AI's electricity use is skyrocketing, the US electricity gap may burst in 2033

The Zhitong Finance App learned that within this decade, the US may fall into an electricity crisis: demand for electricity in data centers is surging, but it is colliding head-on with aging grid infrastructure. Schneider Electric SE (Schneider Electric SE)'s latest warning shows that if the current backup capacity level is maintained, there will be a substantial gap in the country's power generation capacity by 2028 during the peak electricity consumption period; more seriously, its model predicts that the power gap in 2033 may be as high as 175 gigawatts, which is enough to cause large-scale power outages and a systemic power limit crisis.

According to information, the US power system is a patchwork of regional power grids and utilities, and is now facing multiple pressures from data centers, new factories, and electric vehicles. Even before the rise of the AI boom in the past two years, the frequency of extreme weather and the continuous increase in the share of intermittent renewable energy already overwhelmed power grids.

Jeannie Salo (Jeannie Salo), Schneider Electric North America's chief public policy officer, said in an interview: “This may mean we may be lagging behind in the AI race. It will also send a signal to the world that it is difficult for the US to meet the electricity needs of this era, and investment is likely to decline as a result.”

Schneider predicts that rising peak demand will increasingly squeeze so-called “backup capacity” — additional power buffers that can be used during extreme weather or cyber attacks. Diverting emergency electricity to cope with daily spikes will threaten grid reliability and increase vulnerability.

Salou graphically compared backup capacity to an emergency savings fund in a bank. She pointed out that after three years, the US will have to use these “reserves” regularly. At that point, these reserves may be scarce and simply unable to withstand the impact of serious emergencies. Although batteries are currently being installed in many places, the amount of electricity that can actually be stored is still very small, like a glass of water. You need to know that power grids must accurately balance power generation and electricity consumption every second to avoid falling into the dilemma of collapse.

Salou said, “Currently, we are experiencing increasingly frequent peak electricity consumption and power fluctuations. Following this trend, by 2028, we will have to use the electricity reserves that were originally reserved in advance and cannot be used to respond to emergencies. This situation poses a significant threat to the reliability of the power system, as the severity of the power shortage is likely to increase further, and we must take early action in a very short period of time.”

In most parts of the US, the growth in electricity demand stagnated before artificial intelligence, and people originally expected it to gradually rise as the economy electrified. Today, however, the boom in data centers is rapidly breaking this situation. The market predicts that by 2035, data center electricity demand will double, accounting for nearly 9% of the nation's total electricity demand, which will undoubtedly push the grid to the limit of its carrying capacity.

According to relevant information released by North American Electric Reliability Corporation (NERC) this month, the impact of data centers is already evident: they have caused a sharp jump in electricity demand in winter, and the risk of power outages has also increased significantly under extreme conditions. According to forecasts, peak electricity consumption this winter will be about 20 gigawatts higher than the previous winter, while electricity supply will only increase by 9.4 gigawatts during the same period. (Note that 1 gigawatt of electricity is roughly equivalent to the power generated by a conventional nuclear reactor.)

Schneider analyzed data from the North American Electric Reliability Company (NERC) and compared its predictions with the electricity demand of its own customers (including Nvidia, Vantage Data Centers, Compass Datacenters LLC, and tech giants such as Microsoft, Google, Meta), and the conclusion was consistent.

However, some techniques and strategies are currently being applied to mitigate grid congestion. In Texas, for example, its battery capacity is growing rapidly, which has mitigated to some extent people's concerns about the collapse of the power grid due to a surge in demand. Salou argues that priority should be given to deploying these “grid enhancement technologies” rather than large-scale new power generation and transmission projects. After all, the latter will be difficult to meet actual electricity demand in recent years.