Morgan Stanley: U.S. data center power gap reaches 34%; leading chipmakers temporarily immune, midstream and downstream segments under pressure.

GMTEight APP has learned that Morgan Stanley released a research report updating its estimates of the U.S. AI data center power shortage and the impact on the industry chain. The report shows that the total U.S. data center power demand gap for 2026-2028 will reach 57GW. After deducting "plug-and-play" power solutions such as behind-the-meter (BTM) generation and fuel cells, the net gap still reaches 32GW, accounting for 34% of total demand over the same period. The power shortage has gradually replaced chip supply as the core bottleneck for AI computing expansion. However, the report emphasizes that Nvidia and Broadcom, with their global footprint, higher computing output per unit of power, and clear upstream and downstream visibility, will not see their 2027 performance guidance affected by power constraints for the time being, while midstream and downstream segments such as ASICs, storage, optical modules, and analog devices will face greater demand volatility risk. Gap estimate: A 34% power gap will persist for a long time, with computing resources concentrating toward leaders Morgan Stanley estimates that cumulative U.S. data center power demand for 2026-2028 will reach 97GW, of which data centers under construction cover 21GW and available grid capacity covers 19GW. After deducting these two items, the initial power gap is 57GW. After including probability-weighted "plug-and-play" solutions such as BTM gas turbines, Bloom Energy fuel cells, nuclear plant site conversions, and crypto mining site conversions, the net gap under the base case is 32GW, accounting for 34% of total demand; under the optimistic case, the gap can narrow to 15GW, while under the pessimistic case it widens to 42GW. By year, the power gap shows a trend of widening year by year: the net gap is about 9GW in 2026, about 15GW in 2027, expands to 32GW in 2028, and climbs further to 68GW in 2029. The report points out that the large-scale deployment of next-generation high-power GPU racks (such as NVL72) is the core driver of the surge in power demand. The supply side also shows a highly concentrated landscape. Among global new data center capacity in 2026-2028, hyperscale cloud providers and neocloud providers (Amazon, Google, Microsoft, Meta, etc.) will contribute 60%-95% of the increment. Leading players, with stronger credit profiles, revenue certainty, and procurement scale, hold an absolute advantage in the competition for power resources, while small and medium-sized cloud providers, international providers, and low-efficiency chip manufacturers face the risk of being squeezed out. (In 2027, hyperscale cloud providers/neocloud providers will add 32 gigawatts of computing capacity, and in 2028 they will add 39 gigawatts; among them, Google ranks first in the scale of newly commissioned capacity.) Divergent impact across the industry chain: Leading chipmakers are safe, while the midstream and downstream tail end comes under pressure The report argues that the impact of the power shortage on the semiconductor industry chain is significantly tiered. Nvidia and Broadcom together hold about 90% of the AI XPU market, and the two companies' combined 12-month AI revenue guidance is about $800 billion. At present, this guidance is not materially threatened by the power shortage. The core support lies in four aspects: management's performance guidance has fully considered risks from shortages of sites, power, and infrastructure supporting facilities (LPS); there is high visibility into where chips will be deployed across the full chain; they are actively building global computing nodes outside the United States; and Nvidia's computing output per gigawatt is significantly higher than that of peers, giving it a clear comparative advantage in a power-constrained environment. By contrast, ASIC chips have lower token output per unit of power, and their market share faces the risk of being squeezed by GPUs in a power-constrained environment. At the end of the industry chain, low-value subcategories such as storage, optical modules, power management, and analog devices are most vulnerable to the bullwhip effect if the pace of computing deployment slows, customers will prioritize delaying or canceling orders for such components, and the related manufacturers face higher risks of revenue and profit volatility. Limited mitigation paths, and overseas relocation is unlikely to fill the domestic gap Regarding mitigation options for the gap that the market is focused on, the report argues that overseas computing expansion cannot fill the U.S. domestic gap. Although U.S. companies are accelerating computing deployments in Australia, Asia, Europe, and other regions, Europe is constrained by power limitations and approval cycles, the Middle East carries geopolitical risks, and Asia is expected to absorb only 14GW of spillover U.S. supercomputing demand by 2030, a scale insufficient to offset the U.S. domestic gap. On this basis, the report lowers its forecast for the U.S. share of global computing power from 60% to 55%. Behind-the-meter (BTM) generation is currently the main source of incremental power. Under the base case, BTM gas turbines and engines can contribute 19GW of power in 2026-2028, and up to 49GW under the optimistic case, but due to constraints such as shortages of skilled labor, engineering complexity, and local approval processes, there is considerable uncertainty about the actual scale that will be realized. In addition, solutions such as fuel cells and nuclear plant site conversions can also provide some supplement, but they are unlikely to fundamentally solve the long-term gap problem. Overall, the power shortage is reshaping the competitive landscape of the AI computing and semiconductor industry chains. The trend of resources concentrating toward leading manufacturers will further strengthen, and companies with energy efficiency advantages, global deployment capabilities, and the ability to secure power resources will benefit more, while midstream and downstream niche segments need to guard against demand volatility risk.