A new report from The Conference Board analyzes the potential that the rise of AI will fuel a major increase in electricity demand.
"The rise of AI and data centers presents both opportunities and challenges for the energy sector," said Alex Heil, Senior Economist at The Conference Board. "While AI can drive efficiencies in electricity use, the increased demand from data centers could strain our existing infrastructure. It's crucial for businesses to manage their power demand and for policymakers to address grid constraints to ensure a reliable, sustainable energy future."
Among the report's key insights:
Power demand from data centers is projected to significantly increase.
- Analysts project that power demand from data centers could more than double over the next decade.
- This increase is expected to place significant pressure on the existing electricity grid, potentially creating outages and reliability issues.
- The expansion of data centers—and some $500 billion of investment in industrial and manufacturing facilities since 2021—has already substantially upgraded forecasts for overall US power usage.
Nationwide, electricity demand is now expected to grow 4.7% over the next 5 years—nearly double the previous forecast for 2.6% growth.
Despite previous predictions, dramatic increases in energy demand did not materialize as expected, the Conference Board said.
“The past two decades have seen relatively stagnant electricity demand, despite corresponding with dramatic increases in smartphone and internet usage.”
In 2007, the EPA estimated that energy consumption in the US by data centers and related technologies would double between 2005 and 2010, but the actual increase was only 36%.
Advances in energy efficiency, such as LED lighting and improvements in technology, combined with the decline in demand due to the 2008-2009 recession, helped mitigate the predicted spikes in energy demand, the Conference Board said.
“AI will seriously tax utilities and the grid in the decade ahead—but also has the potential to unlock significant efficiencies in electricity use,” it said.
Current-generation AI models like ChatGPT now use over 500,000 kWh per day to operate -- compared to 29 kWh for the average US household.
AI's computational power is projected to double every 100 days, increasing global energy demand attributed to AI by 26% to 36% annually in the coming years.
Implementing AI to manage electricity usage -- for instance, by shifting the most taxing tasks to off-peak hours -- can result in efficiency savings, reducing costs and enhancing the capacity of energy systems, the Conference Board said.
“If the added computational power of AI allows businesses to achieve greater efficiencies and utilities to operate more effectively when managing their own operations, then the increase in power demand due to AI could be lower than projected—as happened with the rise of smartphones and the internet in the 2000s and 2010s—and support climate goals.”
