During the past several years there has been a significant push toward electrification, from EV vehicles to heating and cooling systems that use greater amounts of electricity, and reduction of fossil fuel to generate electricity with the use of more renewable sources of generation to replace it. More recently, Artificial Intelligence (AI) exploded into our lives, and its demand for electricity is significant and expected to grow at a rapid pace.

Two months after OpenAI’s ChatGPT was released it had 100 million active users. The results of an AI search can be very impressive, but it comes with an energy consumption price tag. For example, a single ChatGPT query uses 2.9 watt-hours of electricity. In comparison, a Google search requires nearly one tenth the energy at 0.3 watt-hours. A typical cell phone uses about 5 watt-hours to charge, about the same amount of energy it takes AI to generate a single image.

The International Energy Agency is estimating that electricity consumption from AI, data centers and the cryptocurrency sector could double by 2026. In 2022, global data centers used an estimated 460 tera-watt hours (TWh) in 2022, potentially reaching more that 1,000 TWH in 2026, which is approximately equivalent to the electricity consumption of Japan.

What causes AI to use so much energy?

There are basically two phases of an AI search, training and inference. In the training phase, the model learns from a large set of data to discover patterns and relationships. From that, the inference phase applies those patterns to generate predictions and generate content.

Essentially housed in data centers, AI has to run enormous numbers of calculations quickly on specialized equipment, with image generation performed on graphical processing units. Those processing units generate heat and need to be cooled, in a clean environment and must operate continuously, 24 hours per day, 365 days per year.

Currently, 20% of the data centers in the US are in northern VA. Retrofitting conventional data centers for AI is often cost-prohibitive as they typically aren’t large enough and don’t have access to sufficient power to handle increasing rack energy density, and cannot accommodate advanced cooling technologies. The rapid expansion of datacenters to meet growing demand to accommodate AI will expand across the country as companies seek out any advantages they can to minimize the impacts to needed resources, minimize energy costs and environmental impacts, at the same time very likely putting a strain on the US power sector’s infrastructure in the coming years.

Based on analysis of available disclosures from technology companies, public data center providers and utilities, and data from the EIA, the research team at Barclays estimates that data centers account for 3.5% of US electricity consumption today, and data center electricity use could be above 5.5% in 2027 and more than 9% by 2030.

Can AI have a positive effect on our grid?

While the demand for energy use of AI may push our grid to its limits, it may very well prove to be extremely useful in helping clean energy to meet that demand. In April 2024, DOE released a report outlining how AI can accelerate the development of a 100% clean electricity system. The DOE fully recognizes “the need to build clean energy projects at speed and scale, and is currently developing AI tools to improve the way such projects are sited and permitted at the Federal, state, and local levels. This includes tools to put decades of previously inaccessible environmental data in the hands of scientists and government reviewers to identify opportunities and accelerate decisions.”

Some interesting key points where AI can help to optimize the grid include:

• Data analysis: AI can process vast amounts of data from sensors, smart meters, and other sources to detect anomalies, predict future energy consumption patterns, and identify potential issues within the grid.
• Grid optimization: By analyzing real-time data, AI can optimize power distribution, manage load balancing, and reroute electricity to areas with high demand, improving grid stability and efficiency.
• Renewable energy integration: AI can predict fluctuations in renewable energy generation from sources like solar and wind, allowing for better grid management and integration of variable energy sources.
• Predictive maintenance: AI can be used to predict equipment failures in power plants and transmission lines based on historical data, allowing for preventative maintenance and minimizing downtime.
• Demand response management: AI can analyze consumer usage patterns and incentivize them to adjust energy consumption during peak demand periods, reducing strain on the grid.
• Smart home energy management: AI-powered smart thermostats can learn user habits and optimize heating and cooling schedules to minimize energy consumption.
• Virtual power plants: AI can aggregate distributed energy sources like rooftop solar panels into a virtual power plant, allowing them to be managed as a single entity to respond to grid fluctuations.
• Energy trading optimization: AI can analyze market data to predict price trends and optimize energy trading strategies.

Just as the grid has been significantly impacted by renewable resources including solar and wind, there is no doubt that AI will transform the energy industry, and has already begun to do so. In the short term, it will require a combination of renewable and non-renewable generation, but may help to accelerate the conversion to a greater amount of renewable resources.