Will AI cause more problems for the environment in the next 6 years than solutions?
Wells Fargo is projecting that AI power demand will grow 8100% from 8TWh in 2024 to 652TWh in 2030. More than half of that is due to AI model training.
Google has already stated that AI data centers are derailing their carbon emission goals. And the 8100% increase has just begun.
The AI demand surge correlates in time with a surge in electric cars, continued Bitcoin draining of the grid, and an overall shift in the economy from carbon-based to renewable energy.
At the same time, scaling up renewable energy production takes years. A nuclear power plant takes around 6-8 years to build, but no new ones are being built in the US at this time.
It is also highly unlikely that the output of AI tools would contribute to any substantial impact that would pass regulations, get implemented, and have an actual impact in that timeframe.
The supply and demand factors of the economy imply that the energy price will rise when there is increased demand that is not met with increased production. Since energy is a part of most production, products will to some extent get a higher price.
If supply is met by carbon emission energy sources, the world’s emission goals will take a hit and the greenhouse effect can accelerate.
One trend that will probably continue is banning Bitcoin mining to save energy. Bitcoin uses much of the same resources as AI, like data centers, GPUs, and energy.
Optimally, AI can accelerate the development of solutions like a more energy-efficient cryptocurrency, thus freeing up much-needed energy resources.
However, in the near term, up to 2030, AI will probably have a negative environmental impact by driving up demand for energy. AI will likely not be able to make any large-scale contributions in that time that would result in a net positive effect on the environment.
It could very well be an investment that pays off in the long run. The problem for the environment is that net positive changes need to be made now.