Amazon Web Services has told engineers to conserve computing resources and shut down idle EC2 instances as demand for cloud capacity tightens, with pressure now extending from AI chips to traditional CPU servers. Engineers who once received CPU server resources within hours are now reporting waits of several days, according to The Information, an unusual delay that suggests AWS is working to preserve headroom for paying customers. AWS management gathered engineers in May and said teams needed to do everything possible to conserve compute so the company's core EC2 business can continue meeting customer demand. The effort includes reducing internal resource footprints later this year and reclaiming underused development instances for external customers. AWS said it can still meet the needs of the vast majority of internal and external customers and that it is working with teams to keep EC2 utilization efficient. The strain appears linked to the rapid spread of agentic AI, which increases ongoing demand for general-purpose compute as AI agents run tasks and as companies prepare data for model training. Jing Xie, Co-founder and Managing Director of Elendil Labs, said client per-capita IT spending has doubled because of large-scale AI agent use. Chipmakers are also describing a shift toward heavier CPU use in AI inference: Intel CEO Lip-Bu Tan said in April that the CPU-to-GPU ratio was 1 to 4, while Intel CFO David Zinsner said in July that it had climbed to nearly 1 to 1. AWS said its employee resource guidance has not changed because of tight memory-chip supply, but warning signs are emerging in the Spot Instances market, where discounted surplus capacity has become harder to secure in large batches. Analysts say that could point to a narrowing gap between AWS supply and demand and, for enterprises that rely on spot capacity for large-scale workloads, possible upward pressure on cloud costs. Similar allocation pressures have surfaced elsewhere in big tech, including at Google and Microsoft, as cloud providers try to balance internal AI needs with external customer demand.