Principal Product Engineer, Cloud Platform

About the job

GPU racks pull 120–140 kW today. By 2027, that number hits 600 kW to 1 MW per rack. The entire AI buildout — hundreds of billions in capex — is being erected on a grid that was not designed for it. Design margins have compressed from 30% to 10–15%. The monitoring systems built for the last generation of infrastructure poll at one-second intervals. GPU workloads ramp in eight milliseconds.

AI is accelerating faster than the infrastructure beneath it can be understood.

The incumbent vendors — Schneider, Eaton, Vertiv — were built for a world where loads were predictable and slow. They are not broken. They are mismatched to what AI infrastructure demands. Verdigris captures continuous waveforms at 8 kHz. That is not a software improvement on existing monitoring data. It is a different measurement entirely — one that makes visible what no other system can see: hidden degradation, safe operating headroom, and the real-time electrical behavior of infrastructure running at the edge of its design limits.

We are not a monitoring solution. We are the electrical intelligence layer — the validation layer that sits between the physical environment and the autonomous control systems the industry is building toward. Solving this matters beyond the business case. Carbon-free AI, stranded capacity recovery, and the long-term reliability of the compute layer the world is betting on all depend on getting electrical intelligence right at the physical layer.