Is Artificial Intelligence (AI) Accelerating or Hindering Global "Net Zero" Emissions?

According to Google's recently released "2024 Environmental Report," the company's greenhouse gas emissions in 2023 increased by a significant 48% compared to 2019,due to the expansion of data centers that support AI.

Another tech giant,Microsoft,also admitted in May of this year that its emissions have increased by nearly one-third since 2020,largely due to the construction of data centers.

However,some predictive studies suggest that AI has the potential to improve energy efficiency and aid in climate transition.Microsoft co-founder Bill Gates recently stated that although AI will increase global electricity demand by 2% to 6%,it can also "accelerate the reduction of electricity demand by more than 6%."

Kate Brandt,Google's Chief Sustainability Officer,also predicted that the company's emissions "will continue to rise before declining towards the target."

Xiao Fusheng,a partner at EY's Strategy and Transactions Advisory Services,told Yicai Global that the large amount of electricity required for the expansion of AI computing power is a major issue that cannot be avoided at present.However,AI technology is becoming a powerful driver for energy transition and energy consumption reduction,paving the way for building an energy-efficient future power grid through precise scheduling and demand response.

AI Threatens "Net Zero" Goals

According to data from the International Energy Agency (IEA),a single query using OpenAI's chatbot ChatGPT consumes 2.9 watt-hours,while a Google search only requires 0.3 watt-hours,about 1/10 of the former.

A study by the Brookings Institution states that generative models with extensive functionalities require much higher power consumption per query than traditional applications,and the hardware level exceeds that of task-specific computing systems.The broader the tasks people require the model to perform,the greater the energy and carbon emissions required.According to research by AI startup Hugging Face and researchers at Carnegie Mellon University,the energy required to generate an image can charge a smartphone.

At the same time,a recent report on AI safety supported by the UK government states that the carbon intensity of the energy sources used by technology companies is a "key variable" in the environmental cost of computing technology.However,a "significant portion" of AI model training still relies on fossil fuel energy.The Brookings Institution also stated that over time,the energy consumption associated with AI training will gradually increase rather than decrease.Training early chatbot models (such as the GPT-3 model) generated 500 metric tons of greenhouse gas emissions,equivalent to the travel distance of approximately 1 million miles by traditional gasoline-powered cars.The same model required over 1,200 megawatt-hours during the training phase,equivalent to the energy consumption of one million American households in one hour.Future iterations of large language models may continuously raise these metrics,with newer versions like GPT-4 demanding even more and producing higher carbon emissions.

The International Energy Agency (IEA) predicts that by 2026,depending on factors such as the speed of deployment and the efficiency of computing processes,electricity consumption related to data centers,cryptocurrencies,and AI could increase to between 620 and 1050 terawatt-hours.This means that global electricity demand could "increase by at least the equivalent of one Sweden,and at most one Germany."

Google stated in its latest report that there is "significant uncertainty" about the company's goal of achieving net-zero emissions by 2030,and the future environmental impact of AI is also "complex and difficult to predict." Microsoft also acknowledged this year that due to its strategies involving artificial intelligence and data center construction,the company's "moonshot" net-zero goal for 2030 may not be successful.

At the same time,concerns about the environmental impact of AI are not limited to energy or emissions.The rapidly growing AI infrastructure also requires land and water.According to data cited by British media,by 2027,AI's water consumption could reach as high as 6.6 billion cubic meters,nearly two-thirds of England's annual water usage.The Brookings Institution also stated that some water-intensive digital infrastructure projects in the southwestern United States have put pressure on communities that are already suffering from water shortages.

Google also acknowledged in its latest report that the company's data center electricity consumption will increase by 17% in 2023,accounting for 7% to 10% of the global data center electricity consumption.The water consumption of its data centers also increased by 17% compared to the previous year.

Are tech giants responsible for emissions reduction?

Bill Gates recently stated that large technology companies are "very willing" to pay extra for the use of clean energy to "demonstrate that they are using green energy."

Indeed,according to Amazon's official website,the company has more than 500 solar and wind projects worldwide,investing in over 100 last year alone,making it the largest corporate buyer of renewable energy for the fourth consecutive year.In January of this year,Microsoft also hired a nuclear technology director responsible for developing atomic reactors to power its data centers.In May of this year,Google also announced the signing of two new solar power purchase agreements (PPAs) in Japan,supporting the construction of new solar projects and adding 60 megawatts of clean energy capacity to the Japanese power grid.

However,according to Google's report,due to the termination of some clean energy projects in 2023,the amount of renewable energy available to the company has decreased.At the same time,the electricity consumption of the company's data centers "exceeded" Google's ability to launch more clean energy projects in the United States and the Asia-Pacific region.

At the same time,renewable energy may not be able to keep up with the pace of AI expansion.The IEA warns that although the global renewable energy capacity grew at the fastest rate in the past 20 years in 2023,according to current government plans,global renewable energy may only double by 2030.In addition,onshore renewable energy projects such as wind and solar can be developed in less than six months.However,planning and regulatory processes in many developed countries may add several years to this process.Offshore wind farms and hydropower schemes face similar challenges,with construction times requiring two to five years.Xiao Fusheng told reporters from First Financial that in the future,technology companies can start from aspects such as technological breakthroughs,resource sharing,and the use of new energy to reduce the energy demand of AI.However,all related solutions are not easy to achieve.

He said: "Technological breakthroughs require a lot of manpower,material resources,and capital support,and need to solve the commercialization of new chips and the problem of continuous performance improvement.Resource sharing requires ensuring the reliability and cost-effectiveness of computing power leasing services,and at the same time,it is necessary to establish an effective resource sharing mechanism.The use of new energy needs to overcome the intermittency and instability of renewable energy,and achieve intelligence and efficiency in energy management,while also considering the compatibility and transition issues with existing energy infrastructure."