AI and the energy problem

Robot artist Ai-Da paints at the Global Artificial Intelligence Summit in Geneva, Switzerland, May 30, 2024. (Photo: THX/TTXVN)

Robot artist Ai-Da paints at the Global Artificial Intelligence Summit in Geneva, Switzerland, May 30, 2024. (Photo: THX/TTXVN)

A new report finds that the world’s leading tech companies’ commitment to carbon neutrality is increasingly being questioned as the rise of AI drives energy demand to spiral out of control. Chatbots like OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Facebook’s Llama are software applications, but they require a network of supercomputers around the world to operate. Every time a user asks a question, millions of calculations are performed in data centers, consuming huge amounts of electricity.

A study by MIT Technology Review found that training a large AI model can consume more energy than the average small city consumes in a year. For example, training OpenAI’s GPT-4 AI model consumed as much electricity as 175,000 American homes consumed in a day. Apple, Google, and Meta have set a goal of achieving net zero emissions by 2030, while Amazon has set a target of 2040, and Microsoft has said it will achieve net zero emissions by the end of the decade. But analysts say those claims were made before the AI boom and are now becoming increasingly out of touch with reality.

Thomas Day, one of the authors of the report, published by Carbon Market Watch and the NewClimate Institute, said that the climate goals of tech companies are losing their meaning. If energy consumption continues to escalate without adequate controls or oversight, the chances of achieving the goals are very low.

The report rated the integrity of the climate strategies of large corporations such as Meta, Microsoft and Amazon as poor, while Apple and Google were rated average. In terms of the quality of their emission reduction targets, Meta and Amazon were rated as very poor, while Google and Microsoft were rated as poor. Only Apple was rated higher. The main reason for the sharp increase in emissions is the expansion of AI operations and the accompanying data center systems, which consume huge amounts of electricity. Over the past 3-4 years, the electricity consumption and corresponding carbon emissions of some companies have doubled or even tripled.

According to the report, the operating emissions of the world’s 200 largest technology companies reached nearly 300 million tons of CO2 in 2023. If the downstream value chain is included, this figure could be nearly five times higher. If the technology industry were a country, it would rank fifth globally in greenhouse gas emissions, higher than Brazil.

The International Energy Agency (IEA) says that electricity supplied to data centers has increased by an average of 12% per year from 2017 to 2024, and is expected to double by 2030. However, the majority of electricity is still not coming from renewables, despite ambitious claims from companies.

It’s worth noting that about half of the processing capacity in data centers is now run by subcontractors, but many companies don’t include their emissions in their official calculations. The equipment and infrastructure supply chain, which accounts for at least a third of emissions, is also often overlooked. While investments in renewable energy are increasing, they’re not enough to offset the industry’s growing electricity consumption, says Thomas Day.

The report points out that, given that AI is seen as a driver of economic growth and a strategic tool in industrial policy, it is unlikely that governments will intervene to limit the industry’s growth. However, experts believe there is still room for improvement. Ensuring data centers run on renewable electricity, extending the life of equipment, and increasing the proportion of recycled materials in hardware manufacturing can all help reduce emissions.

According to BINH MINH/Nhan Dan Newspaper

Original article link

Source: https://baovanhoa.vn/nhip-song-so/ai-va-bai-toan-nang-luong-147512.html