Most teams building agents on AWS assume GPU is where the compute goes. It's usually not. 

In this session, Dr. Ryan Ries and Marty Neuhard traced a real agentic AI request on AWS from start to finish and showed where the compute actually lands. 

Marty and Ryan were joined by Ray Prado at AMD about scaling agentic AI with AMD EPYC. These processors are crucial for the advancement and implementation of Agentic AI. EPYC’s performance and core count flexibility within the AWS instance portfolio enables whatever shape your agentic workload requires.  

Key Takeaways: 

Where your AI agents spend their compute 
The real breakdown of CPU vs. GPU work across an agentic AI system, and why it's not what most teams assume. 

How to size compute correctly for agentic AI on AWS 
A practical way to evaluate what each step of your workflow actually needs, instead of defaulting to GPU across the board. 

What this means for your AWS costs 
Where teams typically overprovision, and what to check before your next GPU spend decision.