The Dark Side of AI: Energy Consumption and Resource Depletion

Meta's plans to build hundreds of datacentres, some rivaling the size of Manhattan, have ignited concerns about the energy requirements of artificial intelligence. According to Meta's CEO, Mark Zuckerberg, the company will spend hundreds of billions of dollars on these projects. However, the focus has shifted from reducing carbon emissions to securing clean energy sources as AI's power hunger becomes increasingly apparent. The tech giants, including Google, Microsoft, and Meta, are racing to secure clean energy agreements, such as Google's $3 billion deal to modernize hydropower plants in Pennsylvania and Meta's 20-year deal with a nuclear plant in Illinois. Despite these efforts, a US energy department report warns that power outages could increase 100 times by 2030 if fossil fuel plants due to be shut are not kept on the grid to support AI.

Key Takeaways:

  • AI's power consumption is projected to double by 2030, surpassing the entire electricity consumption of Japan.
  • A typical ChatGPT query uses 0.34 watt-hours of electricity, equivalent to about 1/15th of a teaspoon of water.
  • Tech companies are racing to secure clean energy sources, with Google's $3bn deal and Meta's 20-year deal being notable examples.
  • The closure of coal-fired plants in the US has been delayed to meet the demand for energy caused by AI.
  • Investigations have revealed that datacentre consumption is already causing water shortages in areas such as Dalles, Oregon, and Newton County, Georgia.
  • There are no requirements for tech companies to disclose the resources used by AI, leading to a big variation in estimates.
  • The International Energy Agency has warned that electricity demand from datacentres will increase significantly by 2030.

Statistics:

  • 355m gallons of water consumed by Google's datacentres in Dalles, Oregon, in just one year (equivalent to 29% of the water supply).
  • 100 times increase in power outages by 2030 if fossil fuel plants due to be shut are not kept on the grid to support AI (according to the US energy department report).
  • $3 billion (PS2.2bn) - the value of Google's deal to modernize hydropower plants in Pennsylvania.
  • 20 years - the duration of Meta's deal with a nuclear plant in Illinois.
  • 2030 - the year by which AI's power consumption is projected to double.

Sources:

  • "Google's $3bn deal to modernize Pennsylvania hydropower plants" (BBC News, undated).
  • "Meta's 20-year deal with a nuclear plant in Illinois" (The Verge, undated).
  • "US energy department report warns of power outages" (The Guardian, undated).
  • "International Energy Agency warns of datacentre electricity demand" (The Telegraph, undated).
  • "ChatGPT query uses 0.34 watt-hours of electricity" (OpenAI, undated).
  • "Google's datacentre consumption causes water shortages" (OregonLive, undated).