NEW YORK — The artificial intelligence boom is running into a power problem: United States data-center developers face a net power shortfall of 34% through 2028, equivalent to about 32 gigawatts, Morgan Stanley has estimated, and while chip giants Nvidia and Broadcom look relatively insulated, makers of memory, optical and other secondary components are far more exposed.
Illustration: An AI data center connected by glowing power lines to electricity pylons and wind turbines, representing surging power demand from artificial intelligence.
What Happened
Morgan Stanley said on Monday, 5 October 2026, that Nvidia and Broadcom are relatively insulated from a worsening United States data-center power crunch, but that any resulting delay in artificial intelligence deployments could affect makers of memory, optical and other secondary chip components, according to Reuters. The brokerage's assessment, reported by Reuters, draws a sharp line between the companies that design the most powerful AI processors and the suppliers whose parts are needed to turn those processors into working data centers.
The scale of the problem is striking. Morgan Stanley estimated last month — in September 2026 — that United States data-center developers face a 34% net power shortfall through 2028, equivalent to 32 gigawatts, after accounting for measures such as behind-the-meter generation and fuel cells. Behind-the-meter generation means power produced on site, at the data center itself, rather than drawn from the public grid; fuel cells are one such on-site option. Even after those workarounds, the brokerage concluded, roughly a third of the power developers need is not yet secured.
The warning is not Morgan Stanley's alone. Goldman Sachs has also flagged mounting constraints on the United States data-center buildout, Reuters reported, although the two banks differ on emphasis: Goldman Sachs expects limited near-term impact from political pushback, while Morgan Stanley sees labour, power and political challenges ahead.
Background
The power crunch is a direct consequence of the AI investment boom. Generative AI systems require vast computing power, and that computing power requires electricity — not only to run the processors themselves, but to cool the buildings that house them. The AI boom has triggered billions of dollars in data-center investment, Reuters reported, as technology companies, cloud providers and specialist developers race to build the facilities that train and run large AI models.
That race is now colliding with the physical limits of the American power system. Power grids were not designed for the sudden arrival of facilities that can each consume as much electricity as a small city. Connecting a new data center to the grid can take years, as utilities assess capacity, upgrade transmission lines and navigate permitting processes. Developers have responded by seeking on-site generation, including fuel cells and other behind-the-meter solutions, but Morgan Stanley's estimate suggests those measures close only part of the gap.
Illustration: Computer chips and memory modules on a production line, representing the semiconductor supply chain strained by data-center delays.
The distinction Morgan Stanley draws between chipmakers matters for understanding who bears the risk. Nvidia and Broadcom design the flagship processors and networking chips at the heart of AI systems, and Morgan Stanley said it does not see the bottlenecks putting their 2027 forecasts at risk, citing their visibility into chip placement, geographic expansion and coordination between data centers, semiconductor suppliers and the power supply chain. In other words, the biggest names can see where their chips will go and can plan around constraints.
Secondary suppliers have less room to manoeuvre. Memory chips store the data that AI processors crunch; optical components carry data between machines at high speed; power-management and analogue components regulate the electricity flowing through the system. If chip capacity cannot be deployed — if a data center is built but cannot get enough power to switch on — customers could push out deliveries or cancel orders, Morgan Stanley said, with memory, optics, power-management and analogue components most exposed to inventory disruption. A cancelled order for a flagship processor is a lost sale; a cancelled order for a secondary component can leave a supplier holding stock it cannot easily sell elsewhere.
Why It Matters
The power crunch matters because it could become the binding constraint on the AI boom — the factor that decides how fast the technology actually spreads, regardless of how good the chips become. For years, the limiting factor was widely assumed to be chip supply: could factories produce enough advanced processors? Morgan Stanley's analysis suggests the next bottleneck may be the electricity grid, and that it is already visible in developers' planning through 2028.
