Nvidia and Broadcom are expected to remain relatively protected as a growing shortage of electricity creates new challenges for AI data centers in the United States.
The rapid growth of generative AI has pushed technology companies to build larger data centers filled with powerful chips and computing systems. However, these facilities need huge amounts of electricity, and power infrastructure is struggling to expand at the same speed.
Morgan Stanley estimates that U.S. data-center developers could face a 34% net power shortfall through 2028, equal to about 32 gigawatts, even after accounting for measures such as behind-the-meter power generation and fuel cells. The shortage could make it harder for some planned AI facilities to deploy computing equipment on schedule.
Despite these concerns, Nvidia and Broadcom are considered better positioned than many other semiconductor companies. Their visibility into chip deployments, geographic expansion and coordination across data centers, semiconductor suppliers and the power supply chain could give them more flexibility as power constraints develop.
Morgan Stanley does not currently expect the power bottlenecks to put Nvidia or Broadcom’s 2027 forecasts at risk. At the same time, the wider AI boom continues to drive major investment in computing infrastructure, advanced processors, networking equipment and custom chips.
The bigger risk may fall on other parts of the semiconductor supply chain. If enough power is not available to deploy new computing capacity, customers could delay deliveries or, in some cases, cancel orders.
Memory, optical, power-management and analog component suppliers could be particularly exposed. Delayed AI deployments could leave these companies with more inventory while customers wait for data-center capacity to become available.
The issue highlights an important shift in the AI industry. Building faster AI chips is no longer the only challenge. Companies also need enough electricity, grid connections, cooling systems and physical infrastructure to run increasingly powerful computing clusters.
The pressure is expected to grow as AI infrastructure expands. U.S. research estimates that data centers could account for around 11.8% of the country’s total electricity use by 2030, showing how closely future AI growth is becoming tied to energy availability.
Data-center developers are increasingly looking beyond traditional grid connections to secure enough electricity. Options include behind-the-meter generation, fuel cells, dedicated power projects and locating new facilities in regions where additional energy capacity is easier to obtain.
For Nvidia and Broadcom, their position in the AI infrastructure market may provide some protection from near-term power constraints. But the growing electricity crunch shows that the next stage of AI expansion will depend on more than faster chips. Power generation, grid connections, cooling and data-center construction are becoming just as important to how quickly new AI capacity can come online.
