The AI buildout's carbon bill could undo 15 years of power sector emissions cuts

Data centers have become a contentious issue due to their impact on electricity prices, water supplies, and carbon emissions. A recent report from the Sierra Club estimates that planned US AI data centers could add 618 million metric tons of carbon emissions to the power sector between 2025 and 2030, effectively reversing 15 years of progress in reducing greenhouse gas emissions.

This would bring the sector's emissions to 2.1 billion metric tons by 2030, a 42% increase. Many data centers are now turning to alternative power sources to avoid waiting for existing grids, but these off-grid sources could add over 460 million metric tons of emissions, equal to Mexico's annual emissions. Estimates of the AI buildout's effect on carbon emissions vary widely, ranging from two to four times what the Sierra Club initially projected a year and a half ago.

The gap between the lowest and highest projections is even larger than the combined electricity demand of California and Florida. While some projections suggest that renewables could supply much of the new buildout, leaving a 24% reduction in power sector emissions by 2040 compared to 2025 levels, the actual implementation of these data centers is uncertain.

Earlier this year, Sightline Climate estimated that between one-third and half of the US data centers slated to open this year may be delayed or canceled. Struggles to meet AI's power demands have also slowed the deployment of data center GPUs. Major companies like Amazon have been affected, with some reportedly leaving idle GPUs until the data centers have the power to run them.

Even Amazon's planned Texas data center, which would be powered by a 7.65-gigawatt natural gas plant, could become the largest single source of CO2 emissions in the US. Data centers are increasingly controversial in the US due to concerns about electricity prices, noise, water pollution, and impacts on property values.