🇰🇷 한국어 🇺🇸 English 🇯🇵 日本語 🇪🇸 Español 🇨🇳 简体中文
Published: 2026.11.18 (Wed)
Economy

Grid bottlenecks threaten AI profitability as data centers sit idle

The success of the AI industry is shifting from computing power to power infrastructure, as delays in connecting completed data centers to the electrical grid create significant profitability risks.

Grid bottlenecks threaten AI profitability as data centers sit idle
A man in a white shirt speaks in front of a microphone. (Photo=Screenshot of Understanding: All the Knowledge in the World YouTube video)

Data center investments face a profitability crisis as transmission grid limitations leave completed facilities idle. While US electricity demand is expected to soar to 500–600 TWh by 2030, grid connection delays in regions like Pennsylvania can take up to 12 years.

The success of the artificial intelligence (AI) industry is shifting from computing capacity to power infrastructure. Even when data centers are completed, delays in connecting them to the power grid are creating a "profitability crisis," where massive investments turn into the burden of vacant facilities.

According to an analysis by Professor Kim Hak-ju presented by Understanding: All the Knowledge in the World, the key variable threatening current AI investment profitability is not simple computing power, but rather the "bottleneck in power supply." While constructing a data center typically takes about a year, connecting it to the power grid can take several years. This time lag increases vacancy rates and rent burdens for data centers, serving as a direct cause of deteriorating free cash flow for companies. "Data centers are built and then left sitting idle. Real money is being wasted right now," Kim noted, pointing out that delays in power supply directly impact corporate profitability.

Power Plants are Sufficient, but Electricity Cannot be Delivered: The Reality of the 'Grid Bottleneck'

While many argue for the construction of additional power plants to solve electricity shortages, critics point out that the real issue lies in the capacity limits of the transmission grid rather than generation capacity. Kim explained, "Even now, the reason power plants are not operating at full capacity is that the transmission grid is full and cannot push more electricity through." Although power plants themselves have sufficient capacity to supply electricity, the transmission grid—the "blood vessels" that deliver it—is blocked and cannot keep up with demand.

In the United States, electricity demand is exploding. Current power demand from data centers is at the level of 200 terawatt-hours (TWh) per hour, but it is expected to soar to 500–600 TWh by 2030. Although plans have been made to restart three large nuclear power plants that were previously closed to supplement the shortage, the power secured through this will be "insignificant" compared to the additional demand of hundreds of TWh.

Ultimately, the key lies in expanding the transmission grid. Kim emphasized the importance of grid expansion, stating, "If transmission capacity is improved, we can feed in as much electricity as needed." In the early 2000s, power plant construction and grid connection were completed within two years, but it now takes more than five years, including administrative processing. In data center-dense regions such as Pennsylvania, New Jersey, and Maryland, cases have occurred where grid connection takes up to 12 years. As infrastructure construction periods lengthen, analysis suggests that 70–80% of those who apply for power plant construction end up abandoning the project due to uncertainty in power supply. Solving this issue is difficult as transmission and distribution infrastructure is managed by a limited number of companies that function essentially like public enterprises.

Natural Gas, Nuclear, and Renewables: The Dilemma of AI Power Supply

The advantages and disadvantages of various energy sources for power supply are distinct. Natural gas is considered an alternative that can meet immediate demand due to its short construction period, but it is hindered by fuel price volatility and carbon emission issues. Nuclear power offers the advantage of being a carbon-free energy source with low fuel cost ratios, but the risk lies in the massive financial costs incurred as construction periods lengthen in a high-interest-rate environment.

Renewable energy has limitations in directly supplying data centers, which must operate 24 hours a day, due to the issue of output volatility. Consequently, debates continue within the Federal Energy Regulatory Commission (FERC) regarding "Direct Supply"—a method of supplying electricity produced at power plants directly to data centers without passing through the transmission grid.

This infrastructure shortage also affects the investment structures of AI companies. Currently, big tech companies like Meta are building their own data centers while simultaneously leasing data centers from companies like CoreWeave, which has received equity investment from NVIDIA. This is similar to pre-sale apartment contracts; even if vacancies occur due to power supply issues after completion, companies face the risk of having to pay rent according to their contracts. Ultimately, the lack of power infrastructure becomes a core factor in deteriorating the cash flow of AI companies.

Focus on 'Infrastructure Demand' Rather Than Volatile Individual Stocks

From an investment perspective, it is necessary to focus on infrastructure-related companies with steady demand rather than individual AI stocks with high volatility. Kim specifically mentioned natural gas transportation companies, industrial gas and liquefaction equipment companies, and companies related to the restarting of nuclear power plants as specific beneficiary sectors.

In particular, he emphasized "maintaining investment principles" when responding to market crashes. He warned that it is dangerous to sell out of fear without judging value when stock prices fall, or to use excessive leverage to recover losses. "When buying a stock, you should record why you are buying it, like writing a diary," Kim advised. "What is important in a bear market is not your current emotions, but re-reading the reason you initially invested." He added that it is important to maintain a certain percentage of cash (dry powder) when investing, and that revisiting the original investment rationale helps in making wise decisions when the market is volatile.

Ultimately, the sustainability of the AI industry depends on how it overcomes the physical limitations of power infrastructure. The profitability of data centers is expected to be determined not by the advancement of simple algorithms, but by how quickly the massive "blood vessels" of the power grid can expand.

Source: original video (YouTube)

#AI #data centers #power grid #Kim Hak-ju #Meta #NVIDIA #CoreWeave #FERC
H
Han Kyungsoo
TrendBiz · Reporter

Covers Economy for TrendBiz, and also writes about Company News and Finance.

More by this reporter ›
Copyright ⓒ TrendBiz All rights reserved. Unauthorized reproduction, redistribution, and AI training prohibited.

0Comments

Comments are currently disabled.

Be the first to comment.