OCP Telemetry White Paper -- A Summary of the Work Led by Hyperscalers on DC Telemetry Guidelines

Author: Scott Sharp, Mechanical Engineer, Google Collection: 2025 OCP Global Summit - Data Center Facility

Summary

Scott Sharp, a mechanical engineer at Google and long-time OCP contributor, presents a telemetry standardization initiative led by Google, Microsoft, Meta, and Nvidia under the OCP Data Center Facility track. The effort is producing a white paper (in review at time of presentation, expected to publish within 1-2 weeks) that documents the business case for standardized telemetry integration between third-party data center service providers and their tenants. The core problem: building management systems (BMS) and power management systems (PMS) sit on the service provider's secure network, while tenants need real-time access to infrastructure vitals -- cooling temperatures, pump speeds, CDU pressures, power metering -- to schedule and fine-tune AI workloads. No scalable, standardized, secure method exists today to bridge this gap across different data center operators.

The presentation frames telemetry as the data center equivalent of an EKG machine, essential for monitoring the "vital signs" of facility infrastructure. Sharp explains that AI workloads create fundamentally different power and thermal profiles compared to traditional compute/storage/networking: GPU in-rush currents produce rapid power fluctuations with much greater amplitude, and these fluctuations operate on timescales faster than mechanical systems (chillers, pumps, generators) can react. This mismatch means infrastructure operators need preemptive signals -- ideally from the workload scheduler itself -- to ramp chillers before large training jobs begin. The telemetry white paper aims to codify the data points, reporting frequencies (from 1-second snapshots to 60-120 samples per second for waveform analysis), naming conventions, security protocols, and transport mechanisms (MQTT and REST API are the current candidates, though both may be too latent for real-time building control) required to make this two-way data sharing work at scale.

The white paper and subsequent specification will address four key areas: best practices for integration, standard naming conventions for BMS/PMS points (currently "dealer's choice" with tens to hundreds of thousands of points per building), standard security protocols for bridging the service-provider and tenant networks without compromising either, and an implementation guide. Sharp noted that the greatest challenge is the security boundary: tenants often refuse to allow any networking hardware or pathways to leave their secure environment, making physical data transport a potential necessity. The initiative plans to establish smaller working groups for specific subtasks like point naming once the white paper is published.