Nvidia and Broadcom appear relatively well positioned to withstand a growing power crunch affecting U.S. data centers, although delays to artificial intelligence infrastructure projects could put pressure on other parts of the semiconductor supply chain, Morgan Stanley said.
The investment bank estimates that U.S. data-center developers could face a 34% net power shortfall through 2028, equivalent to about 32 gigawatts, even after accounting for alternatives such as behind-the-meter power generation and fuel cells. The rapid expansion of generative AI has driven billions of dollars in data-center investment while sharply increasing demand for electricity.
Morgan Stanley said it does not currently expect the power constraints to put Nvidia or Broadcom’s 2027 forecasts at risk. Both companies have greater visibility into where their chips will be deployed and are closely coordinating with data-center operators, semiconductor suppliers and power providers as AI infrastructure expands.
However, the wider chip supply chain could face greater risks if power shortages delay new data centers. Customers may postpone or cancel chip deliveries if computing capacity cannot be deployed on schedule, potentially leaving suppliers with excess inventory. Memory chips, optical components, power-management products and analog semiconductors are among the segments Morgan Stanley considers most exposed.
The assessment highlights a growing challenge for the AI industry: chips may be available, but the electricity and infrastructure required to operate them may not be. Morgan Stanley and Goldman Sachs have both identified power, labor and other infrastructure constraints as potential obstacles to the pace of U.S. data-center expansion.
Leave a comment