U.S. equities suffered a sharp reversal on Thursday, concentrated in sectors most exposed to the AI investment cycle. The Nasdaq 100 fell more than 6%, while the Philadelphia Semiconductor Index dropped 12.25%, its worst session since the pandemic-era sell-off. Weakness spread into the broader market, taking the S&P 500 down 4.36% and pushing the VIX above 28.
The sell-off represents an acceleration of a reassessment that was already underway around the AI infrastructure buildout. Investors are increasingly questioning the pace, economics and eventual returns associated with extraordinary levels of AI investment.
Importantly, the rotation began well before this week’s U.S. midterm election. From October 5, when the 10-year Treasury yield peaked at 5.3%, through November 3, Communication Services and Information Technology were the two worst performers among the eleven S&P 500 sectors, declining 15.3% and 10.4%, respectively. The other nine sectors were higher over the period.
This week’s mid-term U.S. election results, which included democratic victories in key congressional and gubernatorial races, added political and regulatory uncertainty to an AI investment story that was already being reassessed.
Data center development featured prominently in several races, reflecting growing concern around electricity prices, grid capacity, water usage and local infrastructure. With state and local authorities exerting considerable influence on permitting, utilities and development, investors must now account for a greater risk of tighter approval conditions, higher costs and project delays.
That matters well beyond data centers. The AI investment cycle has become an important driver of U.S. business investment, earnings expectations and market leadership. The crucial question is whether political resistance ultimately changes the scale of AI investment or primarily its timing, location and cost.
Data centers have moved from a largely local land-use issue to a broader political focus, combining voter angst over affordability and resource use with growing unease about AI itself. Communities increasingly question why they should bear the demands on electricity, water, land, and infrastructure without receiving a commensurate share of the economic benefits.
These concerns cut across party lines and are unlikely to disappear with the election. But greater scrutiny is very different from the end of development. The strategic case for development remains strong across both parties, with AI infrastructure increasingly viewed as important to U.S. economic competitiveness and national security.
So, while policymakers may disagree over where data centers should be built, how resources should be allocated and who should pay for the necessary infrastructure, severely restricting the buildout would carry significant economic and strategic costs.
Geography also provides an important release valve. If one jurisdiction becomes prohibitively difficult or expensive, investment can migrate toward states and municipalities offering power, infrastructure, and a more supportive political environment. The result is more likely to be a redistribution of development than a wholesale retreat from it.
The distinction between the headlines and the underlying policy trajectory is therefore critical. Moratoriums and canceled projects will attract attention, and some jurisdictions may impose significant restrictions. But the broader direction is more likely to be toward changing the terms under which development proceeds rather than stopping it altogether.
Although political opposition is real, the policy debate is increasingly about the terms of development, not whether development should occur at all.
Authorities are likely to demand more from developers before projects receive political and regulatory support. That could mean greater responsibility for grid and infrastructure costs, protections for households from higher electricity bills, enhanced disclosure of energy and water consumption, closer scrutiny of tax incentives, and more explicit investment in surrounding communities. Data center tenants and developers will need to demonstrate that the economic benefits of development are being balanced against its local costs.
The result is unlikely to be a halt to development, but a more conditional path forward. Projects that can address community concerns while delivering the considerable economic and strategic benefits of AI infrastructure should continue to move ahead. Getting there could nevertheless be disruptive. Projects may be delayed, costs could rise, and development is likely to become more geographically dispersed as companies seek locations that combine power availability with political and community support.
Data centers are often treated as a single category, but their infrastructure requirements differ considerably. Gigawatt-scale AI training campuses pose far greater challenges for generation, transmission, water, and local infrastructure than smaller cloud, inference, enterprise, or colocation facilities.
Water offers one of the clearest examples of technological adaptation.
Traditional evaporative cooling can require substantial ongoing consumption. In contrast, newer closed-loop systems require an initial fill, roughly 615,000 gallons, or about one Olympic-sized swimming pool, that is then continuously recirculated. Investors should expect closed-loop cooling and similar technologies to increasingly become table stakes, particularly in water-stressed markets, materially reducing one of the industry’s most visible community concerns.
