AI at the edge isn’t necessarily a threat to centralized data centers — it’s a shift toward hybrid computing. Routine, low-latency and privacy-sensitive workloads can move to phones, PCs, vehicles and enterprise devices, reducing some cloud inference and bandwidth demand. But frontier-model training, complex reasoning, agentic workloads, massive context and advanced inference still require centralized compute. The bigger picture: AI compute is being distributed, not eliminated. Edge AI can actually act as a pressure valve for power- and grid-constrained data centers, allowing scarce centralized capacity to focus on increasingly demanding workloads. Cloud + Edge > Cloud vs. Edge. I don’t view edge AI as an immediate existential threat to hyperscale data-center demand. I view it as an evolution of the architecture. The cloud will increasingly handle the workloads that require enormous amounts of compute. The edge will handle workloads where latency, privacy, bandwidth and cost make local processing more efficient. And the boundary between the two will continue to move. The long-term AI infrastructure story therefore may not be about cloud versus edge. It may be about cloud + edge + increasingly intelligent distribution of compute. The winners may ultimately be the companies enabling that entire transition — from silicon and memory to networking, data centers, power and the intelligent devices sitting at the edge