.png)
January 2, 2025
The rapid development of Artificial Intelligence (AI) is leading to an immense increase in energy demand in data centers worldwide. Especially in the USA, where many large AI companies and their data centers are located, this poses significant challenges for the power supply. In the vicinity of newly built data centers, which are primarily used for AI services, instabilities in the power grid are increasingly occurring, as an analysis by Bloomberg shows.
The investigation is based on data from around 770,000 sensors from Whisker Labs, which monitor power quality in US households in real time. The study, conducted in collaboration with market research institute DC Byte, reveals that more than half of the households with the strongest harmonics in the power grid – and thus the potentially largest disruptions to power quality – are located within a 20-mile (32-kilometer) radius of major data centers. Harmonics are high-frequency waves in the power grid caused by nonlinear loads. Simply put, the electrical voltage is distorted by various factors. Overloaded transformers, power supplies, and the use of power converter valves are considered common causes. The consequences can range from disruptions in the operation of other consumers to damage to household appliances and an increased fire risk due to voltage spikes.
The analysis evaluated by Bloomberg includes data from February to October 2024. Within a 50-mile (80-kilometer) radius of important AI data centers, more than three-quarters of the distorted power readings were even recorded. The consequences of poor power quality are manifold: flickering lights, voltage drops, and power outages can be the result. While grid operators and individual electricity providers question the results of the study and argue that their own measurements are within industry standards, experts warn of the long-term effects of the AI boom on the power supply. Bob Marshal, head of Whisker Labs, sees the harmonics as the first signs of an overloaded power grid. Hasala Dharmawardena of the Institute of Electrical and Electronics Engineers (IEEE) compares the impact of AI on the power grid to a "hammer." The power consumption of a single data center can be equivalent to that of 10,000 households.
AI data centers are being built at record speed, often within just 12 to 24 months. This rapid development makes planning and expanding the power infrastructure difficult. In addition, there are other factors that exacerbate the situation: the aging power infrastructure in the USA, the increase in extreme weather events, and the ongoing electrification of everyday life, for example through the increasing number of electric cars. In Germany, too, the government expects a doubling of electricity consumption by data centers by 2037.
In view of the increasing energy demand, large tech companies are already looking for solutions. Microsoft, Amazon, and Meta are planning to generate the energy for their AI data centers themselves or have it generated by partners. Nuclear power plays an important role in this. There are plans to reactivate decommissioned nuclear power plants or build new ones. This is intended not only to support climate goals, but also to secure the power supply for the energy-hungry data centers. The discussion about the future of energy supply in the age of AI is thus in full swing.
```