The Banking Executive Magazine - April 2025

ISSUE 196 APRIL 2025 the BANKING EXECUTIVE 41 FinTech and AI Chornicle According to figures reported from 2024, global data centres already consume around 415 terawatt-hours (TWh) annually, representing ap- proximately 1.5% of the world’s total electricity usage. The growth rate is a steady 12% per year. Though AI’s direct slice remains relatively modest at around 20 TWh for now, its trajec- tory is steep and unrelenting. Ana- lysts predict that by 2027, the global power requirement for AI data cen- tres will reach 68 gigawatts (GW)— on par with California’s entire grid capacity as of 2022. Looking ahead to 2030, forecasts from OPEC and leading investment firms suggest global data centre consumption may rise to 945 TWh, or even triple to 1,500 TWh—nearing 3% of all elec- tricity produced worldwide. Such numbers are not speculative extremes. They stem from concrete developments: the scale of AI train- ing, the infrastructure race among tech giants, and the sharp increase in generative AI adoption across sec- tors. The energy consumed to train a single advanced model, like GPT-4, can be as much as 50 times higher than that required for GPT-3, which itself used over 1,200 megawatt- hours—enough to power hundreds of homes for a year. Even more strik- ing is the fact that over 80% of an AI system’s energy footprint comes not from training, but from its ongoing use, as millions of users interact with chatbots, recommendation systems, and automation tools. A STRAIN ON GLOBAL RE- SOURCES Energy consumption is only the tip of the iceberg. AI’s environmental im- pact casts a broader shadow, involv- ing freshwater depletion, critical mineral extraction, and electronic waste proliferation. Data centres re- quire constant cooling, often achieved using vast amounts of water—up to 1.7 litres per kilowatt- hour. Google, for example, reported that its data centres consumed over five billion gallons of freshwater in 2022 alone—a 20% increase from the previous year. Projections indi- cate that the global AI infrastructure may soon outstrip the annual water consumption of entire countries like Denmark. The hardware arms race is equally alarming. As AI chips evolve rapidly, obsolete equipment is discarded at a staggering pace. By 2030, AI-related e-waste from data centres could reach five million tons annually. Each AI chip consumes an estimated 1,400 litres of water and 3,000 kWh of electricity to manufacture—inten- sifying reliance on energy-heavy semiconductor factories and global supply chains for rare earth minerals. For countries in the MENA region, many of which already contend with water scarcity and fragile ecosys- tems, these trends carry strategic im- plications. THE ROLE OF RENEWABLE AND NUCLEAR ENERGY Meeting the energy appetite of AI will require a fundamental recalibra- tion of how the world produces and distributes electricity. Renewables— solar, wind, geothermal, and hydro— will play a vital role, but limitations remain. Intermittency is a major ob- stacle: data centres require consis- tent, around-the-clock power, while renewable output fluctuates with weather and geography. The integra-