SODA Foundation joined the cloud native and AI communities in Shanghai for KubeCon + CloudNativeCon China 2026, September 7–9.
We showcased Aitra Meter, an open-source project for measuring AI inference energy consumption and attributing infrastructure usage to workloads.
The demo showed Aitra Meter running with Kubernetes, connecting GPU, inference, and workload-level metrics. The focus is straightforward:
- Measure GPU, tokens, and time
- Attribute energy and resource consumption to workloads and teams
- Optimize efficiency, cost, and carbon
The discussions at the booth highlighted a growing need for better visibility into AI infrastructure. GPU utilization, inference performance, energy consumption, and cost are closely connected, but are often measured separately.
Aitra aims to provide an open foundation for bringing these signals together, working across Kubernetes, AI runtimes, accelerators, and infrastructure.
Thanks to everyone who visited the booth, tried the demo, and shared ideas with the team. We look forward to continuing the work with the open source community.
