On 5 June, the United Nations University released yesterday the report of the Institute for Water, Environment and Health, which called artificial intelligenceInfrastructureRapid expansion is increasing significantlyrenewable energyWater resources and land needs。
THE REPORT ESTIMATES THAT THE GLOBAL DATA CENTRE WILL CONSUME 448 WATT-HOURS IN 2025; BY 2030, THE DATA CENTRE ' S ELECTRICITY DEMAND MAY INCREASE TO 945 WATT-HOURS (1 WATT-HOUR = 1 BILLION DEGREES OF ELECTRICITY), REPRESENTING APPROXIMATELY 31 TP3T OF THE TOTAL GLOBAL ELECTRICITY CONSUMPTION。
The report lists “debate” as the main source of energy consumption. Research shows that the reasoning chain accounts for 80% to 90% of the total energy consumed by artificial intelligence; ChatGPT handles an estimated 2.5 billion alert requests per day, corresponding to annual electricity consumption of about 383 giva。
Different AI JOB ENERGY CONSUMPTION ALSO VARIES GREATLY: 1450 TIMES MORE ENERGY-INTENSIVE THAN BASE TEXT CLASSIFICATION TASKS TO GENERATE AN AI IMAGE AND A SHORT VIDEO EQUIVALENT TO 200,000 SPAM SORTING MISSIONS。
The report predicts that by 2030, the relevant water use for global data centres (including direct cooling and indirect power generation/manufacturing) will reach 9.3 trillion litres, equivalent to the basic water needs of 1.3 billion people a year in sub-Saharan Africa。
THE REPORT ALSO MENTIONS THAT ONLY 32 COUNTRIES WORLDWIDE CURRENTLY HAVE DEDICATED AI DATA CENTRES AND THAT MORE THAN 90% AI DEDICATED CLOUD COMPUTING RESOURCES ARE CONCENTRATED IN THE UNITED STATES AND CHINA。