Morgan Stanley's latest report indicates that the bottleneck for AI infrastructure in the US is shifting from "chip shortages" to "power shortages." The report projects a net electricity deficit of approximately 34% for US data centers between 2026 and 2028, equivalent to 32 GW. The report suggests that power shortages could delay chip deployment as planned, subsequently posing risks of order delays, cancellations, and inventory adjustments for downstream components in the supply chain, such as ASICs, memory, optical modules, and power management. NVIDIA's and Broadcom's 2027 performance forecasts are not substantially affected for now, given their higher GPU power output per unit and high visibility into deployment locations. Oracle's 1.3 GW "Project Lighthouse" in Wisconsin may see its full power supply delayed until October 2028, or even spring 2029, highlighting the practical hindrance of power insufficiency on AI hashrate deployment.