The landscape of enterprise computing is shifting as AI agents replace traditional, human-led chatbot interactions. During the Advancing AI 2026 keynote, leadership from AMD and Cisco highlighted a major transition: for the first time, global AI compute capacity used for inference now outweighs the capacity used for training models. This change is driven by the rise of agentic AI that reasons, calls tools, and accesses data repeatedly to solve complex problems.
Unlike traditional queries that result in a single response, agents operate continuously. This shift creates a persistent demand for infrastructure and bandwidth across the cloud, private data centers, and local devices. AMD CEO Dr. Lisa Su noted that this transition demands a balanced hardware approach. While GPUs handle the intensive reasoning, CPUs remain vital for orchestrating every step of the agentic workflow. To meet this, AMD is deploying rack-scale solutions like the Helios platform, which integrates high-speed networking with advanced computing hardware.
As AI agents move closer to employees, the physical hardware at the desk side is becoming an intelligence node within the enterprise. Cisco President Jeetu Patel emphasized that deploying these local systems requires strict governance and management. The collaboration between AMD and Cisco focuses on providing a unified control plane to monitor agent behavior, enforce security policies, and manage computing resources across a distributed environment.
This evolution necessitates a move from simple inference to comprehensive orchestration. By utilizing an isolated agent sandbox and intelligent routing, enterprises can maintain control even as their AI deployments scale. AMD remains committed to an open ecosystem approach, partnering with organizations like Hugging Face to ensure developers can build and test models locally before deploying them across the wider infrastructure. This move toward distributed, managed AI marks a new chapter for corporate IT strategy.