Lack of electricity, lack of water, lack of manpower and land grabbing! The construction frenzy of American data centers faces resistance.

The data center construction boom, driven by the artificial intelligence revolution, is facing resistance in reality. From electricity grid capacity, water resources, to technical workers, and the competition with residential development for land, this infrastructure frenzy led by tech giants such as Microsoft, Alphabet, Meta, and Amazon is facing multiple constraints, and the execution risks are rising, potentially dampening the market's optimistic expectations for AI investment returns. A recent conversation between Goldman Sachs analyst Brian Singer and Mark Monroe, former chief engineer of Microsoft's data center senior development team, revealed three key bottlenecks: electricity remains the most urgent constraint in the near term, water resource pressure is forcing the industry to turn to higher-energy cooling technologies, and a shortage of technical workers may become the next hurdle. Monroe warned that by 2030, the U.S. would need to add more than 500,000 manufacturing, construction, operations, and transmission and distribution workers to meet the electricity demands of data centers. At the same time, tech giants are buying land at unprecedented prices in multiple locations in the U.S., directly squeezing residential development. According to the Wall Street Journal, Amazon spent $700 million in November last year to acquire residential developer Stanley Martin in Virginia, which was originally purchased for just over $50 million. In Northern Virginia, rural land that once sold for tens of thousands of dollars per acre is now priced at over $3 million, making it impossible for residential developers to compete. The sustainability of this construction frenzy is crucial to the current macro narrative and the valuation of tech stocks - that is, whether data center investments can measurable productivity improvements and support long-term growth. However, supply chain bottlenecks, infrastructure constraints, and community opposition are accumulating, potentially leading to unmet overly optimistic expectations. Electricity bottleneck remains urgent Electricity supply continues to be the most critical near-term constraint for data center deployments. Monroe pointed out that while cloud computing and AI inferencing workloads often need to be close to end-users - leading to power shortages in crowded markets, AI training workloads are not sensitive to geographic location and are moving to remote areas with sufficient power. Flexible load management could release some capacity, but adoption is hindered. A study from Duke University showed that if data centers accept an average annual load reduction of 0.25% (99.75% uptime), they could add 76 gigawatts of load, equivalent to 10% of the peak total demand in the U.S.; if they accept a 0.5% reduction (99.5% uptime), they could add 98 gigawatts. However, Monroe indicated that the widespread adoption of this solution is hindered by the industry's inherent risk-averse culture - repeated switching of IT equipment makes operators uneasy, and may require stronger financial or regulatory incentives. On-site generation becomes an expensive temporary solution. Although only a small percentage of data centers under construction apply on-site generation, Monroe emphasized that these are typically large data centers with significant power demand. These projects mainly deploy natural gas simple cycle generators, which cost 5 to 20 times more than grid electricity. However, considering the huge profitability of large AI data centers, the use of on-site generation to facilitate project deployment remains economically feasible. Monroe mentioned that the ultimate goal for data centers deploying on-site generation is to connect to the grid within three years, either by relocating to other data centers, integrating and selling power back to the grid, or phasing out on-site generation assets. Water resource constraints bring energy costs Pressure from communities, regulations, and advancements in chip technology is driving the industry towards more water-efficient but energy-intensive cooling technologies. Monroe stated that as pressure from communities, regulations, and technology intensifies, the industry is shifting from traditional water-intensive evaporative cooling to designs with less water usage, especially for hyperscale operators. This shift comes with significant energy efficiency losses. Monroe pointed out that the transition to closed-loop and waterless cooling systems could increase the power usage effectiveness (PUE) from the optimal level of 1.08 to 1.35-1.40, leading to an increase in energy expenditure from 8% with evaporative systems to 35%-40%. While innovations like direct chip liquid cooling and high-temperature water cooling can achieve more efficient heat transfer in more geographic locations, colocation data centers, due to their diverse customer base and the need to determine cooling architectures early in the construction process, may still stick to using chiller plant designs. Monroe stated that although the share of evaporative cooling in the overall data center cooling solution may decrease, the demand for chiller plants is expected to significantly increase over the next decade due to the overall growth in data center capacity. Shortage of technical workers becomes the next hurdle Monroe warned that a shortage of technical workers could become the next bottleneck for data center deployments. The difference between data centers and ordinary industrial buildings lies in the specialized electrical and mechanical systems required, making electricians and pipefitters crucial to data center construction. Industry organizations are collaborating with technical universities and colleges to develop training programs to address this gap, and are trying to engage students as early as middle school to make the tech industry a more attractive career path. According to Goldman Sachs estimates, by 2030, the U.S. will need to add over 500,000 net workers in manufacturing, construction, operations, and transmission and distribution fields to deploy all the needed power. Tech giants land grabbing raises prices Data center developers are acquiring land at prices far exceeding other purposes, directly impacting residential development. According to the Wall Street Journal, when Stanley Martin CEO Steve Alloy was preparing to develop 516 new homes in Bristow, Virginia, five years ago, he found that the surrounding land was being bought up by tech giants like Microsoft and Google. By November last year, the company sold the land it had acquired for just over $50 million years ago to Amazon for $700 million, making it one of the largest land deals in U.S. history. Northern Virginia has become the global data center capital. The region has vast land, growing power infrastructure, and a dense fiber optic network laid during the internet bubble era. Loudoun County is home to a cluster of facilities known as the "Data Center Alley," and the world's largest tech companies are moving south along Interstate 95 into Prince William County. The soaring land prices have made it impossible for residential developers to compete. In Northern Virginia, developers have sent letters to landowners offering prices as high as $1 million per acre. Some rural land that was once sold for tens of thousands of dollars per acre is now priced at over $3 million. In Elk Grove Village, a data center hub near Chicago, Stream Data Centers bought and demolished a 55-home neighborhood for nearly $1 million per house in 2024, to build three data centers totaling 2.1 million square feet. Along U.S. Highway 67 near Dallas, land that was selling for $20,000 to $40,000 per acre three years ago is now selling for over $350,000 in some areas. Residential land developer Scott Finfer stated, "Residential developers simply cannot make these numbers work." Looking ahead, the key issue is whether the U.S. can sustain the continuous increase in data center capital spending, considering that these constructions are deeply embedded in the macro narrative and the valuation of tech stocks. The investment argument assumes that ongoing construction will measurable productivity improvements, leading to long-term growth. However, the execution risk ultimately depends on key inputs and infrastructure, such as core components, grid connections, and related supply chain bottlenecks, which could slow down construction and thwart overly optimistic expectations. This article is a translation from Wall Street Horizon, edited by Chen Wenfang.