AI's energy demands could boost CO2 emissions until 2030, but small modular reactors may offer a cleaner solution.
AI energy demands rising fast
The growth of renewable energy sources is not sufficient to meet the energy demands of artificial intelligence, which is expected to lead to an increase in CO2 emissions at least until 2030. The situation could finally change when small modular reactors (SMR) enter the market.AI driving electricity use
The race for artificial intelligence implies a race for energy production. According to the International Energy Agency (IEA), the rise of AI is expected to lead to a significant increase in electricity consumption by data centers worldwide, rising from 460 terawattheures (TWh) in 2024 to 1000 TWh in 2030 and even 1300 TWh in 2035.Fossil fuels still dominant
Currently, coal remains the dominant source in the electricity mix for data centers, producing 30% of the necessary electricity. Renewable energy sources, collectively, come in second place with 27% of this mix. Natural gas is in third place with 26% of the electricity needed for data centers. These figures are based on the actual source of the electricity powering the data centers, not potential agreements between hyperscalers and electricity suppliers.Nuclear potential with SMR
Nuclear energy produces only 15% of the required electricity. However, it is the growth of nuclear energy through small modular reactors (SMR) that could enable a real reduction in CO2 emissions associated with data centers.Renewables not enough
Renewable energy sources remain the fastest-growing means of electricity production and this trend is expected to continue until 2030. The IEA estimates that this increase should be an average of 22% between 2024 and 2030, allowing it to meet half of the increase in electricity needs for data centers. However, this will not be sufficient to meet these increasingly important electricity needs. Consequently, coal and natural gas are expected to meet 40% of the additional electricity needs until 2030, coinciding with a peak in emissions in 2030, reaching 320 million tons of CO2.SMR needed for emission cuts
The beginning of the democratization of small modular reactors (SMR) from 2030 onwards will be necessary to hope for a real decrease in CO2 emissions. It is only under this condition, and with the continued strong growth of renewable energy sources, that emissions should decrease with a decreasing reliance on coal. Indeed, the IEA has estimated the CO2 emissions associated with data centers to be 300 million tons of CO2 in 2035.Data centers still a small share
It should also be noted that despite a spectacular increase in electricity needs, data centers only represent 1% of the world's total electricity production today, and are not expected to exceed 3% in 2030.