How to Prepare Data Centers Power Distribution for AI

As AI applications become more prevalent in data centers, power distribution becomes increasingly complex. And there are many factors to consider, from increased power demands to the need for power redundancy.

Power consumption is a crucial factor to consider regarding AI/HPC applications. In some cases, the power requirements can exceed 40kW, which can cause problems for the equipment. As these AI applications become more sophisticated, the demand for power in data centers has increased substantially.

As a data center manager, it's essential to consider upgrading busway systems and implementing power monitoring points to optimize our power and cooling systems. Doing so can help us run our operations more efficiently and effectively, especially as we integrate Machine Learning.

To ensure a reliable and secure power distribution system, consider using a busway solution to monitor power at various points along the busway—providing a broader range of data for analysis and identifying potential energy efficiency and reliability improvements. By gathering accurate and reliable data, data center managers can make informed decisions, leading to effective strategies that optimize operations and result in significant cost savings and improved performance.

When powering high-density areas where AI servers and hardware are stored, specialized power distribution solutions are a must. They must be both efficient and safe, and cooling systems must be robust enough to handle the heat generated by AI workloads.

While energy-efficient designs and variable frequency drives can help power systems, there are still challenges to overcome. Non-linear components can cause harmonic distortions, leading to efficiency problems, overheating, and safety hazards in data centers. To combat this, busway systems can be designed with an oversized neutral conductor and monitored with a power meter that measures total harmonic distortion.

Starline M70 Power Meter

Engineers are suggesting new infrastructure planning approaches as we rely more heavily on AI. One common distribution topology involves A/B busway runs, which allows for seamless switching between power sources in the event of maintenance or failure, ensuring that our servers remain powered and our apps continue running smoothly, meeting our high-reliability metrics. However, there are differing opinions about whether AI workloads truly require this level of redundancy, and alternative systems like 5N/4 redundant system are being proposed in engineering circles.

It's crucial for data centers to stay up-to-date with the latest developments in power distribution strategies to meet the industry's changing demands. With the rise of AI, engineers must adapt their systems to ensure high reliability and seamless switching between power sources.