John Sterman and Jennifer Turliuk estimate the direct effect of AI at 0.1°C by 2100. The rebound effect would increase the projected warming from 3.3°C to 3.6°C.
The debate on the imprint of artificial intelligence almost always focuses on the electric invoice of the computing centres. An analysis conducted at MIT Sloan by Professor John D. Sterman and Jennifer Turliuk concludes that this scoping is missing the essentials.
The direct figures are modest: IA and data centers absorbed about 1.5% of global electricity in 2024, accounting for about 0.5% of energy-related CO2 emissions, or about 0.1°C of additional warming by 2100. On this scale, improving energy efficiency makes little difference to the result.
The rebound effect changes the order of magnitude
The main risk is indirect. If AI keeps its promises of productivity, it accelerates economic growth — and hence energy consumption and its associated emissions. Based on a global gross product raised by 25%, but below the most optimistic projections, the expected warming in 2100 increases from 3.3°C to 3.6°C above the pre-industrial level.
The authors also note a time lag: emissions are immediate, the climatic benefits of AI innovations will come much later. Sterman's straight formula — Greenland and Antarctica will not freeze themselves while waiting. Finally, they point to accounting blind angles, such as redirecting the production of an existing nuclear power plant to a data center: emissions increase elsewhere on the network without appearing in the company's balance sheet. Their conclusion is political rather than technical: carbon pricing, mandatory publication of environmental impacts, incentives for efficiency and renewables.
Read the full article on the original source
Go to the source →