Many enterprises, excited by the transformative possibilities of AI and machine learning, have begun to pilot use cases for the technology that demonstrate real business value. Yet many are quickly realizing that the benefits of AI carry significant dependencies – namely, power management – that must be addressed before AI can truly deliver on its expectations

As the AI era transitions from hype to reality, data center operators must embrace bold new approaches to managing power to balance aggressive carbon reduction goals with meeting the growing energy needs of AI.

AI innovation brings new challenges

It is now commonly acknowledged that AI infrastructure consumes significantly more power than traditional computing infrastructure, with massive implications for both data centers and the grid. This has led to the recently recognized phenomenon in larger AI data centers known as power “bursting,” a frequent and dramatic swing in utility power draw by as much as +/- 50 percent, repeating as frequently as every few seconds continually.

Faced with potentially higher energy costs and issues with power availability, data center operators are seeking creative solutions, including targeting non-traditional locations (such as the Midwestern US) for the construction of new data centers.

At the same time, operators continue to manage issues that have long been top-of-mind but are now amplified amid AI’s rise, such as skilled labor shortages and an evolving cybersecurity landscape.

Grid operators face challenges of their own. The appetite for energy from both AI data centers and other innovations, such as electric vehicles, is placing new demands on the grid, while utilities work to meet aggressive carbon reduction targets amid the energy transition.

It is becoming clear that data center and grid operators alike must find creative approaches to managing power and designing infrastructure to meet the growing energy needs of AI. Fortunately, emerging innovations can help lower costs and offset data center energy usage while providing critical support for the grid.

A grid-interactive solution

There is now growing industry awareness that it is no longer enough for data centers to simply be consumers of energy. They must become energy producers.

One way to achieve this is to leverage existing technology in the data center – namely, the uninterruptible power supply (UPS). These devices, which serve as a bridge to generator power in the event of an outage or interruption, leverage batteries to provide backup power capabilities that help enhance the resiliency of critical infrastructure.

Recently, lithium-ion battery technology has gained widespread adoption in UPSs, as it offers longer battery life in a reduced footprint, in addition to other benefits. What lithium battery chemistry also offers is energy storage capacity not previously achievable with traditional valve-regulated lead-acid (VRLA) batteries.

What if operators could use the energy storage capabilities of lithium-ion UPS batteries to provide power back to the grid? Or, alternatively, use the battery as a buffer to perform peak shaving or take advantage of time-of-use rates from the grid operator?

It may seem far-fetched, but grid-interactive UPS technology is rapidly gaining steam. By applying sophisticated analytics and digitalization, operators can turn the UPS from a dormant asset in the data center into a distributed energy resource, which can create a fast-responding bi-directional flow of energy to and from the grid.

Using this solution, part of Eaton’s "Data Centers as a Grid" approach, operators can put energy infrastructure to work in new ways by offsetting peak energy usage and providing grid stabilization capabilities through frequency response and other grid services. They can also create a potential revenue source to offset higher energy costs and the infrastructure upgrades required to meet AI demand.

A self-sustaining data center

Another way for data centers to produce their own energy is to deploy full-scale microgrids, which have emerged as an attractive solution for ensuring energy resiliency, creating independence from utilities, advancing islanding capabilities, and optimizing energy usage.

Microgrids are stand-alone electrical power systems that integrate electrical loads and two or more generating assets that can operate autonomously from the grid. They also offer a reliable solution to unexpected power loss by balancing variations in energy demand and usage. With microgrid technology, data centers can better support the integration of renewable energy, such as solar and battery energy storage, to further offset energy costs and enhance resiliency.

Microgrids are not a one-size-fits-all solution, and every operator will have unique applications and assets needed to engineer the right solution. However, as the energy transition continues and AI power needs grow, microgrids will come to be viewed as an essential technology for data centers to support AI use cases while continuing to advance the energy transition.

Embracing digital transformation

Digitalization can play a key role in helping operators optimize performance, resiliency, and sustainability in response to the growing complexity and power usage of AI infrastructure. Therefore, it should not be forgotten in the evolving AI landscape.

The more data operators can collect and analyze, the better they can improve performance, mitigate potential power events, analyze and report key sustainability metrics, and improve asset utilization both in the core data center and across edge environments.

Data center infrastructure management (DCIM) software has had a resurgence recently, offering operators the opportunity to better understand and manage assets in the core data center as their infrastructure and operations become more complex.

Data center performance management (DCPM) software goes beyond traditional DCIM by providing a holistic view of the data center with critical insights into space, power, cooling, connectivity, and other resources to streamline operations, optimize efficiency, improve sustainability, and enhance performance.

The software can also help operators manage and control network-connected assets across distributed environments, such as edge data centers and sites supported by the core data center. It proactively identifies and automates actions to prevent IT equipment from going down, in addition to helping collect and report on power usage data.

These digital solutions will soon be viewed as essential tools for data centers to meet their sustainability goals while managing and reporting on energy usage in AI data centers. It is important, however, that operators seek digital solutions that prioritize cybersecurity in their design. As infrastructure becomes more connected, threat actors will potentially look to connected power devices as targets for cyberattacks, and operators should invest in tools that place cyber protection at the forefront.

Final thoughts

The AI boom and the energy transition are creating tremendous opportunities for data centers to rethink their power management approach to help support both emerging power needs and demands on the grid. As this happens, it is imperative that operators shift from power users to power producers and seek out digital solutions that can help them become smarter about power usage in data centers and distributed environments.

By leveraging technologies such as grid-interactive solutions, microgrids, and emerging digital tools, operators can help deliver on businesses’ expectations for AI while anticipating energy needs and becoming active participants in the energy transition.