Arm Emerges as Standard Platform for Agentic AI
IT DAILY ·
✦ AI Summary
Arm announced that it had been chosen as a core platform for Agentic AI environments.
According to IDC analysis, Arm-based rack-scale servers are surpassing traditional x86 servers and gaining an edge in the accelerated computing server market.
Arm said it supports both agent execution and security and control areas through an ecosystem spanning CPU and DPU.
Arm announced on the 2nd that it had been chosen as a core platform for Agentic AI environments. Arm said these environments require both high-performance computing and independent security controls.
According to IDC analysis, Arm-based rack-scale servers are becoming the main platform in the server market for accelerated computing. Arm-based rack-scale servers are surpassing traditional x86 servers and gaining an edge.
The shift has been attributed to the widespread adoption of Arm among major hyperscalers. Agentic AI expansion lies behind this structural change.
Agents do more than generate one-off responses; they continuously handle system runtime, orchestration and external tool control. As a result, host computing capabilities for stable agent operation and security computing capabilities for blocking and isolating unexpected risky behavior are emerging as essential infrastructure requirements.
Arm presented a framework that supports both areas through an ecosystem spanning CPU and DPU. In the agent execution layer, it is expanding the Arm software architecture ecosystem based on big tech’s own chips. Chips mentioned as targets included AWS's Graviton, Google's Axion, Microsoft's Cobalt and NVIDIA's Vera.
On the execution side, NVIDIA is building runtime boundaries through OpenShell to restrict agent access across Arm CPU environments. Arm said this explains a trend in which the agent execution layer is supported as a CPU-centric ecosystem.
The security and control layer is handled by Arm-based infrastructure processors, DPU. A representative example is NVIDIA's Open Agent Safety Platform. The platform runs in an NVIDIA BlueField-4 DPU environment equipped with 64 Arm Neoverse V2 cores, forming an independent control layer completely separated from the host CPU.
OpenShell links with NVIDIA Sentry on BlueField to detect anomalous agent behavior in real time. The agent is then immediately isolated, to the point that it is separated from the host and the network.
In this process, the Arm CPU handles agent execution, while the Arm DPU independently monitors and protects the agent, creating a two-layer safety net. Arm said it plans to contribute to strengthening Agentic AI safety in line with market trends.
Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=241987
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.