AWS re:Invent 2025 - Architecting for sustainable IT at scale (AIM255)

Architecting for Sustainable IT at Scale

The Growing Challenge of Cloud Sustainability

• Cloud workloads are growing at a rate of 25% year-over-year, leading to increasing environmental impact and operational costs.
• As cloud usage scales, companies must address regulatory compliance around carbon reporting, water usage, and emissions.
• Maintaining brand reputation is also a key driver, as customers, stakeholders, and governments increasingly demand sustainable IT practices.

The Environmental Impact of Generative AI

• Generative AI models like GPT are extremely energy-intensive, with a single GPT training session consuming the equivalent of 150 homes' annual electricity in the US, 300 homes in Europe, or 1,000 homes in India.
• This rapid growth in energy-hungry generative AI workloads must be addressed alongside the scaling of traditional cloud workloads.

AWS Sustainability Well-Architected Framework

• In 2021, AWS introduced the Sustainability Well-Architected Framework, based on 6 key best practices:
  1. Region Selection: Aligning cloud resource placement with demand and renewable energy availability
  2. Scaling: Implementing autoscaling and spot instances to right-size resources
  3. Data Strategies: Optimizing data lifecycle management and transfer patterns
  4. Hardware & Services: Leveraging energy-efficient AWS hardware and services
  5. Software Architecture: Designing software for energy efficiency
  6. Culture & Process: Embedding sustainability as a first-class concern

Quick Wins for Improving Sustainability

• Implement autoscaling, use spot instances, and leverage serverless services to right-size resources and reduce waste.
• AWS Compute Optimizer can provide recommendations to resize resources for 20-30% improvements in cost and carbon efficiency.

AWS Hardware Innovations for Energy Efficiency

• AWS Graviton processors are up to 60% more energy-efficient than standard EC2 instances.
• AWS Inferentia chips are up to 50% more power-efficient for inference workloads.
• AWS Trainium chips offer up to 25% better efficiency for training machine learning models.

Sustainable Generative AI Architectures

• Leverage pre-trained and optimized models from services like SageMaker to minimize the energy impact of training.
• Implement data lifecycle management policies to reduce the storage and processing of unused data.
• Optimize data transfer patterns by aligning cloud resources with user demand and leveraging regional data centers.

Measuring and Monitoring Sustainability

• Use the AWS Customer Carbon Footprint tool to track emissions and energy usage across your cloud workloads.
• Leverage CloudWatch metrics to monitor resource utilization, data transfer, and other key sustainability indicators.

Conclusion

• Sustainability is a critical concern as cloud and AI workloads continue to scale rapidly.
• By following the AWS Sustainability Well-Architected Framework and leveraging AWS hardware and services, organizations can significantly improve the energy efficiency and environmental impact of their IT operations.
• Continuous monitoring, optimization, and a sustainability-first culture are key to maintaining a sustainable cloud footprint.