3 energy-driven networking trends emerging in the AI era

Rising energy demands in the AI era

According to Lee, components such as switchboards, power, cooling and electrical infrastructure, also known as gray space, are becoming increasingly important in data centers in the AI era. For example, data center capacity is projected to increase to 200 gigawatts by 2030, consuming 300% more energy by 2050, according to research from National Electrical Manufacturers Association’s 2025 Year in Review.

"Artificial intelligence is rewriting what's possible, and it's clear our energy systems must evolve to keep pace," she said. "Every model, every breakthrough demands more power than the last."

The increased power requirements have made it more difficult to build a data center, Lee said. In the past, when operators built data centers, the assumption was that sufficient power was available for deployment. Now, in the age of AI, data center operators must consider physical characteristics -- such as grid capacity, water supply and infrastructure build-out time -- as these constraints can hinder the ability to build a plant.

Cisco's three-pillar framework

Energy Networking Systems is Cisco's answer to this power problem. Rather than an architecture platform, Energy Networking Systems is an umbrella term that encompasses the following three main components for managing power-hungry AI-native infrastructure:

1. Energy visibility.
2. Power distribution.
3. Thermal management.

Energy visibility

The first layer of Energy Networking Systems is energy visibility, provided through a dashboard in Cisco Cloud Control called Energy Management. It provides data center operators with the telemetry data necessary for the following:

• Visibility and reporting.
• Insights and analytics.
• Control and automation.

The tools within Energy Management enable organizations to understand their energy consumption, mix and costs, while recognizing the carbon intensity and greenhouse gas emissions of their projects.

Power distribution

Cisco is supporting power distribution with modern technology called Fault Managed Power (FMP). This technology, developed by engineers worldwide, including those at Cisco, delivers more electricity over longer distances. Unlike traditional electricity delivery systems that require the data center to convert energy before it reaches the devices, FMP uses transmitters, cables and receivers to send electricity to devices in a data center rack.

"It is safe to touch, faster to install, requires less material and the limits on wattage are sort of left up to the imagination because it goes up to 450-volt DC," Lee said.

Cisco's focus on power distribution suggests that it is critical to AI data center networking. This improved form of power distribution can not only power AI data centers, but also entire buildings -- including smart buildings -- edge-computing equipment, battery energy storage systems and more.

Thermal management

Cooling is one of the most important parts of this umbrella, because so many resources go toward it. In traditional air-cooled data centers, more than 40% of electricity goes toward cooling. Lee described this as "non-useful power," meaning that a large amount of energy is being directed toward a less significant resource.

Many newer cooling methods are available, such as rear-door heat exchangers or direct-to-chip cooling. Lee noted some of Cisco's cooling approaches, including its network switches with direct liquid cooling -- such as the Cisco Silicon One G300 102.4T switch -- and immersion cooling.

"When you look at all these different emerging architectures for power and cooling, specifically with cooling, it's a double knock-on effect because you no longer have to spend as much power [on] cooling," Lee said. "And that cooling may not take as much power to get the output you need."

With these different options, Lee said, organizations now have multiple choices for cooling. One data center can have different cooling options. This is especially useful in situations where organizations must retrofit their existing data centers.

Workday's phased approach to FMP

A power distribution technology like FMP was especially useful to Workday, according to Ben Paterson, a technology and smart building leader at the company. Workday faced visibility concerns, unable to recognize how much power the organization was consuming.

"A facility needs to be a living and breathing entity," Paterson said. "Everything should provide data [so that] we can track how people are utilizing the space, track flow, or heating and cooling."

Paterson and his team decided to deploy FMP in a phased approach. First, they created a proof-of-concept with Cisco and its partners to see how FMP could benefit them. In the first phase, they installed a temporary setup with FMP in a space within a facility. During this test, FMP powered everything in the room, such as the lights, desk power and monitoring equipment. In a typical electrical distribution system, this would have taken Workday weeks, Paterson said. With FMP, it only took an hour or two.

Now in phase two of the deployment, Workday is expanding FMP across its entire office environment and testing how to improve it within the organization. Paterson added that if he were to build a new data center from the ground up, he would consider how much power his networking equipment and servers use. When these consume less power, the organization experiences less power loss.

The future of data center efficiency

Only a minority of power in data centers goes toward useful compute, according to Cisco research. In addition to the 40% of non-useful power that goes toward cooling in data centers, another 25% is lost in power distribution, bringing the total to 65%. The annual energy cost of running one of these data centers averages around $6 million.

However, according to the same report, switching from air cooling to liquid cooling can reduce non-useful power by more than 20%, bringing the total to 43%. Organizations can also see a half reduction in costs, going from $6 million to $3 million.

Organizations can see even more savings when combining liquid cooling with FMP. Non-useful energy reduces to less than a quarter at 22%, and the costs of running a data center drop to $2 million. This can result in 76% in energy cost savings and enable more than double the compute with the same energy.

With that in mind, however, every setup will be different. Similar to Workday's phased deployment, organizations will need to test Energy Networking Systems technologies to find out how they fit into their existing data center infrastructure.

"The actual benefit of efficiency and how you design, wire [and] architect it is up to you because it varies," Lee said. "Every environment is slightly different. We're encouraging people to get into your environments, get a proof-of-concept going [and] get a proof-of-value going."

Deanna Darah is site editor for TechTarget's Networking site. She began editing and writing at TechTarget after graduating from the University of Massachusetts Lowell in 2021.