Bridging Agentic AI to System Reality

Session Overview

As AI systems evolve from single-turn chatbots to autonomous agents, the core infrastructure challenge is shifting from model execution alone to how agents remember, retrieve, share, and act on context over time. A memory-centric agentic AI service is an AI platform designed around this shift.

In this talk, I will discuss why memory is becoming a central resource in agentic AI workloads. Modern agents need to maintain long task histories, reuse prior context, coordinate across multiple sub-agents, access tools, and preserve user- or service-specific knowledge. This creates new challenges in inference cost, latency, KV cache management, long-term memory design, observability, reliability, and service-level optimization.

I will also explain how memory-centric AI infrastructure differs from conventional GPU-centric serving systems. Instead of treating each request as an isolated inference call, a memory-centric platform manages context, state, and computation as shared resources across agent workflows. This opens new opportunities for inference optimization, agent orchestration, hardware-software co-design, and AI service differentiation.

The talk will conclude with a discussion of how these ideas can shape next-generation AI services, AI data centers, and sovereign AI infrastructure.