AMD has presented the direction of changes in data center infrastructure driven by the spread of agentic AI. The company explained that as AI expands from simple reasoning to autonomous task execution, the structural roles of CPUs and GPUs are also changing.

AMD announced on the 8th via its official blog that the importance of CPUs is increasing as AI systems evolve based on agentic AI.

Agentic AI is a form of AI technology in which AI independently reasons, performs tasks, and manages workflows. Unlike existing chatbot-centric structures, it requires data processing, task execution, and system coordination capabilities.

■ Changes in CPU-GPU Ratio and Structural Reorganization

Existing AI infrastructure was a GPU-centric structure. Generally, the ratio of CPUs to GPUs operated at a level of approximately 1:4 to 1:8.

According to AMD, in agentic AI environments, the ratio of CPUs to GPUs is showing a tendency to approach a 1:1 level. It was explained that in some environments, the proportion of CPU resources may be higher.

This is because AI systems perform complex tasks beyond simple computation, such as data processing, memory management, workflow orchestration, and tool execution.

AMD stated that these changes refer to a change in the configuration method, not the amount of computing resources.

■ Dedicated CPU Layer and Distributed AI Architecture

AMD explained that next-generation AI infrastructure is evolving into a structure based on the separation of roles between CPUs and GPUs. In this approach, GPUs handle high-density computations such as training and inference, while CPUs perform system control and data processing.

In particular, it was emphasized that AI systems require a dedicated CPU layer to handle large-scale agentic workloads. The CPU is responsible for managing interactions between agents, controlling data access, and coordinating task execution.

AMD predicted that future AI infrastructure will move away from a single-server-centric structure and transition to a distributed system.

The explanation is that this process requires a structure in which the CPU, GPU, network, and software are integrated to function as a single platform.

■ Emphasis on System Balance Over Performance

AMD stated that AI performance is determined by the overall system balance, not a single component.

If CPU resources are insufficient, GPU utilization decreases, and if network and data movement become bottlenecks, overall performance may degrade.

In addition, orchestration performance capable of handling large-scale concurrent tasks was also identified as an important factor.

AMD explained that in this environment, optimization of the entire infrastructure design, including CPUs and GPUs, is necessary.

■ Expanding Role of CPUs in Enterprise Environments

AMD stated that the scope of CPU utilization in enterprise AI environments is expanding based on its EPYC processors.

They explained that some AI inference tasks can be performed using only the CPU, and that the CPU plays a role in increasing accelerator utilization.

In addition, it was emphasized that in AI infrastructure investment, a combination of computing resources suited to various workloads is more important than specific processor performance.

AMD stated that companies need to prepare infrastructure designs early in anticipation of the widespread adoption of AI.