A spate of articles have proclaimed that demand for electricity will surge, perhaps faster than our capability to bring on more generating capacity. Concerns about electric reliability are well-founded, but the demand side of the equation is much overstated.

Goldman Sachs predicts electricity demand growing by 2.4% per year after over a decade of zero growth. Key factors frequently cited on the demand side include the growth of cloud computing, the growth of artificial intelligence, and electrification of cars, heating systems, and other energy uses. But the actual data to date show virtually no growth. Since 2008, demand is up just 1.4 percent. That’s the total increase over 16 years, not the average gain. And even that 1.4 percent is within the range of normal variation in the period.

The key error being made is adding up new electric uses without subtracting old practices that are being replaced. This applies to cloud storage and AI, but not electrification of transportation and heating. Reports from the International Energy Agency, academics, and investment companies make this error.

The current era of flat electricity demand came despite the emergence of cloud data storage. Data centers used for cloud storage certainly consume a lot of electricity, but they may not add to net electricity demand. Cloud storage often replaces onsite data storage. In fact, a large cloud data storage facility may be more efficient in its electricity usage than storage scattered across multiple locations. New technologies, Goldman noted, have so far kept data centers’ electric demand level despite large increases in usage. Local storage the old way may involve multiple users storing the same data, and smaller storage units may not be as focused on electricity efficiency as a larger unit.

Artificial intelligence also uses a great deal of electricity, both for model training as well as inference (the use of the model to provide information to users). But we cannot simply add up the electric demand from AI; we must also subtract uses that will be reduced by AI. For example, one widespread use being rapidly adopted is AI for customer service. When the AI can answer a customer’s question, then no customer service rep sits in front of a screen, which is connected to a computer, which is linked to a database, all in a heated building. A 15-second AI response may replace five minutes of the rep’s time. Which solution uses more electricity?

In manufacturing, early use cases include predicting maintenance needs and supply chain management. In both activities, more efficient factory utilization will likely reduce electricity needs per unit of production. AI may provide similar benefits in construction, utilities, and other physical-world activities.

The entire electric system wastes some power in standby capacity that is sometimes excessive. We also have transmission losses. AI may be able to reduce these.

Energy management in buildings currently lags what good AI can accomplish. The same is true for equipment usage across a wide range of activities.

Transportation routing with AI can lower miles driven by both EVs and internal combustion vehicles. Current map applications work well for single trips, but AI increasingly reduces miles driven for multi-stop trips such as deliveries.

We don’t know for sure the net additional electricity and energy demand from AI yet, but it’s likely that in many cases AI will reduce demand. Adding up the new uses will certainly overstated net demand.

The third expected trigger for increased demand is the electrification of travel and heating. Electric vehicle use will very likely increase, unless public policy to discourage fossil fuels reverses. Similarly, use of heat pumps rather than furnaces and electric stoves rather than natural gas will boost electricity demand where governments push for these changes. Absent political actions, though, people will probable stay with current energy sources for years to come.

Shifting the economy from fossil fuels to renewables stresses the current system. Right now we have evidence that electric reliability is suffering because we are trying to end usage of old fuels faster than the new technologies develop. A reliable electric system based on renewables needs a better grid than we currently have and better energy storage technology. Although improvement is likely to come, shifting to renewables too soon will lead to more power failures in the interim.

In sum, the electricity demand forecasts are probably much overblown, though concerns about electricity supply reliability are warranted.

More By This Author: