Enterprise AI is no longer confined to data centers. As artificial intelligence inference shifts from cloud servers to individual employee workstations, companies face an entirely new security and management challenge. Cisco and AMD have announced a joint architecture designed to help large organizations protect, deploy, and manage AI systems running directly on local devices, marking a significant shift in how enterprises think about AI infrastructure.

What Is Driving AI Away From the Cloud?

For years, AI inference (the process of running trained models to generate predictions or responses) happened almost exclusively in cloud data centers. But that model is changing rapidly. According to AMD CEO Lisa Su's keynote at the company's Advancing AI event, inference now accounts for 60% of global AI compute capacity in 2026, and much of that workload is moving to edge devices and personal computers rather than remaining in the cloud.

The reasons are practical. Running AI locally reduces latency, meaning responses happen faster. It keeps sensitive company data closer to users instead of sending it to remote servers. And it reduces dependence on constant cloud connectivity. For enterprises deploying agentic AI (AI systems that can act autonomously on behalf of users), these benefits compound quickly. As Cisco's senior director of engineering and research Yash Sheth explained, the shift creates new operating challenges that traditional cloud-centric infrastructure cannot solve.

How Are Cisco and AMD Addressing Enterprise AI Security?

The partnership combines AMD's Ryzen AI Halo hardware with Cisco's enterprise networking, security, and monitoring tools. The Ryzen AI Halo is designed to support local AI inference on an AI PC using its CPU (central processing unit), GPU (graphics processing unit), and XDNA neural processing unit (NPU), which is a specialized chip designed specifically for AI workloads.

Cisco wraps this hardware in a comprehensive security and observability layer that includes:

• Full-Stack Observability: Splunk Agent Observability and Splunk Infrastructure Monitoring track agent behavior, token usage (a measure of how much AI computation is being consumed), and compute operations across all local AI nodes
• Model and Agent Security: Cisco AI Defense protects the AI models themselves from tampering or unauthorized access
• On-Device Policy Enforcement: DefenseClaw enforces security guardrails directly on the device, within the agent harness, rather than relying on cloud-based policy servers
• Unified Control: Cisco Cloud Control provides IT administrators with a single dashboard to manage policies and control across thousands of distributed AI agents

"To make deskside and local AI computing work at enterprise scale, every AI node must be treated as a secure, managed node in the enterprise network," stated Yash Sheth, senior director of engineering and research at Cisco.

Yash Sheth, Senior Director of Engineering and Research, Cisco

Why Is This Architecture Different From Previous Approaches?

The key innovation is treating each employee's local AI system as a managed, monitored endpoint rather than an isolated device. Previous generations of edge AI focused on performance or efficiency in isolation. This partnership explicitly addresses what happens when thousands of AI agents run continuously across an enterprise, each potentially acting on sensitive data and making autonomous decisions.

The architecture also acknowledges that AI inference is becoming a continuous, always-on workload rather than a one-time computation. Traditional cloud-based AI systems process requests on demand. But agentic AI systems can run 24/7, continuously analyzing data and taking actions. That permanence creates new requirements for network infrastructure, token efficiency (ensuring AI systems don't consume excessive computational resources), agent behavior monitoring, and security.

What Do the Market Projections Tell Us About AI's Future?

The timing of this partnership reflects broader market trends. AMD's Lisa Su cited several striking statistics during the Advancing AI event that underscore why enterprises are investing in edge AI infrastructure now:

• AI Accelerator Market Growth: The AI accelerator market is projected to reach $1.4 trillion by 2030, nearly tripling previous forecasts, with GPUs expected to dominate but CPUs gaining new growth vectors due to agentic AI adoption
• Server CPU Expansion: The server CPU market is forecasted to grow over 50% to $200 billion by 2030, fueled by rapid agentic AI adoption and the need for massive CPU infrastructure
• Industry-Wide Adoption: AI adoption is accelerating across all industries, with agentic AI driving a surge in compute demand and shifting workloads from training to inference

How Can Enterprises Prepare for Distributed AI Deployment?

Organizations planning to deploy local AI systems should consider several practical steps to ensure security and manageability at scale:

• Assess Your Infrastructure: Evaluate whether your current network and security tools can monitor and control thousands of distributed AI endpoints, or whether you need new observability and governance platforms designed specifically for edge AI
• Plan for Data Sovereignty: Identify which data must remain local for compliance or sensitivity reasons, and design your AI deployment strategy around keeping that data on-device rather than transmitting it to cloud servers
• Establish Token Budgets: Set limits on how much computational resources individual AI agents can consume, similar to how organizations manage cloud spending, to prevent runaway costs and ensure fair resource allocation
• Design for Resilience: Build systems that continue delivering useful AI services even when cloud connectivity is limited or models need to change, ensuring your AI infrastructure doesn't fail when network conditions degrade

The Cisco and AMD partnership signals that enterprise AI is entering a new phase. Rather than debating whether AI should run locally or in the cloud, organizations are now asking how to securely manage thousands of local AI agents while maintaining visibility and control. That shift from "where should AI run?" to "how do we govern AI at scale?" represents a maturation of enterprise AI strategy.

As agentic AI moves from experimentation to real enterprise workflows, the infrastructure required to deploy it responsibly is becoming as important as the AI models themselves. Cisco and AMD's joint architecture addresses that gap by pairing high-performance local AI compute with the observability, governance, and control infrastructure enterprises need to deploy AI at scale without sacrificing security or data privacy.

" }