AI’s power constraint is reshaping the chip supply chain

Power availability is no longer a data-center issue alone.

Today, Reuters reported Morgan Stanley expects the worsening U.S. power shortage to affect the AI chip ecosystem unevenly.

NVIDIA and Broadcom remain relatively protected due to:

• Strong visibility into chip deployments
• Geographic flexibility
• Large customer commitments
• Coordination across data centers, suppliers, and power providers

The risk increases for suppliers of:

• Memory
• Optical networking components
• Power-management chips
• Analog semiconductors

If power delays a data center, GPUs often move to another project or location. Supporting components face greater exposure to postponed orders and excess inventory.

The infrastructure evidence supports this concern.

The PJM region faces a power-supply shortfall exceeding 6,800 megawatts as demand grows faster than new generation connects. A separate analysis found only 5 of 16 gigawatts announced for 2026 delivery were under construction.

My takeawayAI infrastructure strategy must connect compute, memory, networking, power, geography, and deployment timing.

A component shortage limits one part of the architecture.

A power shortage delays the entire system.

Is your AI infrastructure plan based on hardware availability, or on where the hardware will receive power?