AI and energy dilemma

The rapid rise of artificial intelligence (AI) has brought about remarkable technological advances, reshaping the world in unprecedented ways. However, scientists are raising concerns that this booming industry consumes an enormous amount of electricity, potentially becoming a key driver of global energy resource depletion.

Artist robot Ai-Da paints at 2024 AI For Good Global Summit, Switzerland, May 30, 2024. (Photo: Xinhua/VNA)

A recently released report casts doubt on the carbon neutrality commitments of major tech corporations, as the explosive growth of AI pushes energy demand beyond manageable limits. Chatbots like OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Meta’s Llama may be software applications, but they rely on a vast, constantly operating global network of supercomputers. Each user query triggers millions of computations in data centres, consuming immense amounts of power.

According to research by MIT Technology Review, training a large AI model can consume more energy than an average small city does in a year. For instance, the training of GPT-4 by OpenAI used electricity equivalent to the daily consumption of 175,000 US households. Apple, Google, and Meta have pledged to reach net-zero emissions by 2030, while Amazon aims for 2040, and Microsoft targets the end of this decade. However, analysts warn these commitments were made before the AI boom and now appear increasingly unrealistic.

Thomas Day, one of the authors of a report by Carbon Market Watch and the NewClimate Institute, said tech companies’ climate goals are losing relevance. Without better oversight and control, soaring energy demand will make these targets nearly impossible to achieve.

The report rated the climate strategy integrity of major firms like Meta, Microsoft, and Amazon as poor, while Apple and Google received average ratings. In terms of emission reduction targets, Meta and Amazon were ranked very poor, Google and Microsoft poor—only Apple performed better. The primary reason for the emissions surge is the rapid scaling of AI operations and associated data centres, which consume vast electricity. Over the past 3–4 years, electricity use and corresponding carbon emissions of some firms have doubled or even tripled.

In 2023 alone, the operational emissions of the world’s 200 largest tech companies reached nearly 300 million tonnes of CO2. When downstream value chains are included, this figure could be nearly five times higher. If the tech industry were a country, it would rank fifth globally in greenhouse gas emissions—higher than Brazil.

According to the International Energy Agency (IEA), power supplied to data centres has grown at an average annual rate of 12% from 2017 to 2024 and is projected to double by 2030. Yet, despite ambitious corporate pledges, most of this electricity does not come from renewable sources.

Notably, around half of current data centre computing capacity is operated by subcontractors. However, many companies exclude their partners’ emissions from official calculations. Equipment supply chains and infrastructure, which account for at least one-third of total emissions, are also frequently overlooked. According to Thomas Day, although investments in renewable energy are rising, they remain insufficient to offset the tech sector’s accelerating power consumption.

The report highlights that, given AI’s role as an economic growth driver and a strategic tool in industrial policy, government intervention to curb the sector’s expansion is unlikely. Nonetheless, experts believe there is room for improvement. Ensuring data centres are powered by renewable sources, extending equipment lifespan, and increasing the use of recycled materials in hardware production could all help reduce emissions.

NDO