Insights
Compute and energy notes
Notes on energy measurement, workload evaluation and infrastructure.
Measuring energy per workload
Why power draw alone cannot tell you whether a system is more efficient.
From host measurements to facility impact
A clear way to connect component-level improvements to the bigger picture.
Planning an infrastructure evaluation
Define the workload, the measurement and the decision before you begin.
External research
Primary sources on the energy challenge. These organisations are referenced for their published work, not as CARMIX customers or endorsers.
- IEA · Key Questions on Energy and AI
2026 analysis of data-centre electricity demand and infrastructure bottlenecks.
- Google · Measuring AI’s environmental impact
A measurement approach that considers the wider system supporting AI inference.
- NVIDIA · From Megawatts to Tokens
A systems perspective on compute production within available power.
- OpenAI · Building compute infrastructure
The infrastructure and energy ecosystem behind the expansion of AI compute.
- Anthropic · Electricity and infrastructure
Energy investment, grid impacts and the responsibilities of AI infrastructure.
Discuss CARMIX.
Start a conversation about energy efficiency in your computing environment.