Cisco and AMD have expanded their partnership with a new suite of hardware and security software aimed at assisting enterprise customers in protecting, deploying, and managing distributed artificial intelligence resources. The announcement was made during AMD’s Advancing AI event this week, where Cisco president and chief product officer Jeetu Patel discussed the future of AI inference.
Patel noted that inference will become widely distributed, necessitating an architectural stack of software and tools that the two companies are jointly developing.
The joint architecture integrates AMD’s compact, high-performance Ryzen AI Halo hardware with a range of Cisco technologies covering networking, observability, governance, and security. According to a blog post by Cisco senior director Yash Sheth, the foundation of this platform is the AMD Ryzen AI Halo hardware, which provides an isolated agent sandbox and services for local-first inference.
These services include model routing and token limits via AMD’s Semantic Router, alongside local inference capabilities on Lemonade.
AMD states that Ryzen AI Halo is designed to support local AI inference on AI PCs by utilizing its CPU, GPU, and XDNA neural processing unit. The company emphasizes that a resilient AI platform must continue delivering useful services even when connectivity is limited, models change, or workloads shift.
Cisco wraps this platform in a secure harness that includes Splunk Agent Observability and Splunk Infrastructure Monitoring. These tools provide full-stack observability, tracking agent behavior, tokenomics, and compute operations.
Additional Cisco technologies in the package include AI Defense for model and agent security, DefenseClaw for security policy enforcement to ensure guardrails are applied directly on-device, and Cisco Cloud Control for unified policy management. Sheth wrote that every AI node must be treated as a secure, managed node in the enterprise network to make deskside and local AI computing work at scale.
Sheth explained that the need for token efficiency and data sovereignty is driving a new class of computing called deskside computing. This approach places AI agents directly with users and teams, moving inference to a hybrid architecture with thousands of ambient deskside agents. AMD stated that as agentic AI moves from experimentation to real enterprise workflows, organizations require more than powerful endpoints.
AI agents can run continuously and act on enterprise data, creating new requirements for network infrastructure, tokenomics, agent behavior, and security.