Concerns about the energy used by digital technologies are not new. Near the peak of the dot-com boom in the 1990s, a Forbes article lamented, “Somewhere in America, a lump of coal is burned every time a book is ordered online.” The authors of the article, which became widely cited in subsequent years in debates about energy policy, estimated that “half of the electric grid will be powering the digital-Internet economy within the next decade.” However, the estimate was wrong, with errors in both its facts and methodology. In hindsight, there is no longer any dispute, as the International Energy Agency (IEA) estimates that today’s data centers and data transmission networks “each account for about 1–1.5% of global electricity use.”
This mistake was not an isolated event. Numerous headlines have appeared over the years predicting that the digital economy’s energy footprint will balloon out of control. For example, as the streaming wars kicked off in 2019—with Apple, Disney, HBO, and others announcing video streaming subscription services to compete with Netflix, Amazon, and YouTube—multiple media outlets repeated claims from a French think tank that “the emissions generated by watching 30 minutes of Netflix is the same as driving almost 4 miles.” But again, the estimate was completely wrong (it is more like driving between 10 and 100 yards), resulting from a mix of flawed assumptions and conversion errors, which the think tank eventually corrected a year later.
With the recent surge in interest in artificial intelligence (AI), people are once again raising questions about the energy use of an emerging technology. In this case, critics speculate that the rapid adoption of AI combined with an increase in the size of deep learning models will lead to a massive increase in energy use with a potentially devastating environmental impact. However, as with past technologies, many of the early claims about the consumption of energy by AI have proven to be inflated and misleading. This report provides an overview of the debate, including some of the early missteps and how they have already shaped the policy conversation, and sets the record straight about AI’s energy footprint and how it will likely evolve in the coming years. It recommends that policymakers address concerns about AI’s energy consumption by taking the following steps:
- Develop energy transparency standards for AI models.
- Seek voluntary commitments on energy transparency for foundation models.
- Consider the unintended consequences of AI regulations on energy use.
- Use AI to decarbonize government operations.