Karine GourdonKeller

Are we building AI infrastructure for the past or the future? For years, "hyperscale" meant massive, centralized data centers. That worked for training models. But training was just the prologue. The agentic era changes the math entirely. Unlike a simple chatbot, an AI agent might run dozens of iterative inference loops to complete a single task. If every step requires a 200ms round-trip to a distant mega-region, compounding latency and egress costs destroy the business case. According to @IDC research, by 2029: 🔷 1B+ active AI agents will execute 217B+ actions daily. 🔷 Inference workloads will consume 3.7 teratokens every single day. Centralized setups can't sustain this. True scale is shifting from mass to velocity and proximity. The metric of the next decade isn't how much compute you can concentrate in one place—it’s how effectively you can distribute it. I dive into this architectural shift in this guest blog for @Akamai akamai.com/blog/cloud/new-de…