

-- Zheng Xigen, Chief Technology Officer of SK Telecom and Head of the Artificial Intelligence Innovation Center