A 32-gigawatt power shortfall could delay America’s AI data-center boom, but Nvidia Corp. (NASDAQ:NVDA) and Broadcom Inc. (NASDAQ:AVGO) may be surprisingly well protected, according to Morgan Stanley.

Morgan Stanley estimates U.S. data-center developers face a 34% net power shortfall through 2028, equivalent to 32 gigawatts, even after accounting for workarounds such as fuel cells and power generated directly at data-center sites, according to Reuters.

The bank does not see the bottleneck putting Nvidia or Broadcom’s 2027 forecasts at risk. Instead, some of the pain could move further down the semiconductor supply chain.

Why Nvidia and Broadcom May Be Protected

Morgan Stanley pointed to the companies’ visibility into where chips are being deployed, geographic expansion and coordination with data centers and the power industry. That could let Nvidia send scarce GPUs to projects that have power if one facility is delayed, while suppliers tied to that project may have less flexibility.

The bank has separately argued that Nvidia’s advantage increasingly comes down to computing output per gigawatt. If electricity becomes the constraint, customers have more reason to squeeze as much computing power as possible from every watt.

Nvidia CEO Jensen Huang has described the bottleneck in similar terms, saying AI developers now need to “secure the land, power, and shell” before filling data centers with computing equipment. He said that process can take two to three years.

Nvidia is already backing that infrastructure, providing credit support for 4.25 GW at SB Energy’s Ohio campus and investing $1.5 billion in the developer.

A Memory Shortage Meets a Power Shortage

Morgan Stanley said delayed deployments could force customers to push out or cancel deliveries of memory, optical, power-management and analog components.

The bank did not specifically name Micron Technology Inc. (NASDAQ:MU), but its warning comes as the memory industry is telling investors a very different story.

Micron says more than 75% of its 2027 output is already committed, while investor Ross Gerber said the AI-driven memory shortage could last a decade.

The two views are not necessarily contradictory. Morgan Stanley is warning about when memory gets deployed, not whether AI will ultimately need it.

A data center delayed six months may still require the same memory, but customers could push out deliveries while they wait for electricity, creating inventory disruption even during a broader memory shortage.

Prediction traders aren’t betting on the power crunch breaking the AI boom yet. Polymarket puts the chance of an AI-industry downturn at about 7% by year-end and 19% by June 30, 2027.

Goldman Sachs has said political opposition has barely slowed near-term U.S. data-center construction. Morgan Stanley’s analysis suggests electricity itself may prove the harder constraint.

Image: Shutterstock