Tracking the carbon footprint of global generative artificial intelligence

Dear Editor,

In recent years, generative artificial intelligence (GAI) has gained unprecedented attention. Unlike conventional AI, GAI can generate innovative and meaningful content across texts, images, and videos. The success of OpenAI’s ChatGPT has driven global tech companies to develop high-performance models and integrate GAI into products.1 This AI arms race continues, as shown by OpenAI’s text-to-video model, Sora, and Anthropic’s new large language model, Claude 3. The release of DeepSeek V3/R1 has sparked a global AI cost revolution. GAI is transforming key sectors like business, finance, law, and healthcare, heralding a new era in AI technology.

However, the widespread use of GAI has sparked concerns about potential energy crises and environmental impacts.2 At the Davos World Economic Forum, OpenAI chief executive officer Sam Altman highlighted the challenges of energy systems faced with advancing GAI. NVIDIA’s Jensen Huang and Tesla’s Elon Musk also emphasized the critical need for comprehensive energy management and the risks of shortages by 2025. Commentary suggests prioritizing energy efficiency and reducing unnecessary consumption. Some studies indicate that large AI systems may consume as much energy as entire nations, which conflicts with global carbon neutrality goals set for the mid-century.3 Despite the increased awareness of AI’s substantial energy consumption, there is a lack of quantified analyses and in-depth predictions for the energy consumption and carbon emissions induced by global GAI.4 Furthermore, the energy consumption and environmental impacts of vertical GAI models across various industries remain unclear. Addressing these issues is crucial for understanding GAI’s role in the global low-carbon transition and climate mitigation.