About Phaidra
Phaidra provides an AI control platform tailored for the orchestration of power, cooling, and workload management systems within high-density data centers, often referred to as AI factories. The primary objective of the system is to maximize energy efficiency, specifically targeting the "tokens per watt" metric. By integrating various components of the facility's tech stack—from the Building Management System (BMS) to the individual racks—Phaidra ensures that the entire infrastructure operates as a single, coordinated unit of compute.
The technology centers on the use of agentic AI that autonomously manages critical hardware like liquid cooling Distribution Units (CDUs) and chiller plants. These self-learning agents are designed to maintain thermal stability and have been shown to reduce thermal spikes associated with AI workload ramps by approximately 80%. Additionally, the platform features a conversational LLM agent that acts as a specialized assistant for facility operators. This allows staff to query operational data and troubleshoot issues using natural language, significantly accelerating root-cause analysis.
This tool is intended for data center operators and facility managers overseeing large-scale, mission-critical environments. It is specifically optimized for facilities deploying advanced hardware, such as NVIDIA gigawatt-scale AI factory designs. Because it allows for higher cooling temperature setpoints without compromising GPU safety, it is highly valuable for organizations looking to reclaim power from their facility infrastructure to support higher IT capacities and revenue-generating compute.
What sets Phaidra apart is the background of its creators and its focus on closed-loop, autonomous control rather than simple monitoring. Founded by former Google and DeepMind engineers who were responsible for pioneering data center energy efficiency projects, the platform applies reinforcement learning to industrial systems. Unlike traditional rule-based controllers, Phaidra’s agents are self-learning, allowing them to adapt to changing environmental conditions and complex system interdependencies that manual management might overlook.
Phaidra FAQs
How does Phaidra improve thermal stability in data centers?
The platform's AI agents reduce thermal spikes by nearly 80% when workloads ramp up. This autonomous control ensures that cooling systems react instantly to changes, protecting hardware and maintaining uptime.
What is the benefit of the conversational LLM agent?
The LLM agent allows facility operators to troubleshoot and analyze facility data using natural language. It can reduce the time required for complex operational analyses from days or weeks to just a few minutes.
Does Phaidra help increase IT capacity?
Yes, by optimizing cooling efficiency and enabling higher temperature setpoints safely, it reduces the power required for facility support. This freed-up power can be redirected to increase the number of revenue-generating GPUs.
Can Phaidra manage liquid cooling systems?
Phaidra is designed to optimize the entire cooling stack, specifically including liquid cooling Distribution Units (CDUs). This ensures that liquid cooling infrastructure operates efficiently alongside the broader chiller plant.
Who developed the technology behind Phaidra?
The technology was developed by a team including the engineers who built DeepMind's AlphaGo and those who led Google’s data center cooling optimization projects, resulting in significant energy savings.
