Scholarlink

<p>The expansion of energy-intensive data centres to support the use of artificial intelligence (AI) has driven concerns about the resulting greenhouse gas emissions, pressure on power grids and stress on freshwater resources. In this Review, we examine the challenges of mitigating the growing environmental impacts of AI, and potential solutions. Hardware design choices, such as using recycled or older components, can reduce the overall emissions of data centres by 10–20% by reducing embodied emissions. Grid-integrated AI data centres can lower emissions across the AI model life&#xa0;cycle by aligning AI workload management with grid operating conditions, potentially reducing carbon emissions by about 10% in grids with high renewable-energy penetration. Water use by data centres can be negatively coupled with CO<sub>2</sub>-equivalent emissions, with methods of reducing water consumption increasing carbon emissions in some cases. These trade-offs highlight the need for strategies that balance energy and water usage, carbon emissions and societal needs when designing and managing AI infrastructure.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Strategies and design for increasing AI sustainability

  • 期刊论文

  • 发表日期：

    2026年7月9日

查看全文
  • Andrew A. Chien,
  • Udit Gupta,
  • Shaolei Ren,
  • Akshitha Sriraman,
  • Bill Tomlinson

摘要

The expansion of energy-intensive data centres to support the use of artificial intelligence (AI) has driven concerns about the resulting greenhouse gas emissions, pressure on power grids and stress on freshwater resources. In this Review, we examine the challenges of mitigating the growing environmental impacts of AI, and potential solutions. Hardware design choices, such as using recycled or older components, can reduce the overall emissions of data centres by 10–20% by reducing embodied emissions. Grid-integrated AI data centres can lower emissions across the AI model life cycle by aligning AI workload management with grid operating conditions, potentially reducing carbon emissions by about 10% in grids with high renewable-energy penetration. Water use by data centres can be negatively coupled with CO2-equivalent emissions, with methods of reducing water consumption increasing carbon emissions in some cases. These trade-offs highlight the need for strategies that balance energy and water usage, carbon emissions and societal needs when designing and managing AI infrastructure.