A deep report from Morgan Stanley is delivering a warning to the market amid the frenzy surrounding artificial intelligence infrastructure development.
Morgan Stanley public policy strategist Ariana Salvatore distilled the core constraints facing AI infrastructure into three dimensions 鈥?quot;people, power and politicsquot; 鈥?and examined each potential downside pathway that could derail a capital expenditure program reaching as high as $1.6 trillion. While the bank maintains an overall constructive view on AI infrastructure, the report explicitly states that this expectation comes with an increasingly lengthy list of preconditions.
The report arrives as the market is already pricing in the most unfavorable political scenario for AI sentiment. Morgan Stanley argues that three risks 鈥?U.S. political risk, a reversal in supply-demand price cycles, and a power shortfall 鈥?reinforce one another, and if any single link weakens, the entire capital expenditure cycle could reverse at a quot;catastrophicquot; pace.
U.S. Political Risk: Local Power Poses Greater Threat Than Federal Authority
Morgan Stanley's report makes clear that a comprehensive federal pause on AI infrastructure is unlikely 鈥?the existence of a national security framework and the president's potential veto power over restrictive legislation make it improbable that Washington would directly halt construction. However, the truly binding leverage lies at the local level: interconnection applications, large-load electricity pricing design, site selection, permitting approvals and tax incentives are all determined by state governments and public utility commissions.
Salvatore states directly in the report that U.S. political headwinds have become a central issue in midterm elections and could serve as a catalyst affecting AI sentiment. She notes that non-incumbent candidates broadly tend to propose stricter restrictions, leading Morgan Stanley to conclude that market sentiment would be most favorable under a Republican sweep, largely neutral under a divided government, and most unfavorable under a Democratic sweep.
The report includes a race-by-race scorecard covering every election 鈥?from governor to elected utility commissions 鈥?that could tighten rules.
Three key takeaways stand out. First, the debate has shifted from quot;whether to restrictquot; to quot;how to restrict.quot; Second, regardless of which side wins, eliminating tax incentives has become a bipartisan consensus 鈥?Arizona has already suspended related exemptions through June 2029. Third, utility commissions may have the most far-reaching influence, because their rulings carry legally durable effect.
Take Ohio as an example: AWS, Meta, Google and Microsoft have all built facilities there, but Democratic candidate Amy Acton supports a quot;conditional pausequot; on new data center construction, while Republican candidate Vivek Ramaswamy has proposed an quot;Ohioans First Data Center Commitmentquot; requiring developers to bear full property taxes, provide free electricity to surrounding residents and meet strict environmental standards. Morgan Stanley counts nearly 60 gigawatts of data centers across the flagged states that are under construction, planned or on purchased land, with Texas alone accounting for 24 gigawatts.
Price Cycle: Volume and Price Decline Risk Underestimated by the Market
Morgan Stanley characterizes the second major risk as a classic capital cycle problem and argues it currently receives the least market attention.
At the current stage, scarcity has spawned what the bank calls a quot;procurement panicquot;: buyers expect computing power to be more expensive tomorrow, so they rush to purchase at a premium today, while suppliers layer on additional profit margins, creating a mutually reinforcing positive cycle of volume and price. But once supply catches up with demand or demand growth normalizes, this mechanism will rapidly reverse 鈥?earlier stockpiling and urgency will crush pricing power, and suppliers will by then be locked into high-cost inventory.
Morgan Stanley draws an analogy to the memory market downturn at the end of 2022: at that time NAND shipments fell only about 5%, yet prices dropped 20% to 30%, and supplier margins collapsed far faster than the shipment decline alone could explain. The report applies this logic directly to AI, noting that even if hyperscaler capital expenditure merely enters a quot;plateau rather than a collapse,quot; it could still cause pricing power and upstream margins to evaporate while shipment volumes remain resilient.
The report's conclusion is bluntly worded: quot;We believe the risk of simultaneous volume and price declines is severely underestimated in the current market.quot; Notably, Amazon attributed the main reason for its 2026 capital expenditure increase to rising memory costs rather than an expansion in construction scale 鈥?meaning a considerable portion of the current capital expenditure quot;boomquot; is driven by price factors, and memory is precisely the segment most likely to reverse first.
Power Shortfall: The Most Immediate and Tangible Risk Today
Morgan Stanley calls the power problem quot;the most palpable risk at present.quot; The bank's own quot;AI Power Supply Modelquot; shows that even accounting for various immediate power solutions including nuclear energy, gas turbines, engines and Bloom fuel cells, the United States still faces a net power shortfall of 32 gigawatts through 2028, roughly 34% of the 97 gigawatts needed; by 2029, that gap will widen to approximately 72 gigawatts.
Morgan Stanley characterizes this as a timing mismatch rather than a demand collapse. Hyperscalers can continue purchasing GPUs before power is available and stockpile them in data centers that are not yet energized 鈥?this supports chip revenue in the short term but could lead to quot;hoarded and stranded inventory,quot; and quot;continuously buying power-dependent equipment that cannot be turned on is ultimately unsustainable.quot;
The report further distinguishes risk exposure across supply chain segments: components shipped with racks 鈥?such as memory, networking, optics and ODM servers 鈥?carry the highest risk, with memory particularly sensitive because recent capital expenditure increases are partly price-driven; electrical, cooling and construction equipment face slower backlog conversion due to interconnection delays; relative beneficiaries include behind-the-meter power generation (such as gas turbines and modular nuclear reactors), fuel cells and suppliers of already-energized shells. Morgan Stanley also expects that if domestic U.S. power tightness persists, hyperscalers will increase capital expenditure allocations outside the United States.
Three Risks Compound: Conditional Construction Remains the Base Case
After laying out these three major risks, Morgan Stanley still maintains a constructive base case: conditional construction progress, a power shortfall that delays rather than destroys demand, and a price cycle that can be managed through long-term agreements.
Industry-level responses are already underway: several hyperscalers have signed a quot;Ratepayer Protection Pledge,quot; committing to build or procure their own generation facilities and bear grid upgrade costs; Google has pledged to replenish more water to local systems than its data centers consume by 2030; Amazon disclosed global data center water usage data for the first time; and Meta has established a $1 billion fund to support teachers, first responders and infrastructure in host communities.
However, a mutually reinforcing transmission mechanism exists among the three risks: political resistance slows interconnection applications and site selection, which widens the power gap, causing chips purchased at peak prices to sit idle and ultimately triggering the quot;simultaneous volume and price declinequot; that Morgan Stanley warns about. The report notes that an infrastructure wave that must pass the triple test of the ballot box, the power grid and the bond market faces far greater uncertainty in actual execution than the headline figure of $1.6 trillion suggests.
Comments