‘Capital alone no longer clears a site’: Morgan Stanley says data centers’ big money era is over
· Fortune
Something has changed in the data-center story, Morgan Stanley warned this week: Money doesn’t talk the way it used to.
For years, the formula for building a data center in America was straightforward: Find cheap land, line up enormous amounts of electricity, and arrive with a sizable tax-incentive package. Not in the election year of 2026.
Visit moryak.biz for more information.
“Capital alone no longer clears a site,” the investment bank wrote in an Aug. 17 research report, arguing community resistance to data centers has grown from a public-relations challenge into a material development risk. Citing two expert sessions on the topic, the team led by strategist Michelle Weaver, the bank argued political backlash must now be “underwritten like land, power, or labor.”
The warning lands at a fraught moment in the broader race to build AI. The U.S. and China are competing to dominate a technology that most analysts believe will reshape the global economy, military balance, and scientific capacity for decades. Data centers are the factories of that race—the physical infrastructure on which AI models are trained, run, and deployed at scale. China’s government can build them where it wants, when it wants, with limited friction from the communities that surround new facilities. American developers, increasingly, cannot.
In a companion report published the same day, Morgan Stanley analysts laid out what they call the “AI sovereignty” imperative: Governments are now treating compute capacity, energy, data, and supply chains as strategic national-security assets rather than commercial infrastructure, much as earlier generations treated electricity grids or shipping lanes. The bank’s base case is the federal government will not impose a nationwide moratorium on data-center development—doing so would directly weaken U.S. competitiveness against China—and that Washington will continue to support the buildout through permitting reform, energy policy and industrial incentives. The constraint, in Morgan Stanley’s telling, is not federal policy. It is the accumulation of local decisions that federal policy cannot reach.
The financial stakes are significant. The six largest hyperscalers—Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave—are forecast to spend roughly $785 billion on capital expenditures this year and approach $1 trillion in 2027, according to Moody’s Ratings, with total data-center lease commitments across the group now exceeding $1.2 trillion.
Morgan Stanley projects a 38-gigawatt U.S. data-center power shortfall between 2026 and 2028 before new “time-to-power” solutions are added, estimating that some regions already face grid-interconnection waits of five to seven years. Its preferred fixes include rapid deployment of natural-gas turbines and fuel cells, colocating data centers with operating nuclear plants, and converting former Bitcoin-mining sites that already have access to power infrastructure.
The existential risk
The communities raising objections have legitimate grievances: higher power bills, strained water supplies, disrupted farmland, noise, and hundreds of miles of new transmission lines crossing backyards and farm fields. Zoning boards, utility commissions, and state legislatures are democratic institutions doing what democratic institutions do—responding to constituents. But the cumulative effect of a thousand local objections, each individually reasonable, is a national constraint on infrastructure the AI industry says it cannot build fast enough.
Axios calls it “the new existential threat to AI,” reporting that Republicans and AI CEOs are in “full panic mode” watching politicians awaken to the appeal of data-center opposition. If the pipeline of new data centers slows, computing power will be constrained, and investors will (further) question whether the market has legs.
Morgan Stanley’s report puts financial numbers on the constraints. Hyperscalers, cloud providers, and AI companies racing to build the power-hungry facilities needed to train and run their systems now face a collision that is increasingly visible in state capitals and local planning meetings. The bank identifies affordability, environmental concerns, and quality-of-life impacts as the three principal drivers of resistance, and argues developers face a crucial divide between complaints that can be addressed through project design—such as water use, farmland conversion, and emissions—and “societal objections” that cannot.
The new political reality is already changing the economics of expansion.
Texas and Virginia
In Texas, Gov. Greg Abbott earlier this month directed the Public Utility Commission of Texas and grid operator ERCOT to audit all data-center projects moving through the state’s interconnection process. The governor’s office cited more than 474 gigawatts of pending connection requests, about 90% of them tied to data centers. Developers must disclose details including their electricity needs, onsite generation plans, water use, public incentives, ownership, and community impacts. Morgan Stanley characterized the move as more likely to cause a near-term timing disruption than a fundamental collapse in demand.
Virginia is moving in the same direction. A July 31 order by the State Corporation Commission requires Dominion Energy to develop a mandatory contribution-in-aid-of-construction mechanism for direct-connect transmission assets, including substations and lines built specifically for a new or expanding large customer—a move Morgan Stanley described as a shift toward a more explicit “cost causer pays” framework.
The bank’s more consequential point is that the biggest projects may now be the hardest to approve.
“Opposition scales non-linearly with size,” according to the report, making a strategic case for developers to keep projects under regulatory thresholds where possible. It pointed to a 50-megawatt threshold in New York, a proposed 100-megawatt trigger in Delaware and a 20-megawatt trigger in Maine as signs that policymakers are beginning to distinguish projects by scale.
Colocation operators may be relatively better positioned. Morgan Stanley said Equinix and Digital Realty are less exposed because their facilities average roughly 5 to 10 megawatts, far below the gigawatt-scale AI campuses planned by hyperscalers and neocloud operators. Their smaller size, critical-infrastructure status, and longer records of community engagement could produce “softer political pushback,” the bank said.
The bank’s warning to developers: The first sign a project is in trouble may not come from a regulator or a lawsuit. It will appear in a municipal agenda, a zoning hearing, or a local Facebook group—well before litigation, and often too late to change.
For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.
This story was originally featured on Fortune.com