2025年发展可持续生成式人工智能（GenAI）-实践路线图及其环境影响研究报告

Developing Sustainable Gen AI Report Summary

Core Findings & Challenges

1. Environmental Impact:

   • Gen AI has a substantial environmental footprint with high energy consumption, water usage, and e-waste generation throughout its lifecycle (model training, inferencing, etc.). Training large models like GPT-4 can consume electricity equivalent to powering ~5,000 US homes for a year.

2. Organizational Awareness:

   • Only 12% of surveyed executives actively measure Gen AI's environmental footprint, while 74% rank performance, scalability, and cost as top considerations instead of sustainability.

3. Emissions Impact:

   • 48% of executives believe Gen AI has increased their organization’s GHG emissions. Emissions could rise by 6% on average over the next 12 months.

4. Future GHG Reduction:

   • 66% of executives expect Gen AI-driven sustainability initiatives will reduce emissions by >10% in 3–5 years.

Proposed Roadmap for Sustainable Gen AI Implementation

1. Identify the Right Technology

• Use hybrid intelligence (combination of traditional AI, Gen AI automation, etc.) tailored to specific use cases for optimal efficiency and lower environmental impact.

2. Assess and Mitigate Environmental Impact

• Monitor and report Gen AI footprints transparently.
• Optimize through:
  • Model efficiency (smaller, task-specific models)
  • Energy-efficient hardware and green data centers
  • Sustainable infrastructure (low-carbon energy sources)

3. Build Use Cases Focused on Sustainability

• Prioritize sustainability-focused use cases like ESG reporting, waste reduction, circular economy, material substitution, etc.
• Focus on quick wins (e.g., ESG automation, sustainable product design, logistics optimization).

4. Govern Sustainably

• Establish strong governance frameworks for sustainable Gen AI deployment.
• Foster collaboration between Gen AI teams and sustainability teams.

Key Challenges

• Lack of transparency in Gen AI provider reporting.
• Measurement difficulty due to incomplete tracking of resource usage across development and deployment stages.
• Governance gaps in sustainability practices and data visibility.

Conclusion

Gen AI offers significant business potential but poses substantial environmental risks. Organizations must integrate environmental considerations into Gen AI strategy to support the UN SDGs and global net-zero goals.