Morgan Stanley Says Nvidia and Broadcom Protected From Data-Center Power Crunch
Morgan Stanley has concluded that Nvidia ($NVDA) and Broadcom ($AVGO) are well-positioned to weather the growing power challenges facing U.S. data centers. In a Monday note, the bank highlighted that these two chipmakers have structural advantages, such as the ability to track product installations and collaborate closely with data-center operators and power-sector partners. These capabilities provide a buffer that smaller suppliers lack, allowing Nvidia, for instance, to redirect chips to projects with secured power if others face delays.
The bank's report last month estimated a net power shortfall of 34% for U.S. data-center developers through 2028, equating to around 32 gigawatts, even after accounting for behind-the-meter generation and fuel cells. Despite this, Morgan Stanley does not believe these bottlenecks will jeopardize Nvidia's or Broadcom's 2027 forecasts. The analysis also noted that customers under electricity pressure are likely to prioritize the most computationally efficient hardware, further sharpening Nvidia's competitive edge.
However, the bank warned that a slowdown in AI buildout could disrupt suppliers of memory, optical, and related chip components, potentially leading to order delays or cancellations. Memory, optics, power-management, and analog components were identified as the product categories with the highest inventory risk in such a scenario. Nvidia CEO Jensen Huang previously emphasized that data-center developers must secure real estate, electricity, and building structures before installing any computing gear, a process that typically takes two to three years.
The power crunch is part of a larger strain on U.S. electricity infrastructure, driven by the deployment of gigawatt-scale AI data centers. Hyperscale campuses running Nvidia's AI training chips can consume 120 to 140 kilowatts per rack, and the transmission infrastructure needed to power these facilities takes five to ten years to plan and energize, significantly longer than the 18 to 36 months required to build a data center. Goldman Sachs ($GS) has also noted mounting constraints on the U.S. data-center buildout but expects limited near-term impact from political opposition.