It also matters for investors trying to judge which companies are genuinely exposed to an AI slowdown. The brokerage's message is that exposure is uneven: the headline chip designers may be shielded by their scale and planning visibility, while the suppliers further down the chain carry the inventory risk. That distinction could reshape how markets value different parts of the semiconductor industry, and it echoes a broader lesson of supply-chain shocks — that the weakest link, not the strongest, often determines the outcome.
There is a political dimension too. Data centers have become a focus of local and national debate, as communities weigh the jobs and investment they bring against their demands on power, water and land. Morgan Stanley flagged political challenges alongside labour and power constraints, and Goldman Sachs noted political pushback as a factor, even if it expects limited near-term impact. How governments handle grid connections, permitting and energy policy will therefore shape not only where data centers are built, but how quickly the AI buildout can proceed.
Finally, the crunch raises a question about energy strategy. Behind-the-meter generation and fuel cells help, but they are partial answers. Longer-term solutions — grid upgrades, new generation capacity, smarter siting of data centers near abundant power — take years to deliver. The gap between the speed of AI investment and the speed of energy infrastructure is, in Morgan Stanley's estimate, 32 gigawatts wide.
What Happens Next
What to watch first is whether deployment delays materialise in company results. If data-center developers begin pushing out or cancelling orders, the first signs should appear in the order books and inventory levels of memory, optical, power-management and analogue suppliers — the groups Morgan Stanley identified as most exposed. Nvidia's and Broadcom's 2027 forecasts, by contrast, are not expected by the brokerage to be put at risk by the bottlenecks, so any divergence between the headline chipmakers and their suppliers would be a test of the analysis.
Illustration: Electricity transmission towers beside a glowing data center at night, representing the energy demands of artificial intelligence infrastructure.
The second thing to watch is power. Developers, utilities and policymakers will be looking for ways to close the 34% shortfall: faster grid connections, more on-site generation, data centers sited where power is plentiful, and perhaps new rules on how large new facilities can be and where they can be built. Each of those paths involves trade-offs between speed, cost and local consent.
For now, Morgan Stanley's warning stands as a reality check on one of the decade's biggest investment stories. The AI boom is real, and the chips are getting faster — but the electricity to run them is not keeping pace, and the companies most likely to feel the squeeze first may not be the famous names at the top of the chain.
Sponsored: explore today's featured sponsored offers
FAQ
What is the AI power crunch?
It is the gap between the electricity that US data-center developers need and the power they can actually secure. Morgan Stanley estimated in September 2026 that developers face a 34% net power shortfall through 2028, equivalent to about 32 gigawatts, even after counting on-site generation and fuel cells.
Why are Nvidia and Broadcom less exposed?
Morgan Stanley said the two companies are relatively insulated and that it does not see the bottlenecks putting their 2027 forecasts at risk, citing their visibility into chip placement, geographic expansion and coordination across data centers, suppliers and the power supply chain.
Which suppliers are most at risk?
Makers of memory, optical, power-management and analogue components are most exposed to inventory disruption, according to Morgan Stanley, because customers could push out deliveries or cancel orders if chip capacity cannot be deployed.
What is behind-the-meter generation?
It is electricity generated on site at the data center itself, rather than drawn from the public grid. Fuel cells are one example. Morgan Stanley's shortfall estimate already accounts for such measures, and still finds a 34% gap.
Does this mean the AI boom is over?
No. The analysis points to a deployment constraint, not a collapse in demand. The question it raises is how fast data centers can be powered and switched on, and which suppliers bear the cost of any delays.
Sources
- Reuters, "Nvidia, Broadcom shielded as AI power crunch hits chip supply chain, says Morgan Stanley" (5 October 2026): 34% net power shortfall through 2028, equivalent to 32 gigawatts, after behind-the-meter generation and fuel cells; Nvidia and Broadcom relatively insulated; memory, optical and other secondary components exposed; Goldman Sachs also flagged constraints; labour, power and political challenges.
- Morgan Stanley brokerage estimates reported by Reuters (September–October 2026): 2027 forecasts for Nvidia and Broadcom not seen at risk from the bottlenecks; delivery push-outs or cancellations possible for secondary components.
Comments
Post a Comment