Agentic AI Will Redefine Inference, Making Speed Irrelevant for Compute Infrastructure
Ben Thompson explains that as AI agents act autonomously, inference will no longer be a human‑in‑the‑loop task, so latency matters less and compute can be batch‑oriented. This shift will drive a move from GPU‑centric, low‑latency systems to heterogeneous, high‑throughput architectures, reshaping hardware demand for AI startups.