Who said artificial intelligence would help businesses become more energy efficient? Lots of misinformed people. In fact, artificial intelligence – and its resulting applications – are so energy-intensive that it becomes difficult to cover costs and faithfully meet everyone's needs. In January 2024, the International Energy Agency (IEA) published its annual report on electricity analyzes and forecasts until 2026. And the least we can say is that the figures are rather worrying.
In 2022, data centers, cryptocurrencies and artificial intelligence consumed approximately 460 TWh of electricity worldwide, or almost 2% of total global electricity demand. “Electricity consumption of data centers, artificial intelligence and the cryptocurrency sector could double by 2026,” indicates the international agency, specifying that this total electricity consumption of data centers could reach more than 1000 TWh in 2026. To give a more concrete image, this demand is approximately equivalent to the electricity consumption of Japan.
Advertisement
Tech giants, the first affected
Market trends, including the rapid integration of AI into software programming across various industries, are increasing overall data center electricity demand, the IEA finds. And the tools developed in recent years by technological giants such as Amazon, Google or Microsoft do not really move in the direction of energy sobriety.
“Search tools like Google could see their electricity demand increase tenfold if AI were fully integrated, states the report. Comparing the average electricity demand of a typical Google search (0.3 Wh of electricity) to OpenAI's ChatGPT (2.9 Wh per query), and considering 9 billion searches daily, this would require nearly 10 TWh of additional electricity in a year.
Microsoft or the typical example of tech struggling with its climate commitments
Microsoft is also in the grip of an internal struggle, torn between its massive investments in AI and its climate commitments. These, announced in 2020some time before the boom of AI among the general public, are ambitious to say the least. “By 2030, Microsoft will be carbon negative, and by 2050, Microsoft will eliminate from the environment all the carbon that the company has emitted either directly or through electricity consumption since its founding in 1975.”
The plan is bold. Achievable, less so. The Verge reports as well as the Redmond firm's greenhouse gas emissions were actually about 30% higher in fiscal year 2023. After investing more than $13 billion in OpenAI, Microsoft started working hard seriously on the development of proprietary tools made quickly available to developers and the general public.
Advertisement
All of the Copilot functions announced in recent months are part of this strategy. Only problem: if this work pays off to stand out on the market, the underside of the plan is less glorious. Microsoft pumped out 15.357 million tons of carbon dioxide in its last fiscal year, comparable to the annual carbon pollution of Haiti or Brunei.
Behind the AI, overheating servers
In its study, the IEA also points the finger at other actors who contribute to the overheating of the planet. “The electricity demand for AI can be forecast more comprehensively based on the quantity of AI servers expected to be sold in the future and their rated power.” And in this market, one player dominates: Nvidia. With an estimated market share of 95%, the chip giant has delivered 100,000 units in 2023 which consume an average of 7.3 TWh of electricity per year.
And as the International Energy Agency repeats, “By 2026, the AI sector is expected to grow exponentially and consume at least ten times its 2023 demand.” In its report, however, the IEA provides some ideas: “Updated regulations and technological improvements, particularly in efficiency, will be key to moderating the increase in data center energy consumption.”
The EU is stepping up to better manage energy consumption
For example, in Europe, the European Commission's revised Energy Efficiency Directive includes regulations applicable to the European data center sector, promoting more transparency and accountability to improve electricity demand management. As of this year, operators have reporting obligations for the energy consumption and emissions of their data centers, and large-scale data centers are required to have waste heat recovery applications, where appropriate. technically and economically feasible, while respecting climate neutrality by 2030.
A previous European regulation, applicable since 2020, sets efficiency standards for data centers allowing better control of their environmental impact. A European self-regulatory initiative created in 2021, called the Climate Neutral Data Center Pact, sets targets to achieve climate neutrality in the sector by 2030. More than 60 data center operators have signed the pact, including major operators like Equinix, Digital Realty and Cyrus One.
The United States and China: each has its own method
In the United States, the Energy Act of 2020 requires the federal government to conduct studies on energy and water use in data centers, in order to create enforceable energy efficiency measures and best practices that drive efficiency, as well as public reporting on historical energy and water usage in data centers. At the state level themselves, regulators in Virginia and Oregon – where a large number of data centers are based – have already mandated best practices in sustainability and carbon reduction.
Chinese regulators will require all data centers acquired by public bodies to improve their energy efficiency and be powered entirely by renewable energy by 2032, starting with a 5% share for renewable energy in 2023 .
Do you want to stay up to date on the latest news in the artificial intelligence sector? Register for free to the IA Insider newsletter.
Selected for you