Powering Agentic AI with the Arm AGI CPU Software Ecosystem

Arm introduces Arm AGI CPU, a new class of Neoverse-based processors designed to meet the demands of agentic AI, which requires orchestration, tool execution, and multi-step workflows beyond traditional model inference. The processors aim to deliver performance, scalability, and power efficiency for highly concurrent agentic workloads, addressing the evolving infrastructure needs of autonomous AI systems. At the OCP Global Summit, Arm presented a new CPU evaluation methodology alongside partners Fujitsu, Meta, and OpenAI, with results showing up to 2x performance per rack compared to leading x86 server processors.

To support agentic AI at scale, Arm emphasizes the importance of a robust software ecosystem spanning agent frameworks, orchestration, data infrastructure, networking, and application services. The company highlights its mature, production-proven software foundation built through years of collaboration with open-source communities, software vendors, and cloud providers. Arm AGI CPU leverages Arm64 support across cloud-native and AI software, with Neoverse-based CPUs already deployed at scale on leading cloud platforms, enabling developers to build and scale agentic AI applications efficiently.

Arm’s agentic AI software stack includes frameworks like LangGraph and LangChain, tool execution environments such as Firecracker, and data infrastructure tools like Redis and Elasticsearch. Developers can access enablement resources through the Arm Software Ecosystem Dashboard, which provides supported versions and practical guidance for building agentic AI applications on Arm64. The ecosystem supports diverse workloads, from coding agents requiring sandboxing to enterprise workflows dependent on API interactions and database transactions.

Arm demonstrates the practical application of its ecosystem with a multi-agent customer-service system using LangGraph and LangChain, where a small language model routes requests to specialized agents handling tasks like scheduling meetings and processing refunds. The company also offers learning resources, including tutorials on llama.cpp and vLLM, and collaborates with partners like AWS to provide platform-specific skills for optimizing AI applications on Arm-based infrastructure.