- +1-408-508-6255
- 830 Hillview Court, Suite 100 Milpitas, CA 95035
- info@prophetstor.com
Milpitas, CA, Oct. 30, 2026 — ProphetStor Data Services, Inc. today announced that it has been approved for a U.S. patent covering a method and system for optimizing cooling resources in a data center based on workload prediction. The method forecasts GPU workload demand and adjusts cooling resources, including fan speeds, liquid cooling flow rates, and air conditioning settings, section by section before the heat arrives. Data center operators can replace facility-wide, fixed-setpoint cooling with cooling that tracks each section’s predicted workload, cutting energy use and operating costs while keeping high-density GPU infrastructure within thermal limits.
In modern GPU data centers, cooling and air conditioning account for 30 to 40 percent of total electricity use. Facility teams run cooling through operational technology (OT) systems on fixed parameters and temperature feedback, while information technology (IT) teams schedule workloads separately. Because GPUs generate heat unevenly across the facility, this split leaves idle areas over-cooled and busy areas running hot.
ProphetStor’s patented method streams IT telemetry, such as software workload, server, and GPU performance metrics, together with OT telemetry, such as temperature, humidity, power, and air and liquid cooling data, into a single prediction engine. Using Multi-Layer Cascade Causal Analysis, the engine traces how a change in workload ripples from the software layer to hardware temperatures and then to cooling demand at the facility layer, predicting future cooling requirements over both time and space for every section of the data center.
For each predicted workload demand, the system compares two options and picks the one that uses less energy:
The system also reduces cooling in sections running below capacity, pre-cools sections predicted to heat up, and hibernates unused servers while scaling back cooling around them. As workloads change, it retrains its machine learning prediction and correlation models, closing the feedback loop.
“If you’ve run GPU workloads at any real scale, you know how fast things move. A training job spins up, inference traffic doubles overnight, and suddenly you’re either overpaying for idle GPUs or scrambling to allocate more,” said Eric Chen, CEO of ProphetStor. “What this patent covers is a way to get ahead of that. We use correlation analysis and time-series forecasting to predict demand and allocate accordingly, so GPU clusters stay utilized without the constant fire drills.”
ProphetStor builds the newly approved patent into Federator.ai Cortex™, its AIOps full-stack solution for next-generation GPU data centers. The patent puts Federator.ai Cortex’s unified IT and OT closed-loop approach into practice: Smart Liquid Cooling, a capability of Federator.ai Cortex, uses workload predictions to set coolant flow and CDU control, matching liquid cooling to the GPU heat load each high-density rack is about to produce. Operators can then make compute and thermal decisions in one closed loop instead of managing them in separate systems.
These capabilities help AI factory operators improve energy efficiency and PUE, extend hardware reliability, and advance ESG and sustainability goals while scaling GPU capacity.
Founded in 2012 and headquartered in Milpitas, California, ProphetStor Data Services is an AI-driven infrastructure optimization company purpose-built for next-generation AI factories and GPU data centers. Its Federator.ai platform delivers full-stack solutions spanning GPU performance maximization, predictive workload-aware liquid cooling, and IT/Cloud resource optimization — helping organizations maximize compute performance, improve resource efficiency, and achieve ESG and sustainability goals.
For more information, visit prophetstor.com.
Please select the software/ platform you would like a demo of: