Morgan Stanley estimates that U.S. data center developers face a net power shortfall of about 32 gigawatts through 2028, and it says the gap poses more risk to suppliers of memory and optical chips than to Nvidia or Broadcom, Reuters reported Monday in a story carried by Investing.com. Shares of both Nvidia and Broadcom were up about 2% at the time of the report.

What Happened

The brokerage put the shortfall at 34% of the power developers need through 2028. That figure already accounts for mitigation measures such as behind-the-meter generation, in which power is produced on site rather than drawn from the grid, and fuel cells.

Morgan Stanley said it does not expect the power bottleneck to hurt either Nvidia's or Broadcom's 2027 forecasts. Its reasoning is that both companies have visibility into where their chips will be installed and into plans for geographic expansion, and that they coordinate with data center operators, other semiconductor suppliers and the power supply chain.

RelatedFirmus Pulls A$44 Billion ASX Listing After Weak Investor Demand

The bank sees the most exposure among makers of memory, optical, power-management and analog chips. If new data center capacity cannot be energized on schedule, customers could postpone or cancel orders for those parts, leaving suppliers with inventory they cannot ship.

Why It Matters

For most of the AI buildout, chip supply was the main constraint on new capacity. The Morgan Stanley note reflects a shift in which electricity has become a limiting factor, and it argues that the effect of that shift is uneven across the semiconductor supply chain.

In Morgan Stanley's view, the companies that design the main AI accelerators and custom chips have enough insight into deployment plans to see delays coming. Makers of supporting components sit further from the final site and have less visibility into when a given facility will actually be powered. If a gap of 34% translates into delayed projects, component orders tied to those projects could slip with them.

Context

Morgan Stanley pointed to labor, power and political challenges facing data center construction. Goldman Sachs has also flagged mounting constraints on the U.S. buildout, according to the report.

Power availability has become a central issue for U.S. AI infrastructure as grid interconnection queues lengthen and rising electricity costs draw public attention. Developers have responded by adding their own gas turbines, fuel cells and other on-site generation, which is the mitigation Morgan Stanley factored into its estimate. Its conclusion is that those measures narrow the gap but do not close it.

What to Watch

Upcoming earnings from memory, optical-component and analog chipmakers will show whether customers are already delaying orders because finished capacity is waiting for power. For Nvidia and Broadcom, the test is whether 2027 guidance holds as Morgan Stanley expects. Any change in utility interconnection timelines or in the pace of on-site generation projects would also shift the size of the estimated gap.