- Статьи
- Internet and technology
- Electrocuted: How Exorbitant AI Ambitions are Dampening the U.S. Electric Power Industry
Electrocuted: How Exorbitant AI Ambitions are Dampening the U.S. Electric Power Industry
The US authorities have declared an emergency in the country's largest power grid, PJM, covering 13 states and the District of Columbia. The head of the Ministry of Energy, Chris Wright, approved an emergency protocol for this region. The decision was dictated by a sharp surge in consumption amid the threat of a shortage of electricity and a shortage of reserves for its generation and transportation. The rapid growth of the AI companies market significantly outstrips the opportunities and potential for expanding energy infrastructure. Experts warn that such a skewed system is fraught not only with problems with the power supply to other consumers, but also with major troubles on the stock market. Details can be found in the Izvestia article.
A spark of desire
The likelihood of an energy crisis in the United States, associated with sharply increased consumption of AI infrastructure, was warned two years ago. The construction of data centers (data centers), carried out at a reactive pace in the United States by technology corporations, puts excessive pressure on the American power supply system.
The scale of the problem is confirmed by statistics. According to the U.S. Energy Information Administration, by 2028, the share of energy consumed by data centers could reach 6.7–12% of the total in the country. Electricity consumption by these facilities in the country has almost tripled over the past decade: from 58 TWh in 2014 to 192 TWh in 2024 (4.7% of total consumption). By 2030, according to forecasts of the Lawrence Berkeley National Laboratory, this figure will reach 649 TWh, which will be 11.8%. Between 2024 and 2030, data centers will account for 33% of the total projected increase in electricity consumption, more than any other sector.
It is also important that the regional load is unevenly distributed, because IT giants (also known as hyperscalers) make decisions about building data centers based on various factors, not only based on the capabilities of the power grid: from local laws to political views. For example, Virginia has the largest number of data centers, which consume from 26 to 40% of the state's total electricity contained in a single supply system. For this reason, the power surge that occurred in July 2024 led to the simultaneous shutdown of 60 data centers, which created an excess of 1,500 MW of energy and required emergency intervention to prevent large-scale blackouts.
In turn, the authorities of Texas, which has an independent ERCOT power grid, have already planned to modernize power transmission lines worth more than $30 billion specifically to serve new high-load consumers.
The surge in demand naturally hits the wallets of ordinary consumers. The recent auction of the affected PJM power grid for the sale of capacity for 2025-2026 resulted in additional costs of $9.3 billion precisely due to the demand for new data centers. As a result, this increased the bills of PMJ households in Ohio and Maryland, and the price of grid capacity soared from $28.92 to $329.17 per megawatt-day in two years.
The Smoke of Opportunity
In an attempt to anticipate future problems, leading IT corporations are actively creating autonomous power systems. An illustrative example is the xAI Colossus data center near Memphis, Tennessee, used to train the Grok model. If the facility is operating at full capacity, it will consume as much electricity as 200,000 households in the neighborhood. To launch it, xAI built its own power plant, installing 35 gas turbines on it, each the size of a railway carriage, which immediately caused a flurry of criticism from environmentalists due to the giant clouds of combustion products thrown into the atmosphere. According to Elon Musk, after commissioning, Colossus and two other nearby data centers will need almost 2 GW of energy - they will consume almost twice as much electricity annually as, for example, the whole of Seattle.
Another solution is the introduction of small modular nuclear reactors (MMRs) as a way to ensure continuous operation while reducing carbon dioxide emissions and price volatility. However, such projects are fraught with serious logistical and technological challenges — obtaining regulatory approvals, environmental impact assessment, the need for careful design of integration with cooling and redundancy systems, as well as overcoming phobias associated with nuclear energy.
Moreover, this initiative is not only commercial. The White House has been actively promoting it through various projects since the first presidential term of Donald Trump. In May last year, the American leader signed a decree "On the use of advanced nuclear technologies for national security purposes." The contractors are private companies, they are also engaged in similar developments for hyperscalers.
From reality to dreams
However, there is a significant gap between ambition and reality, and it is measured in trillions of dollars. The staggering pace at which large corporations are investing in artificial intelligence and investing in the construction of data centers across the country is too far ahead of the pace of reconstruction and creation of energy infrastructure. According to Sightline Climate's forecast, almost half of the planned 16 GW of capacity in the United States in 2026 will be postponed. Currently, only 5 GW are under construction. The gap is caused by problems obtaining permits at the individual state level, resistance from diverse local communities, and dependence on overloaded global supply chains for critical components such as transformers.
According to forecasts, the imbalance will only increase by 2027: only about 6.3 GW of the announced 21.5 GW will be built. After 2028, the uncertainty will increase even more: 37 GW of the planned infrastructure does not have an exact completion date, and only 4.5 GW of them has already begun work. Therefore, according to Bank of America estimates, between 2026 and 2030, the country will face an annual electricity shortage of 100 GW. To understand the level of the problem: in 2025, the volume indicated in the connection requests was almost one and a half times higher than the average consumption of the entire country in 2023.
The joy of waiting and the sadness of finding
The market is already responding to this gap. If stocks from the AI Data center and TMT AI baskets (the same hyperscalers) are ahead of the market, then securities from the Power Up America basket (companies involved in the modernization of the American energy system) are significantly lagging behind. Investors naturally doubt that the United States will be able to meet its needs for AI computing, which could have disastrous consequences for record budgets for capital investments in artificial intelligence. Think tanks suggest that even more active government involvement is needed. However, even the approximate cost is higher than any possible. According to JPMorgan, it will take at least $5 trillion to stabilize the growing gap on a global scale. Moreover, for this money to bring a 10% return, the AI industry must generate $650 billion in annual revenue. Now this amount barely exceeds $75 billion.
Another major risk factor is time. The construction of gas turbines for power plants in the USA takes 3-4 years, and nuclear power plants take more than 10 years. Some companies like the already mentioned xAI are launching their own production of individual components to reduce these deadlines.
Experts warn that in order for huge investments in artificial intelligence to pay off, appropriate infrastructure and production are needed, allowing it to be constantly reconstructed and modernized. However, the physical capabilities and resources of the American energy industry cannot keep up with the enormous demand for AI in terms of money, timing, or ambition. There is a risk that emergency episodes like these will become commonplace in the United States. But, even more worryingly, the gap between wishful thinking and reality may one day pierce the overgrown AI bubble.
Переведено сервисом «Яндекс Переводчик»