AWS is telling engineers to shut down idle EC2 servers and trim internal computing use as AI demand tightens CPU and GPU capacity, The Information reported Friday. The strain is showing up inside Amazon’s own cloud unit: some AWS engineers are waiting days – not hours – to get CPU server capacity, after teams were told to decommission idle Amazon Elastic Compute Cloud (EC2) instances and cut compute use. The report links the squeeze to AI-driven infrastructure limits that are hard to fix quickly: not enough data center space, plus bottlenecks in key components like memory chips. AWS said it’s still meeting most customer demand and that its “be efficient, shut down what you’re not using” guidance isn’t new. Still, an internal push like this suggests the business is treating spare capacity as scarce. Markets largely shrugged: Amazon shares closed up 1.69% to $276.85 on Friday. When a cloud provider runs with less slack, the first impact is often hidden in the fine print: tighter availability for certain machine types, longer waits, and less generous discounts on big contracts, rather than a clean jump in published list prices. That can shift negotiating leverage toward AWS in enterprise renewals and keep revenue per unit of compute relatively steady. The catch is that the same constraints – CPUs, GPUs, components, and physical space – can also cap how fast AWS can add new workloads in the near term.