Light the Spark
Step 1
Start Simple
step 2
Build A Complete Approach
Step 3
Build A Complete Approach
The transformation to sustainable computing can start in any IT function and go in any order. There are also green principles for every role. There is also an essential shift in organizational culture, where sustainability becomes a consideration on par with reliability or delivery deadlines.
1
Make Sustainability
Reviews Routine
Reviews Routine
2
Adopt Green Software Development
3
Adopt Green DevOps/Cloud Ops
4
Make Green AI
Decisions
Decisions
5
Plan the Way
Forward
Forward
Make Sustainability Reviews Routine
Practitioner’s Tip:
“Implementing sustainable practices doesn’t require a complete system overhaul—it starts with small, actionable steps that create lasting change.”
Make sustainability a routine part of work processes and planning by making it part of the standing agenda. Projects could all have a Software Carbon Intensity (SCI) score included. Add a few questions, such as:
- What was the SCI for this application before and after this revision?
- What is the carbon awareness built into this code?
- Are we using hardware most efficiently, and can we extend the life of the hardware we are using?
Adopt Green Software
Practitioner’s Tip:
“Once engineers realize how their software components affect carbon emissions, they look more closely at what their team is doing.”
Integrate sustainability into the software development life cycle:
Development Practices
- Review and optimize CPU-intensive operations
- Eliminate redundant code, computations, and API calls
- Minimize data transfers with compression, caching, and CDNs
- Eliminate unnecessary data creation and storage
- Choose lightweight libraries and energy-efficient frameworks and languages
- Shut down inactive dev systems to eliminate idling
Testing and Deployment Strategies
- Test sparingly and eliminate redundant test steps
- Use energy-efficient testing environments
- Parallelize tests to reduce pipeline runtime
- Shut down inactive test systems to eliminate idling
- Track efficiency metrics
Green Software Design Culture
- Add power and GHG emissions metrics to development dashboards
- Establish end-to-end sustainability targets for all IT functions
- Share green patterns and information on internal repositories
- Eliminate excess workstation purchases and shut down idling equipment
- Evaluate new features by how much their compute and storage will cost
- Include efficiency criteria in code reviews
Adopt Green DevOps
and Cloud Ops
Practitioner’s Tip:
“Give a few basic metrics to visualize cloud GHG emissions, even just using your cloud provider’s carbon tool, and provide an equivalent such as driving distance or phone charges. It doesn’t need to be perfect – begin with whatever data you can access, and build from there.”
Make your organization’s data center use cheaper and greener with optimized resource management:
Resource Optimization
- Right-size VMs, containers, and databases
- Compress stored data and transmitted data
- Vary service levels according to business needs
Workload Management
- Set up tools to shift workload according to carbon intensity and high-demand periods
- Enable auto-scaling and load-balancing
Idle Resource Reduction
- Minimize idling with managed services, serverless environments, and virtualization
- Implement automated shutdown schedules for non-production environments during non-work hours
- Regularly decommission unused resources
Performance Monitoring and Resource Consumption Measurement
- Track work done per watt and processor utilization rates
- Implement dashboards tracking power usage and GHG emissions metrics
- Monitor grid carbon intensity per compute hour across different workloads
Make Green AI Decisions
While these software design principles apply to all applications, GenAI’s significant impact on the environment requires additional considerations:
Think Carefully About Using GenAI
- Question whether a GenAI application is necessary for your use case
- A different technology may fulfill requirements with lower resource consumption and GHG emissions
- Consider financial and sustainability costs—they may exceed the benefits
- If it is the only solution, focus on efficient design and deployment
Smaller Open-Source Models
- Select smaller models trained on a data set targeting your required capability or subject matter
- Use pre-trained open-source models when possible—green software groups like Hugging Face offer a range of options
- Look for or add climate impact data on model cards (HuggingFace has instructions)
Optimized Model Training
- Set low accuracy criteria and generous retraining thresholds to minimize testing and retraining
- Consider stopping model training at 20% completion and using accuracy forecasting to find the optimal point (mixed-quality model)
- Optimize training data pipelines with pruning and quantization
Applications Designed for Efficiency
- Design applications for power conservation which lowers directly SCI
- Remove unnecessary GenAI features to reduce inferences and data storage
- Limit inference runs in user interactions with well-design user experience
- Eliminate unnecessary data creation to curtail data storage growth
Edge Locations to Curtail Data Transmission
- Large data transmission volumes from GenAI apps drive GHG emissions from networking equipment operations
- Locating models and processing on edge locations brings down SCI, if edge computing is suitable
Plan the Way Forward
Your organization’s adoption of cloud sustainability is part of a broader transformation. Extend your impact by getting involved at a policy and industry level:
- Support advocacy campaigns to contest utilities expanding gas power plants
- Advocate for tighter data center emissions rules such as EU CSRD
- Question the rapid adoption of resource-intensive technologies like billion-dollar GenAI models
- Join advocacy campaigns for green computing and cloud sustainability
By recognizing and reducing the environmental impact of cloud computing in your organization and in the broader industry, you are positioning your organization as a sustainability leader.
Next Steps
(1) Caballar, Rina Diane. “We Need to Decarbonize Software.” IEEE Spectrum, March 23, 2024. https://spectrum.ieee.org/green-software.
(2) “Hugging Face – The AI Community Building the Future.” https://huggingface.co/.
(3) “Displaying Carbon Emissions for Your Model.” Accessed February 6, 2025. https://huggingface.co/docs/hub/model-cards-co2.