The exponential growth in AI is contradicting with environmental sustainability, with the possibility of coal power persisting to maintain electricity demands. As nations across the globe push for net zero transition, AI technology’s immense electricity consumption, particularly generative AI, poses a challenge.
AI models, which were localized and small-scale a few years ago, are now used by millions of people globally, thereby leading to a massive rise in electricity consumption. Recent data reveal that AI’s power usage touches the scale of a small country. For instance, the BLOOM model consumed 433 MWh for training, whereas GPT-3 required a substantial 1287 MWh. To make it understandable, 1287 MWh could provide power to nearly 128,700 average households for a day or can lighten up more than 200,000,000 LED bulbs for an hour.
The ecological impact of AI is not limited to energy consumption, but extends to water usage, particularly in data centers. A 15-megawatt data center can use up to 360,000 gallons of water daily. Predictions hint that the electricity consumption of the US data center may triple by the end of this decade, accounting for about 7.5% of the country’s projected electricity demand. The same trend is expected in the EU, as it is felt that the energy demand from data centers will double by 2026.
Rapid adoption of AI in northern Virginia, renowned as “data center alley”, resulted in a pause in connections to new data centers due to over demand. Some regions had to halt plans to decommission coal plants due to the excess power needed for data centers and electric-vehicle battery factories. The risk of blackouts is high, especially if infrastructure improvements fail to keep pace.
The construction of semiconductor, EV, and battery factories, driven by legislation and incentives, are also causing a shoot in electricity demand. Even though technology firms and clean tech manufacturers prefer renewable energy, the reality of effectively offsetting this energy usage remains a challenge.
Global adoption and increasing demand of AI technologies is generating potential problems with the power grid’s ability to handle it. The question is, can a more efficient, refined, and intelligent AI help in overcoming the environmental challenges it creates? One promising solution could be the development of bio-inspired neuromorphic AI chips which simulate natural synaptic functions.
There is also room for an “AI tax” proposal, where entities benefiting the most from AI could contribute to reducing its environmental impacts. However, how these expectations are met and how these burdens are shouldered remains uncertain