AI infrastructure is accelerating its migration to the Arm architecture. Today, Arm technology is deployed across all major hyperscale cloud service providers' platforms. According to IDC data, the market size for rack-level servers based on the Arm architecture has now surpassed x86, becoming the mainstream platform in accelerated computing.
This shift becomes increasingly important as AI enters the era of agents. AI agents not only generate answers but also run continuously, call tools, access data, and autonomously execute tasks. As the scale of agents expands, two fundamental computing needs arise: the compute power required to run agents and the compute power needed to protect them.
Arm supports both types of computing needs. On one hand, high-performance CPUs can be used to run agent workloads; on the other hand, Arm-based infrastructure processors, such as Data Processing Units (DPUs), provide independent observability, isolation capabilities, policy enforcement, and security protection.
NVIDIA's recently launched NVIDIA Open Agent Safety Platform is a practical implementation of this architectural concept.
Arm Computing Platforms Drive Agent Operations
CPUs are the core of agent AI infrastructure. Agent runtimes, execution frameworks, orchestration systems, tool calls, and various applications all rely on CPUs. As agent capabilities strengthen, runtime durations extend, and concurrency scales up, the demand for CPUs will continue to grow.
Currently, Arm already provides compute power for most computations in agent and cloud workloads. Whether it's Arm AGI CPUs, Amazon Graviton, Google Axion, Microsoft Cobalt, or NVIDIA's custom-developed Vera CPU, although these products have unique designs, they all share Arm's unified software ecosystem. NVIDIA OpenShell extends agent secure execution capabilities to this broad Arm CPU ecosystem.
OpenShell establishes secure runtime boundaries for agents, controlling the resources they access and the operations they perform. This means that a unified approach can be adopted to ensure the secure operation of agents within the increasingly rich and diverse Arm cloud computing and AI infrastructure ecosystem.
Arm Computing Platforms Govern and Control Agent Behavior
In addition to running agents, agent infrastructure requires independent computing resources for observability, isolation, and policy control outside the agents' own runtime environments.
Arm-based DPUs and infrastructure processors provide this independent layer. By separating infrastructure services such as networking, monitoring, isolation, and security from the host CPU, they create independent control nodes that can always operate independently of the agents and applications they protect.
This architectural pattern has expanded to a broad Arm DPU ecosystem. Since these infrastructure processors are also built on Arm CPU technology, developers and infrastructure providers can build security capabilities based on a unified computing architecture.
NVIDIA BlueField embodies this pattern within the NVIDIA Open Agent Safety Platform. Powered by NVIDIA Grace with 64 Arm Neoverse V2 cores, BlueField-4 provides an independent infrastructure runtime environment outside the host. OpenShell works in conjunction with NVIDIA Sentry deployed on BlueField, extending policy enforcement capabilities from the agent runtime to this independent trust domain. Thus, infrastructure and cybersecurity controls can operate continuously, independent of the agents themselves.
As a real-time monitoring mechanism, NVIDIA Sentry adds another layer of protection, continuously monitoring agent behavior and isolating agents when unsafe behavior is detected. Together, these capabilities demonstrate a more comprehensive system-level security architecture: Arm CPUs are responsible for hosting and running agent workloads, while Arm-based DPUs can independently observe, govern, isolate, and protect them.
Arm CPUs host and run agent workloads, while Arm-based DPUs independently observe, govern, isolate, and protect them.
Safeguarding the Scalable Growth of Agent AI
As agent capabilities strengthen, with greater continuous operation capabilities and autonomy, the demand for compute power in agent AI will continue to grow. Meanwhile, security capabilities must scale in sync with compute power, without requiring trade-offs between security and performance.
The NVIDIA Open Agent Safety Platform demonstrates this direction: various computing platforms in the Arm ecosystem run agent workloads, while independent Arm-based DPUs assist in observation, isolation, and protection.
As agents expand from cloud infrastructure to edge devices and physical systems, these security boundaries must also extend to cover the various scenarios where agents operate and execute tasks. As Arm increasingly becomes a core component of AI infrastructure, Arm is driving the entire ecosystem to achieve synchronized scaling of performance and security.
The age of agents is being built on Arm, and Arm is committed to laying a solid security foundation for this era from the very beginning.