Power is a harder challenge, but the industry is adapting here as well.
Grid interconnection timelines can stretch beyond five years in some markets, underscoring the importance of behind-the-meter generation. Moody’s expects it to supply at least 30% of the growth in U.S. data center power demand through 2030, allowing some projects to reduce their dependence on conventional grid capacity. Batteries, onsite generation and workload management can further reduce or shift grid demand during periods of system stress.
Political and physical constraints could therefore influence not only how quickly and where capacity is delivered, but what type of capacity gets built.
Early industry responses offer some indication of how this could evolve. OpenAI’s proposed framework places greater responsibility on developers for grid costs, resource disclosure and community investment. At the same time, Europe offers examples of governments reconciling local concerns with the strategic need for digital infrastructure. As the AI infrastructure cycle matures, the ability to combine available power with community support, appropriate infrastructure and acceptable economics will become an increasingly important competitive advantage.
The election has increased the probability of greater regulatory friction around AI infrastructure. It has not demonstrated that the underlying AI investment cycle is breaking.
Technology earnings remain strong, while there is still limited evidence of a decisive earnings handover to the consumer-facing parts of the market. A durable reversal in market leadership would likely require political pressure to translate into measures that materially reduce the scale of AI capex, rather than simply alter its timing, location or cost.
Near-term volatility could nevertheless persist. Positioning and valuation gaps between AI leaders and market laggards remain substantial, leaving the former particularly sensitive to challenges to the prevailing investment narrative. But with the broader economic backdrop still relatively firm, an AI-led correction need not translate into a broader deterioration in economic fundamentals.
Political and infrastructure constraints are changing assumptions around how quickly, where and on what terms new computing capacity can be delivered.
Access to land and power will remain critical, but the ability to navigate grid connectivity, infrastructure costs, regulatory requirements and community concerns will increasingly determine which projects can move forward. That should create greater differentiation across companies, projects and locations.
Near-term, greater barriers to new development could increase the scarcity value of existing data centers, particularly assets with secured power and established grid connections. Over the medium term, tighter standards around power, water and community impacts should also push the sector toward a more sustainable development model. The transition was likely coming regardless; what has changed is the speed at which it is arriving.
The winners may not necessarily be the largest projects or those with the most ambitious development pipelines. They are more likely to be those best able to match computing demand with power availability, appropriate infrastructure, sensible economics and community acceptance.
For broader markets, this argues for greater selectivity within the AI theme rather than abandoning it. Strong underlying demand can coexist with greater regulatory friction, higher costs and substantial volatility as the investment cycle matures. Political and infrastructure constraints may redistribute the economics of the AI buildout rather than undermine the buildout itself.
Political resistance is unlikely to stop the AI infrastructure buildout. But it is increasingly likely to determine where it happens, how quickly it proceeds, and who ultimately captures the returns.
Investing involves risk, including possible loss of Principal. Past Performance does not guarantee future return. Potential investors should be aware of the risks inherent to owning and investing in real estate, including value fluctuations, capital market pricing volatility, liquidity risks, leverage, credit risk, occupancy risk and legal risk. All these risks can lead to a decline in the value of the real estate, a decline in the income produced by the real estate and declines in the value or total loss in value of securities derived from investments in real estate. Data center investment risks include power constraints and rising costs, technological obsolescence, potential overbuilding, regulatory hurdles, supply chain vulnerabilities, physical and cyber security threats, and increased competition. Data center properties will only be attractive to a unique type of tenant. Additionally, a property designed to be a data center property may be difficult to relet to another type of tenant or convert to another use and will be more likely to become functionally obsolete when compared to other properties.
Views and opinions expressed are accurate as of the date of this communication and are subject to change without notice. This material may contain ‘forward-looking’ information that is not purely historical in nature and may include, among other things, projections and forecasts. There is no guarantee that any forecasts made will come to pass. Reliance upon information in this material is at the sole discretion of the reader.
The information in the article should not be construed as investment advice or a recommendation for the purchase or sale of any security. The general information it contains does not take account of any investor’s investment objectives, particular needs, or financial situation.
6096475
No information