What changed
Agentic AI is changing the business development strategies of chipmakers and server builders, shifting focus from graphics processing units (GPUs) to central processing units (CPUs) that can operate within data centers’ power budgets.
Scott Tease, vice president of business development for high-performance computing and sovereign AI at AMD, noted that customers who added more GPUs to a single CPU found their accelerators were often idle due to data starvation.
Tease emphasized that agentic AI is not just a GPU concern, as it requires high-speed processing and concurrency, making CPUs a critical component.
What this means in practice
Dell and AMD are collaborating on rack-scale AI platforms, with power efficiency becoming a key priority. Dell’s PowerEdge R9825, built on AMD’s sixth-generation EPYC 9006 processors, is optimized for power and cooling efficiency.
According to David Schmidt, vice president of PowerEdge product management at Dell, this server can replace 13 legacy machines while reducing cooling power needs by 80 percent.
The reclaimed power is expected to support agentic workloads, where agents can scale compute jobs from a single researcher’s ideas to thousands of tasks.
Limitations and availability
The shift to CPU-centric agentic AI is still in development, with power constraints remaining a binding factor for data center expansion.
Dell’s PowerEdge R9825 is currently available, but the full impact of agentic AI on CPU demand and power management will depend on how these systems are adopted in enterprise environments.
The supplied announcement does not specify exact power consumption figures for the EPYC 9006 processors or the exact number of compute jobs that can be scaled through agentic workflows.
