The SCI-AI specification, developed by the Green Software Foundation, provides a standardised approach to measuring the carbon emissions of AI systems. As AI workloads grow rapidly, having a consistent methodology is essential.
What SCI-AI Measures
SCI-AI extends the Software Carbon Intensity (SCI) specification specifically for AI:
- Input tokens — The computational cost of processing prompts
- Output tokens — The energy consumed generating responses
- Cache tokens — Cache creation and read efficiency
- Model identification — Different models have different carbon intensities
- Compute metrics — Latency and processing time
Why Standardisation Matters
Without a standard, organisations measure AI carbon in different ways, making comparison impossible. SCI-AI creates a common language for AI sustainability reporting.
Tailpipe and SCI-AI
Tailpipe’s AI telemetry SDK captures all SCI-AI required fields automatically. Whether you’re using OpenAI, Anthropic, Google Gemini, or Mistral, the same standardised data is collected — without ever capturing prompts or responses.