Forecasting Gets Complicated
Screenshot_20250203-051451.png
DeepSeek's latest claim that artificial intelligence can make energy more efficient is just the latest reason why the energy industry needs to rethink all its assumptions about future demand. Through nearly 200 charts, clean energy expert Nat Bullard lays out a host of other surprising factors influencing decarbonization projections in his annual presentation published online.
Bullard, an advisor and co-founder of energy information platform Halcyon, says new information about rapidly developing fields like AI and diabetes drugs that could reduce food consumption adds layers of uncertainty to the outlook.
The world is already on a complicated path to net zero, according to Bullard, who previously held roles at BloombergNEF — including head of content — through 2023.
For example, 2024 may have been a record year for renewables, but demand for coal and gas is also up.
"We're burning more fossil fuels and emitting more CO2 than ever before," he says. "At the same time, we are producing and installing more wind turbines, solar panels and batteries than ever before. We live in complex times."
Here are some other highlights.
Data is not everything.
Global electricity demand growth is expected to accelerate in the coming years, but only a fraction of that demand is expected to be devoted to AI. In a scenario designed to reflect the current political landscape, the International Energy Agency showed in an October report that data centers will play a limited role in the increase in electricity demand between 2023 and 2030.
Screenshot_20250203-051434.png
AI has greater potential for efficiency
Data centers in the United States and Europe consume electricity and raise concerns about their climate footprint. In Virginia, home to more than 300 centers, the industry will account for about a quarter of electricity consumption in 2023, while in Ireland the sector will use more electronics than urban households, Bullard says.
But all that could change. Chinese AI startup DeepSeek says its basic open-source models require only a fraction of the training hours required by equivalent U.S. versions.
Bullard says the development likely means a wave of infrastructure assets like power plants and data centers in the U.S. could peak and then recede.
[attachment deleted by admin]