The energy dilemma

We can all see the energy consumed by artificial intelligence—simplified versions of which we interact with daily on our smartphones. We constantly send queries to our devices, and through data processing, we receive increasingly complex responses. Behind this interaction lie data transmission networks and vast storage systems, all of which require electricity to function. Even though our contribution to energy consumption is small, we still add to it when we recharge our phone’s battery each night. The most modern smartphones can store around 20 watt-hours of energy, enough to last a full day under normal use. To put this into perspective, a 20-watt LED bulb, which produces as much light as a 120-watt incandescent bulb, consumes 20 watt-hours in just one hour.

 

Meanwhile, a complete wash cycle in a washing machine or dishwasher uses between 1,000 and 2,000 watt-hours.

The energy consumed per operation on our devices is minimal, considering we touch our phones over 2,000 times a day. However, behind what appears on our screens is a vast flow of data and the processing power of massive data centers. A race is underway to secure new electricity supplies for these centers as the exponential growth of data processing increases both energy consumption and the demand for cooling systems to keep the machines running. For example, a simple Google search uses about 0.3 watt-hours of energy in a data center. As the volume of these searches—and all other web-related activities—soars, electricity consumption is rising rapidly. A single query with ChatGPT’s AI, which requires far more memory management and processing, can consume up to 3 watt-hours—ten times that of a Google search. Artificial intelligence has become the new frontier for investment, but, perhaps unexpectedly, the availability of reliable electricity is becoming increasingly scarce.

 

 

 

Reliable electricity is in short supply

 

Currently, global electricity consumption for data management is estimated at around 400 billion kilowatt-hours—2 percent of the total—exceeding Italy’s entire consumption of 300 billion kilowatt-hours. By 2030, forecasts predict a 5- to 40-fold increase, potentially pushing data-related electricity consumption to 20 percent of the global total. All of this additional energy will need to be supplied by systems that can provide stable power around the clock, not just when the wind is blowing or the sun is shining. This distinction is crucial, as the Big Five tech companies—Meta (Facebook, Instagram, WhatsApp), Alphabet (Google), Amazon, Apple, and Microsoft—have pledged to use only renewable energy within a few years. Their vast financial resources, buoyed by substantial profits, and their global reach underscore their genuine commitment to addressing climate change.

 

However, they must also prioritize their shareholders. As a result, they have made massive investments in artificial intelligence—nearly $100 billion in the first half of 2024 alone. But they now face a significant challenge: the supply of electricity to power the ever-growing data centers is not enough.

 

Electric utilities themselves admit that they cannot meet connection requests within the required timeframes—often not until several years down the line, sometimes beyond 2030—while investors are pushing for results within months. As a result, Big Tech, with some discomfort, is forced to strike deals with operators of older power plants, whether coal-fired or even nuclear. One notable example is Microsoft’s agreement in September 2024 with Constellation, the company that owns the infamous Three Mile Island nuclear plant, the site of a major radiation accident in March 1979. The accident occurred in reactor two, but the deal involves reactor one, which remained operational for another 30 years until 2019. Now, driven by the demand for artificial intelligence, reactor one will be modernized and restarted with a $1.5 billion investment. A portion of the energy generated will supply Microsoft’s large data center in nearby Virginia. Meanwhile, both companies will also explore the development of small modular reactors, which are being heavily relied upon to revive the global nuclear industry.

 

Openai focuses on fusion

 

The most striking initiative, however, involves ChatGPT itself, or rather, the company behind it, OpenAI. It is partnering with Helion, one of the most prominent start-ups in nuclear fusion—a challenging technology that mimics the sun’s energy-producing process on Earth and, unlike nuclear fission, does not produce hazardous waste. The decision to explore nuclear fusion reflects the urgent need for reliable energy supplies while also honoring commitments to reduce emissions. By pursuing groundbreaking innovations like fusion, this collaboration underscores the role that advanced technologies—both human and artificial—will play in the energy transition. While the expectations for fusion are immense, they remain unproven, but that is a story yet to unfold.